Ranking
- Current Issue
- All Issues
- 1
Structure recognition guided texture suppressing image smoot...
217 - 2
Review on brain functional parcellation based on resting-sta...
170 - 3
Gradient guided higher-order model based on Riemann geometry...
167 - 4
Adaptive bilateral logarithm transformation with bandwidth p...
167 - 5
Hidden information extraction from the ancient painting usin...
146 - 6
Video shadow elimination algorithm by combining HSV with tex...
145
- 1
Adaptive bilateral logarithm transformation with bandwidth p...
41 - 2
Review on brain functional parcellation based on resting-sta...
39 - 3
Gradient guided higher-order model based on Riemann geometry...
37 - 4
Saliency detection combining background and foreground prior
30 - 5
Video shadow elimination algorithm by combining HSV with tex...
27 - 620
- 1129180
- 211558
- 311119
- 49876
- 5
A Survey on Image Segmentation Using Active Contour and Leve...
9385 - 68510
About the Journal
- Current Issue
- Online First
- Special Issues
Review
- Review on brain functional parcellation based on resting-state functional magnetic resonance imaging data Hu Ying, Wang Lijia and Nie Shengdongdoi:10.11834/jig.170081 23-09-2017 170 39 Abstract：Objective Brain functional parcellation based on resting-state functional magnetic resonance imaging data has better functional consistency than traditional structural parcellation(such as AAL atlas and Brodmann atlas) in the construction of functional networks.However,a substantial part of parcellations of brain functional modules are rough at this stage,and it needs more precise and accurate brain functional parcellations,which can define the basic functional unit at macro scale.Considering its theoretical value,this study comprehensively reviews the existing brain functional parcellation methods and their applications in the field of brain science.Method With the widespread investigation and massive literature,we clear the general steps for the segmentation of brain functional areas based on the relationship between rs-fMRI data and brain functional network.The brain functional network can be further acquired only by defining the results of the brain parcellation.Then,the state-of-the-art ideas,methods,and the improvement of the original methods about brain functional parcellation are described in detail.This paper also classifies and analyses these algorithms of brain functional parcellation,whose good points,bad points,and sphere of application are stated in detail.In addition,the evaluation criteria of brain functional parcellation quality is introduced.Finally,the development trends and existing problems of this technology are highlighted.Result The brain functional parcellation can be divided into the whole brain functional parcellation and the regional brain functional parcellation according to the situation of brain regions.The whole brain functional parcellation can generate functional atlas and analyze the functional characteristics of the brain on the whole brain space scale.Also,according to the brain network,which can be constructed by functional atlases,it can promote the development of brain cognitive and brain-like artificial intelligence technology,occupying an important position in brain science research.The regional brain functional parcellation is aimed at dividing larger brain area with mixed functions into sub-regions with strong functional consistency,reflecting the functional distribution of the regional brain with more pertinence.Therefore,regional brain functional parcellation can be used to study brain-related diseases,and detection of abnormal changes in brain regions in time,which is significant for the prevention and diagnosis of brain diseases.The algorithms of brain functional parcellation are divided into two parts,namely,data driven and model driven,and the advantages,difficulties,and challenges of each algorithm are discussed.The data-driven parcellation methods always use limited parameters in the parcellation process and are less demanding on the shape of the image data to be processed.Most of the clustering methods have better performance on non-Gaussian data sets.Therefore,the application of data-driven parcellation methods are quite extensive,but the stability and accuracy of need improvement.Compared with the data-driven method,the model-driven method involves more parameters.During model establishment,the distribution of clusters and the model division of voxel signals require strong domain knowledge and reliable theoretical basis.These requirements limit the development of the model-driven methods to a certain extent.However,considering its uniqueness of the corresponding image,it can reflect individual differences and be applied to brain functional parcellation at group level.Moreover,the existing evaluation criteria can be a preliminary assessment of the functional parcellation results.For a more accurate evaluation,there is a lack of rigorous theoretical support for the overall evaluation system.Conclusion Some meaningful progress and valuable research results have been obtained in the study of brain functional parcellation based on resting-state fMRI.Therefore,researchers could use multimodal brain function data,and even combine with brain structure data,for brain functional parcellation.In addition,improving these methods is necessary so that more precise and accurate functional atlas could be studied in the mechanism of human brain,applied in early prevention,accurate diagnosis and curative effect evaluation of brain injury and neuropsychiatric diseases and brain-inspired and brain-computer interface intelligence technology.For example,based on a more detailed brain subregions,functional analysis of subregions can help to identify new biomarkers of specific diseases.Moreover,traditional partition algorithms can be combined with innovative ideas such as prior knowledge,spatial domain information,spatial constraint,sparse coding,feature selection,and sample learning in the follow-up research.These confluent brain functional parcellation algorithms are dedicated to finer brain functional regions and constructing more detailed and accurate brain functional networks to analyze advanced functions of the brain.
Image Processing and Cording
- Gradient guided higher-order model based on Riemann geometry for color image denoising Lu Bibo, Li Yang, Wang Yongmao and Gao Tianlingdoi:10.11834/jig.160668 23-09-2017 167 37 Abstract：Objective Image degradation is commonly unavoidable during acquisition and noise makes the later processing difficult and inaccurate.Partial differential equation methods,especially low order methods,are efficient for grey-scale images denoising.Instead of the heuristic channel-coupled method,researchers begin to deal with color images under Riemann geometry framework.This framework uses the arc element to measure the rates of change and the eigenvectors of the metric tensor to describe the direction of the change.However,these methods generate low-order partial differential equations and their extension to the higher-order model remains an important challenge.The high-order models have ability to eliminate the undesired staircasing effect that accompanies the use of a model based on first-order derivatives.In this paper,a geometry-driven higher-order model for removing noise from color images is proposed.The proposed method introduces a gradient-based weigh function to improve edge detection and preserving ability while removing noise.Method Within the Riemann geometry framework,the norm of the arc element is calculated and generates a quadratic form called the first fundamental form.This norm can be interpreted as the distance of the ellipse to its center.The eigenvalues of the metric tensor correspond to the semi-major axis and the semi-minor axis of that ellipse.Therefore,its Jacobian matrix allows the measurement of edges in the vector-valued images.Various matrix norms can be established based on this matrix to characterize reasonable measurement for constructing variational models for color image denoising.However,as a low-order model,its corresponding low-order partial differential equation also suffers from staircasing effect.Inspired by the low-order model,a second-order equation is derived from the area element within the geometrical framework for image processing.Its norm square is decided by a second-order-based matrix and the Frobenius norm for this matrix is obtained.Based on this special norm,a higher-order variational model is proposed and a high-order partial differential equation is derived using variational principal.The gradient information is used to guide the higher-order diffusion to preserve edges during the diffusion process.The gradient is first convoluted by Gaussian kernel to predict edge locations to reduce the effect by additive Gaussian white noise.Analysis on the nonlinear diffusion term shows that the diffusion is controlled by the following information:the guide information based on first-order derivatives,the diffusion information based on second-order derivatives,and the second-order derivatives ration between the single-color channel and three-color channels.Result Experiments are conducted for various data,including one-dimensional signal, synthetic images,and standard test images.In every experiment,the test data are corrupted by additive white Gaussian noise with different variances.The results obtained by the proposed model are quantitatively and visually compared with the related methods.Peak signal-to-noise ratio(PSNR) and structure similarity index(SSIM) are used for quantitative comparison.Zoomed images and residual images are used for visual comparisons.The ability of recovering piecewise linear is verified for one-dimensional synthesized signal.The proposed method has an obvious improvement in both objective index and visual perception.The PSNR of the recovered signal processed by the proposed method increases from 32.64 dB to 33.16 dB compared with the low-order model when the standard deviation of the noise is 35.The SSIM result of the proposed method increases from 0.969 5 to 0.991 8 compared with the channel-coupled mean curvature method under the same condition.The proposed method is also compared with the decorrelated vectorial total variation model.The PSNR increases by 1.37 dB,and the SSIM increases by 0.005 8.The proposed method has the best objective index compared with related methods.The results are plotted in the cubic RGB color space.The same color is mapped to the same point in this space,and the linear segments in the RGB space correspond to the smoothing change of color.The result of the proposed method gives the best performance in the recovery of both constant area and the linear part of the noisy signal.Experimental results on a set of color image data are also given.Compared with the current related methods,the average improvement is 2.33% for PSNR and 0.4% for SSIM.Conclusion The proposed model can suppress noise efficiently from a piecewise linear image while avoiding the staircasing effect and giving a better performance at the edge of the color image.The proposed method is efficient for removing noise with different variances.
- Image encryption algorithm based on chaotic maps and bit reconstruction Ping Ping, Li Jianhua, Mao Yingchi and Qi Rongzhidoi:10.11834/jig.170049 23-09-2017 113 13 Abstract：Objective Many of the current image encryption algorithms are based on bit level,whichhave security flaws.Many bit-based image encryption algorithms divide the plain-text image into eight-bit planesaccording the eight binary pixels,and then scramble the eight bit planes,which results in 0 bits and 1 bits of each bit plane to not change,onlymakingthe position change of the 0bit and 1bit in each into bit plane,so that there are security flaws.In this paper,a new image encryption algorithm was proposed according to the existing problems of security flaws in these popular encryption algorithms.The proposed algorithm can resist the chosen-plaintext and the chosen-ciphertext attacks,and solve the position restriction of 0bit and 1bit in the bit plane to bring about global reconstruction.Method The scrambling of the image encryption algorithm is divided into two stages:the first stage of the scrambling is to be transformed into a binary pixel matrix for global scrambling;the second stage of the scrambling is to block the pixel matrix after the global scrambling and scramblefor each bit plane.First,the Tent chaotic map is used to generate a pseudo-random sequence.Then,each pixel of the plain-text image is converted into binary bits,andpseudo-random numbers are sorted in ascending order to generate a new set of sequences.For example,in any sequence {0.3,0.7,0.5,0.4,0.8,0.2},the sequence is sorted in ascending order to obtain the ordered sequence {0.2,0.3,0.4,0.5,0.7,0.8},and then the corresponding position sequence is {6,1,4,3,2,5}.The new sequence is used to carry out the whole row and column scrambling.The pixel matrix is divided into eight blocks to perform the Henon map scrambling,and then the final cipher image is obtained by the diffusion operation.Result The distribution of the pixel value of the plain-text image after encryption changes from non-uniform to uniform distribution,and the correlation between the pixels of the plain-text image is broken.Thus,the original image has no statistical characteristics,the number of pixels change rate(NPCR) and the unified average changing intensity(UACI) is close to the ideal values,and the algorithm can resist the differential attack.Experimental results show that when the algorithm key changes little,more than 99% of the pixels in the resulting cipher images are changed.This algorithm belongs to the symmetric encryption algorithm,anddecryption algorithm is also used in the same key so thatthe decryption key also has the same conclusion.Thus,the encryption algorithm is sensitive to the key.The key space of the algorithm must be large enough to resist the exhaustive attack.The key of the encryption algorithm is composed of two parts,which are the keys used to generate the chaotic sequence and the parameters needed in the diffusion phase.The key space of the algorithm is 2^{192},which can resist the exhaustive attack.The algorithm can guarantee the security of encryption and have lower computational complexity.The cipher image obtained by the encryption algorithm of the plain-text image can be obtained by the information entropy formula.The information entropy of the cipher image is 7.996 2,which is very close to the ideal value of 8.Experimental results show that the encryption algorithm can avoid the information leakage duringimage encryption,and the image encryption algorithm has good anti-entropy analysis attack.The algorithm of image encryption proposed in this paper mainly includes two steps:scrambling and diffusion.The complexity of the algorithm is mainly reflected in the scrambling process.The running time of this algorithm is shorter than that of other image encryption algorithms,and the diffusion process is added to the algorithm.Therefore,the algorithm is more secure than other image encryption algorithms and is more resistant to differential attacks.Conclusion The proposed algorithm is also a relatively classic "scrambling-diffusion" structure.Compared with other encryption algorithms with the same structure,this algorithm is based on bits to scramble.When 1bit in a certain pixel and 1bit of another pixel change in position,the changes not only include the position of the pixel but also the value of the pixel.Experimental results show that the proposed image encryption scheme has numerous characteristics,including large key space,low correlation of adjacent cipher pixels,and high sensitivity to the plain-text and key,which can effectivelyprotect the security of the encrypted image.In this algorithm,the global bits of the plain-text are first scrambled to avoid the scrambling of the bits in the same bit plane,which results in the weight of the 0bit and 1bit tonot change.The scrambling sequence used in the scrambling process is related to the plain-text image,so it shows partial diffusion effect.The experimental results also show that the algorithm is safe and practical and has good application prospects in image encryption and other applications.In the future work,we will continue to explore the new image encryption algorithm,and now compared with the popular encryption algorithm to improve the efficiency of the encryption algorithm while ensuring the security and practicality of the algorithm.
- Adaptive bilateral logarithm transformation with bandwidth preserving and low-illumination image enhancement Mao Dongyue, Xie Zhengxiang, He Xiangqian, Jia Yuanyuan and Zhou Lihuadoi:10.11834/jig.170094 23-09-2017 167 41 Abstract：Objective In a low-illumination environment,such as in nighttime video surveillance and some special scenes,the limitations of the image acquisition device,non-professional photography,and loss of information in video transmission often results in the acquisition of image with low brightness and poor contrast.Such conditions bring great challenges to image post-processing,such as image recognition,segmentation,and classification.Therefore,enhancing a low-illumination image in the preprocessing step is necessary.The aim of low-illumination image enhancement is to enhance the dark area,suppress the highlighted area,and realize the image clarity.At present,most of the methods are based on Histogram Equalization(HE),Retinex theory,and homomorphic filtering.Particularly,HE can adaptively improve the dynamic range of the image gray scale but it also can lead to unnatural over-enhancement of image contrast;and the merging gray levels cause loss of some details of the image.Retinex-based algorithms can enhance image contrast to a certain extent,but the computational complexity is generally high,the computational speed is slow,and the image color is also easily distorted.The premise of the enhancement algorithm based on homomorphic filter is that the illumination is uniform,so this method is unsuitable for low-illumination images with uneven illumination.Moreover,this method also lacks self-adaptability because the dynamic range depends on the frequency of the filter.Although the logarithmic transform can show more details of the dark area,but it also loses some details of the bright region.The Retinex algorithm and the enhancement algorithm based on homomorphic filter all involve logarithmic transformation,but none of the logarithmic base are specified.Only one-way logarithmic transformation is carried out,which can only improve the image contrast of the dark area.To overcome the shortcomings of the existing algorithms,and inspired by the characteristics of logarithmic transformation,this paper proposes a low-illumination image enhancement algorithm based on adaptive bilateral logarithm transformation with bandwidth preserving.Method The proposed method includes four steps.First,the low-illumination image is transformed into a standardized image by a special gray transformation called the standard transformation,which can stretch the image contrast to some extent.The purpose of image standard transformation is to make the image gray/color spectrum width equal to 256 to preserve full bandwidth.Compared with non-standardized image,the contrast and brightness of standardized image had increased.Standard transformation lays foundation for further image quality optimization.The second step of the algorithm is computation of the Average Luminance(AL) of the standardized image.Then,the adaptive bilateral logarithm transformation with preserved bandwidth is performed according to AL.More concretely,if AL is less than 127.5,reverse logarithm transformation with bandwidth preserving is carried out first,and then the forward logarithm transformation with bandwidth preserving is performed.Otherwise,forward logarithm transformation with bandwidth preserving is carried out,and then the reverse logarithm transformation with bandwidth preserving is performed.Through calculation,the value of logarithmic base is set to 1.021 983 956 89,thus achieving logarithm transformation with bandwidth preserving.Through this step,image details both in dark area and bright area can be displayed.Finally,the image is rounded out to obtain the enhanced image.Result In the experiment,29 high-quality images in the LIVE database release 2 are used as reference images,and then processed into low-illumination images by Photoshop CS5.After that,the proposed algorithm is utilized to enhance these low-illumination images and compared with the enhanced results obtained by HE,Multi-scale Retinex(MSR),and Natural Preserved Enhancement Algorithm(NPEA).Qualitative and quantitative analyses are conducted to evaluate the proposed algorithm.Experimental results show that the overall contrast and brightness of the proposed method are improved subjectively,and the enhancement effect is better compared with the other three enhancement algorithms.Simultaneously,the Peak Signal-to-Noise Ratio(PSNR) and Structure Similarity(SSIM) value obtained by the proposed method is higher than the other three algorithms.The average PSNR and SSIM values obtained by the proposed method is 22.75 dB and 0.86,whereas the average PSNR and SSIM value of the other three method are 16.16 dB and 0.58(HE),15.82 dB,and 0.62(MSR),18.62 dB,and 0.78(NPEA),respectively.In addition,the average running time of the proposed algorithm is relatively short(74 ms);however,the running time of MSR and NPEA are respectively 11.28 s and 11.58 s under the same conditions.Conclusion The proposed method makes up for the defects of retinex algorithm and homomorphic filtering method,which can improve the dark area and bright area contrast of the image at the same time.Consequently,it can enhance the low-illumination image effectively.Moreover,the algorithm can eliminate the halo artifact caused by Retinex,and it does not merge gray levels as HE.The enhanced image is more natural and more consistent with the human visual system.Meanwhile,the proposed algorithm is simple and easy to implement,which can greatly improve the operational efficiency.The proposed method can be widely applied to image enhancement in low-illumination environment under backlight or uneven illumination.However,the limitation of the proposed algorithm is the contrast and brightness of the enhanced image should be further improved.The future work will focus on applying the average luminance transformation to the enhanced image to do further enhancement.In addition,for the sake of further improving the robustness of the algorithm,more tests and verification are required for the nighttime video monitoring field.
- Structure recognition guided texture suppressing image smoothing Shao Huan, Fu Xinyi, Liu Chunxiao, Wu Min, Gong Chen and Yu Zongjiedoi:10.11834/jig.170200 23-09-2017 217 17 Abstract：Objective Natural scenes generally contain different scale objects and textures,which carry rich information in regard to human perception.Texture usually signifies pixel values,which change with high frequency.Generally,images are composed of many important structures,texture,edges,etc.Therefore,mining the meaningful structure from textures or complex background images is a critical task in vision processing.The core of image smoothing lies in the separation of structure and texture.Effective preservation of the structure while suppressing the texture with strong gradient or varying scales is a challenging problem.Most of the existing image smoothing methods tends to deal with weak gradient texture images;if the texture gradient is strong,then these methods will fail.To solve the abovementioned problem,a structure recognition guided texture smoothing algorithm is proposed,which deals with the structure and the texture separately and detect structure before image smoothing.Method First,this paper argues that the fundamental difference between structure and texture is the repetition pattern.Particularly,the structure should be sparse and the texture should be a region with a repeating pattern.According to this characteristic,the discriminative features for distinguishing between structure and texture are designed and extracted based on the multi-scale analysis of inherent variation.At least two reasons are available for presenting the multi-scale approach.One reason is that structure and texture are relative.When the scale is small,the texture may not show up,and thus the scale needs to be enlarged and the essence of the texture is released.The other reason is that the texture in the image is diverse,and the adaptive scale in different regions is difficult.Furthermore,textures with various attributes may exist in the same image,a single scale can only solve the partial texture with the default scale parameter and the recognition of other textures will lose.Therefore,multi-scale analysis of inherent variation is proposed to ensure that different textures can display their own repetitive pattern attributes.Second,the core part in the field of pattern recognition is feature extraction.Therefore,the feature extracted must be more robust to guarantee the discrimination ability is strong enough and the stability is good enough.To obtain more accurate features,we need to consider the multi-scale inherent variation in the macroscopic view and grasp its general rules.After we analyze the trend of multi-scale inherent variation curves at different pixel locations,several discriminative features are extracted.Then,these features can be used for subsequent structural recognition.We regard the separation of texture and structure as a typical two-class issue,and the support vector machine is a classical two-class classification method.Compared with many existent machine learning methods,it is a relatively lightweight classifier,which can obtain desirable classification results without a large sample.Consequently,this paper prefers to use the support vector machine to distinguish the texture and structure,with the help of support vector machine,a classifier is trained with the extracted feature pixels,and utilized to classify structure and non-structure pixels efficiently.However,due to the block effect in edge compression and the computational mechanism of inherent variation,pixels nearby the structure will always be affected by the real structure and its multi-scale inherent variation curve is similar in structure.Hence,the support vector machine classification results cannot reach a single pixel.We observed large amounts of data and find that the non-structured pixel appeared symmetrically on both sides of the window.Although the support vector machine classification results are coarser,in the middle of the skeleton should be considered as the real structure.In this paper,a morphological thinning method is adopted directly to get a thinner structure,but the results of thinning still have some weakness.To dispose the shortcomings of the support vector machine classification results after thinning operation,we design two steps of post-processing work,including outlier rejection and deburring,which solve the burr and mistaken isolate.As such,the finer structure recognition maps can be obtained.Finally,based on the fine structure obtained in the previous step,a structure guided bilateral image smoothing method is put forward to remove texture while preserving structure.Result The multi-scale inherent variation features proposed in this paper achieve a correct rate of 96.12% with support vector machine,and our structure guided image smoothing results can effectively suppress the texture details with strong gradient or varying scales while preserving the structure.These excellent experimental results are compared to some results of previous methods,which reveal that the proposed methodology yields better image smoothing.Conclusion In view of the limitations of existing similar methods,this paper analyzes the characteristics of inherent variation deeply and proposes an algorithm to distinguish the structure and texture by means of multi-scale inherent variations.Based on the support vector machine classification results,a post-processing is used to obtain a finer structure recognition map.Then,a structure guided bilateral image smoothing method is applied to remove texture while preserving structure.Our algorithm outperforms the state-of-the-art image smoothing methods,especially for those images containing texture with strong gradient or varying scales,which could strongly promote such technical fields as structure extraction,detail enhancement,image segmentation,tone mapping,image fusion,and object recognition,which reflect the potential practical application values.
Image Analysis and Recognition
- Video shadow elimination algorithm by combining HSV with texture features Wu Minghu, Song Ranran and Liu Mindoi:10.11834/jig.170151 23-09-2017 145 27 Abstract：Objective In the application of video surveillance target detection,the shadow will directly affect the accuracy of the target detection in the scenes,so the shadow suppression algorithm is particularly important.The traditional algorithm which is based on hue,saturation,and value(HSV) to detect shadow is popular.Inspired by color perception mechanism of human visual system,this algorithm detects the shadow by the luminance ratio between the current video frame and background model.We propose a shadow elimination algorithm based on HSV spatial feature and texture features to overcome the shortcoming of the luminance ratio between them,which causes the moving target to be mistaken for the shadow.Method The Gaussian mixture model can effectively overcome the interference caused by the change of illumination and periodic disturbance of background image.First,the mixed Gaussian model(essentially 3 to 5) is used to characterize each pixel in the input images,and the updating mixed Gaussian model is obtained after the new input frame image.Each pixel of the current image is matched with the mixed Gaussian model,and if it is successful,the point belongs to background;otherwise,it belongs to foreground.The algorithm based on HSV color space can detect the shadow accurately by calculating the luminance ratio between the current video frame and the background model because the hue and saturation values are approximate in the shadow compared with the ones in the background,and the luminance value of shadow pixels is lower than the luminance value of the background pixels.The luminance ratio between them is usually 0.7 to 1.Therefore,the moving target can be obtained by combining the foreground detected by Gaussian mixture background model with the shadow detected by the method based on HSV color space.The traditional algorithms based on HSV color space can obtain the accurate detection results,but the moving target is often mistaken for shadows seriously in the video frames.To overcome this problem,we use texture featuresthat conclude local binary pattern(LBP) and OTSU to extract the moving target.LBP is an operator of gray-scale variation.A smaller threshold is selected,which is compared with the difference value between gray value of the central pixel and gray value of its corresponding neighborhood pixel.If the difference is greater than the threshold,it is marked as 1;otherwise,it is marked as 0.Therefore,we can obtain a description of the texture change at the location of the central pixel.The LBP operator extracts local texture features by the original gray level of the image.OSTU is a maximum interclass variance method.If the shadow and the target have large variances,then the two parts have much difference.When the partial target is regarded as shadows or part of the shadow is regarded as the target,the difference of two parts becomes smaller.Thus,the largest variance segmentation between classes can result in a minimum probability of misclassification.According to the gray feature of image,the image is divided into the shadow and the target by OTSU.The complete moving target is obtained by OR operator of combining the foreground,which is respectivelyextracted by OTSU and LBP operator with the result that is extracted by HSV.Result The proposed algorithm is applied to several different shadow videos,which are included in CVPR-ATON standard video library and CAVIAR standard video library.Experimental results show that when the threshold of luminance ratio,which is applied to detect the shadow in HSV color space,remains unchanged,the moving target is extracted accurately and its shadows are basically eliminated.Compared with other traditional algorithms,such as statistical parametric(SP) approach,statistical nonparametric(SNP) approach,and two kinds of deterministic non-model(DNM1,DNM2) approach,the proposed algorithm obtain the better result.Experimental results show that the proposed algorithm has much better increment of about 10% than the forehead algorithms in terms of the average of shadow detection rate and shadow discrimination rate.Although in the Intelligent Room video,the shadow discrimination rate is 1.6% lower than that of the DNM2 algorithm and the shadow discrimination rate is 2.9% lower than that of the SP algorithm in the video of Laboratory;thus,the algorithm improves the shadow detection rate by 29% and 27.2%,respectively.In the real-time test,this algorithm can process 12~15 frames per second,which can satisfy the real-time needs.Conclusion Although the traditional algorithms that use HSV has great effect in the shadow elimination,the moving target may easily be interpreted as the shadow.The texture featuresthat include LBP and OTSU can make up for this shortcoming,wepropose the video shadow elimination algorithm by combining HSV with Texture features.Compared with other algorithms,our method can obtain more accurate shadow detection result and has much better advantages in terms of average shadow detection rate and shadow discrimination.Our method can be applied to intelligent video surveillance,remote sensing images,and human-computer interaction.Our future work will focus on improving the real-time performance.
- Saliency detection combining background and foreground prior Yao Zhaojian and Tan Taizhedoi:10.11834/jig.170114 23-09-2017 133 30 Abstract：Objective Saliency detection aims to automatically identify and localize the important or attractive regions from an image.In the recent years,many researchers have given particular attention to saliency detection and took it as an important step in image processing.Saliency detection has been applied to many computer vision tasks and applications,such as image retrieval,object detection and recognition.Although saliency detection has been studied for many years,there are still certain shortcomings.For example,the detection in complex scenes is inaccurate or the results of the detection contain background noises.Considering that several existing methods of image saliency detection cannot suppress the background regions effectively,or cannot highlight the complete object regions clearly,a novel saliency detection method combining background priori and foreground priori information was proposed to further improve accuracy.The background prior is an assumption that the regions along the image boundaries are background regions,and the foreground prior is to calculate a convex hull to locate the foreground object regions.Method The region saliency of an image is defined as its similarity to the foreground in addition to being defined as its contrast to the background.Therefore,background and foreground can be extracted with prior,and all the regions of an image can be compared with these background and foreground to generate a saliency map.First,we selected the superpixels from image boundaries as the background regions to compute a background-based saliency map based on the dissimilarity between each region and the background regions.Second,we applied the convex hull from interest points to approximately locate the foreground object.Convex hull of original image and filtered image were calculated because there were not only salient regions inside the convex hull,but also the background regions,and the intersection regions of the two convex hull regions were obtained to remove background regions to some extent.Then,the intersection regions were combined with the background-based saliency map to select the foreground object regions,so foreground-based saliency map could be generated based on the similarity between each region and the foreground object regions.Finally,we integrated the two saliency maps utilize their respective advantages because the background-based saliency map could highlighted the object more uniformly and the foreground-based saliency map could better suppress the background noises.Then,the unified saliency maps was further refined to obtain a smoother and accurate saliency map.Result To test the performance of the proposed algorithm,experiments were conducted on the MSRA10K datasets,which contained 10 000 images and was one of the largest publicly available datasets.The results demonstrated that the saliency map of the proposed algorithm are closer to the ground truth and the proposed method performed favorably against the state-of-the-art methods in terms of accuracy and efficiency.The average precision,average recall,F-measure,MAE,and average running time of the proposed method are 87.9%,79.17%,0.852 6,0.113,and 0.723 s,respectively.Conclusion Saliency detection is a promising preprocessing operation in the field of image processing and analysis.We proposed a new method to detect saliency based on a combination of two kinds of prior information.The detection results of the proposed algorithm could not only effectively suppress the background noises,but also clearly highlight the object regions,thus improving the accuracy of the detection.
Image Understanding and Computer Vision
- Rapid saliency detection method using LapSVM Wang Chen, Fan Yangyu and Xiong Leidoi:10.11834/jig.170051 23-09-2017 110 13 Abstract：Objective With the development of Internet technologies,the amount of information has grown exponentially.The information can change the traditional ways of people's lifestyle.They can bring great convenience in daily amusement,education,and commerce,but they also lead to many new challenges on the existing processing technologies.On one hand,people need to handle the amount of information,which greatly exceeds the processing capability of computers.Thus,allocating the limited computational resource to the important visual information is important.On the other hand,people hope computers can simulate the functions of human eyes,which can effortlessly select a small amount of important information for further complex processing.Visual saliency of images can reflect the degree of stimulation of the human visual system to different regions.The reliable saliency methods can automatically predict,locate,and mine the important visual information.Thus,it can help computers in effective selection of important information from the massive visual data,which is suitable for image segmentation and image retrieval.At present,the robustness and real-time performance of the algorithm have been a very active research area.In this paper,we present a rapid saliency detection method based on Laplacian Support Vector Machines(LapSVM).The proposed saliency method can extract the salient regions in image within a relatively short period of times and reach a better accuracy.Method First,we segment the source image into many regular regions using the Simple Linear Iterative Clustering(SLIC) algorithm.The SLIC algorithm is a simple and efficient method to decompose an image in visually homogeneous regions.These regions are called superpixels,which provide a convenient way to calculate local features.They can reduce the complexity of image processing by obtaining the redundancy of images.In this paper,we use superpixels instead of image pixels to participate in the calculation of the algorithm,thus reducing the amount of computation required.Then,we construct the graph using the similarity between character of regions.Second,we define the rough-labelled samples using the boundary feature of image regions,and classify using LapSVM algorithm.LapSVM has shown the state-of-the-art performance in semi-supervised classification.Following the manifold regularization approach,the LapSVM used is trained in the primal.We speed-up the training by using an early stopping strategy based on the prediction on unlabeled data or,if available,on labeled validation examples.This allows the algorithm to quickly compute approximate results with roughly the same classification accuracy as the optimal ones,considerably reducing the training time.The computational complexity of the training algorithm is reduced from O(n^{3}) to O(kn^{2}),where n is the combined number of labeled and unlabeled samples and k is empirically evaluated to be significantly smaller than n.Thus,the LapSVM trained in the primal is the primary tool of the proposed saliency detection algorithm.Third,we extract the more robust labelled samples based on analysis of the result of classification.Then,the classification must be done again based on LapSVM.The classification result of every superpixel is the probability of the category to be owned,and we define the probability as the saliency value of the superpixel.Finally,we get the saliency map using the energy function to optimize the classification result.The saliency map is a gray image and its intensity is between 0 and 1.We can use the ground truth to verify the accuracy of results.Result Compared with seven other well-known saliency detection algorithms on ASD dataset.The ASD dataset is the subdataset of the MSRA dataset,and it contains 1 000 images.It is widely used in the saliency detection experiments of many algorithms.Experiments show that the proposed algorithm exhibits impressive performance with real feature and it maintains also good robustness.The running time of the proposed algorithm is shortened to about 0.03 s and Mean Absolute Error(MAE) is about 4%.Conclusion We propose a novel rapid saliency detection method based on LapSVM.Using the boundary feature of regions and analysis of classification,we get the more accurate background and foreground samples.Experimental results prove that the proposed algorithm maintains the robustness compared with the latest algorithms,and greatly reduces the running time.Therefore,compared with other algorithms,the proposed algorithm is more suitable for real-time application,such as detection and tracking.
- Hierarchical multi-object tracking algorithm based on globally multiple maximum clique graphs Wang Xueqin, Jiang Jianguo and Qi Meibindoi:10.11834/jig.160527 23-09-2017 111 14 Abstract：Objective The variability of the illumination and the correlation of the object,the robustness of multi-object tracking algorithm is relatively poor because of the complexity of the background.The main challenges in multi-object tracking include occlusion,false positives,complexity of object motion,and ambiguity caused by similar features in appearance.To solve the above problems,a hierarchical multi-object tracking algorithm based on globally multiple maximum clique graphs is proposed.Method The method is based on global association of data association,hierarchical,and network flow theory and uses a two-layer framework.Each layer uses a shorter trajectory to form a longer trajectory.An undirected graph is first constructed according to the network flow theory.The nodes are composed of several track segments and the weights are obtained using the linear combination of the motion model and the appearance model of the object.Then,the occlusion object is processed by the aggregation dummy node,and the spatial constraint is added to solve the problem of identity transformation.Finally,the mixed-binary-integer programming is used to solve the undirected graphs on the superposition problem.Simultaneously,a plurality of maximal cliques is obtained.Result Experiments are conducted on public datasets through TUD-Stadmitte,TUD-Crossing,PETS2009,Parking Lot 1,Parking Lot 2,and Town Center to verify the method,and all the datasets show desirable results.The number of identity transformation handled by the dataset Town Center is 12,and is higher by more than 5% for the dataset TUD-Stadmitte.Conclusion Based on the idea of data association,this paper proposes a hierarchical multi-object tracking algorithm based on globally multiple maximum clique graphs.The spatial constraint of key can effectively deal with multi-object tracking problem,especially the problem of occlusion effect.This method has practical application value in intelligent video surveillance.
- Superpixel object tracking with adaptive compact feature Tian Jian and Wang Kaijundoi:10.11834/jig.160619 23-09-2017 136 20 Abstract：Objective Object tracking is the basic theory of computer vision that has been given increasing attention.Object tracking encounters several natural challenges,such as illumination change,scale variations,occlusion,deformable,fast motion,random movement,object presence,analogues or busy background,and low resolution.Recently,superpixel to model object appearance has been employed for object tracking.However,existing superpixel object tracking algorithms(RST) have provided uniform feature confidence to superpixel blocks belonging to the object and similar interference objects in same category,which is difficultly distinguished between object and similar interference objects.A superpixel tracking algorithm with adaptive compact feature(ACFST) is proposed to solve similar interference objects.Method In every frame,the surrounding region of the target is segmented to many superpixels and each superpixel has feature confidence due to the objective model in the last frame.The new method creates a smaller compact search scope to adapt to the object size,and then the feature confidence corresponding to superpixels inside the scope remained unchanged,and the outside scope had decreased.The size of the compact region is controlled by a set of parameters whose values adapt with every change of each frame.The similar interference objects in the background around the object are partitions into the outside of compact search scope and marked as inadequate objective.As such,the feature confidence of the superpixels in interference objects is decreased to reduce miscalculation.Object is composed of multiple superpixels with different feature confidence.When tracking an object in every frame,the candidate sample around the target location of last frame have different confidence.Then,the Bayesian inference is used to find the sample that correspond to the maximum a posteriori probability estimation in the current frame to be regarded as an object.The feature confidence outside of the scope decreases because of the compact search scope,which means that the degree of interference objects is low so that misjudgment did not occur.Result The proposed tracking algorithm is verified using two video sequences with a background similar to the object,namely,Basketball and Girl.The new superpixel object tracking algorithm(i.e.,ACFST) is compared with the original superpixel tracking algorithm(RST) from three aspects,namely,mean center location error,success rate,and precision ratio.In terms of mean center location error,the proposed algorithm can be significantly reduced to 5.4 pixels and 7.5 pixels in the two sequences.In terms of success rate,the ACFST is 11% higher than the RST.With the location threshold limit,the precision ratio of the ACFST is better than that of the RST in the two sequences,an improvement of 10.6% and 21.6%,respectively.Compared with the RST that do not distinguish similar interference objects,the proposed tracking algorithm produces more accurate tracking results.Conclusion The proposed method creates an adaptive compact region and set adaptive parameters to control the size of the compact region,thereby reducing the misjudgment between the real object and the similar interference objects during tracking,resulting in excellent robustness.The effectiveness of this algorithm is verified in video sets with similar interference objects.Experiment showed that when the similar interference objects disturb the object or overlap the object,the existing superpixel object tracking algorithms fail to track object and the new method could track accurately.The tracking precision of the algorithm is improved and the robustness is strong,which is more suitable for complex environments,such as background clutter,target occlusion,and deformation.
Medical Image Processing
- Renal cortex segmentation with fully convolutional network and GrowCut Shi Yonggang, Qian Mengyao and Liu Zhiwendoi:10.11834/jig.170190 23-09-2017 120 17 Abstract：Objective Kidney segmentation plays an important role in the diagnosis of kidney diseases.The volume and thickness of renal cortex are effective assessment criteria in early clinical diagnosis for renal neoplasms,chronic arteriosclerotic nephropathy,and acute rejection after kidney transplant.However,most existing methods focus on the whole kidney segmentation.This paper presents a fully automatic renal cortex segmentation based on fully convolutional network and GrowCut for multi-modality kidney images.Method Generalized Hough Transform(GHT) is used to detect non-analytic shape represented by R-table.GHT localizes the kidney and then the region of interest(ROI) is extracted.The fully convolutional network(FCN-32s) for semantic segmentation is introduced into renal cortex segmentation.Data augmentation is employed to expand labeled data and transfer learning is applied due to lack of sufficient training data.The initial parameters of the proposed network are taken from pre-trained model of VGG-16.All the fully connected layers of VGG-16 are converted into convolutional layers.The filter size of fc8 is changed from 1 000 to 2 because the proposed network regards cortex and background as two classes.The filter sizes of fc6 and fc7 are modified from 4 096 into 1024 to reduce parameters.Pooling and down-sampling layers can extract more abstract features for image classification tasks.For image segmentation,however,too much trivial information will be lost.The proposed network retains the first three pooling layers of VGG-16.Dropout,as a regularization method,prevents over-fitting,and loss function is optimized with Stochastic Gradient Descent.The proposed network is a first implementation for realizing renal cortex segmentation based on the fully convolutional network.However,the obvious disadvantage of fully convolutional network is that it is a pixel-wise classification and does not consider the spatial relationship among pixels.It will cause some unexpected results.The segmentation regions have burrs at the edge,and cortex of some slices is missegmented.The proposed method combines fully convolutional network and GrowCut for superior cortex segmentation.GrowCut is an interactive segmentation method based on the labeled seeds,and the pixels compete to gain labels in the neighborhood.The performance of GrowCut relies on the initial seeds marked by the user.In this paper,the seeds are generated by the proposed network,which realizes an automatic implementation and frees manual interventions.The images of test set are firstly segmented by the proposed network,and the feature maps of the last convolutional layer are extracted as a labeled map.The mislabeled seeds that always appears in spine,spleen,and other adjacent tissues can be corrected by GrowCut with priors of kidney.GrowCut can achieve more accurate cortex segmentation based on the correctly labeled map.A set of contrast-enhanced CT images and corresponding ground truth labeled by experts is given for training fully convolutional network.The ROI will be normalized after it is extracted from the original image,and then expanded after cutting and reflections.These images are split into a training set(5 000) and a validation set(300).Parameters of the network will be adjusted in the training processing.The proposed network is trained on a deep learning framework Caffe,with maximum iteration of 10 000,initial learning rate of 0.001,which is multiplied by 0.1 at every 2 000 interactions,momentum of 0.9,weight delay of 0.000 5,and batch size of 1 for online learning.The proposed network is quantitatively compared with three pre-trained fully convolutional networks by three metrics:pixel accuracy,overall accuracy,and mean region intersection over union(IU).The proposed method is compared with four methods,including two fully convolutional networks and two typical methods.Two metrics are used,namely,IU and Dice Similarity Coefficient(DSC),which is a common metric in medical image segmentation.Result The experimental dataset contains 30 clinical CT and MRI images.The proposed method achieved IU of 91.06%±2.34% and DSC of 91.79%±2.39%.The proposed fully convolutional network has fewer parameters and higher accuracy than basic fully convolutional network for renal cortex segmentation.GrowCut algorithm considers the neighborhood information between pixels,thus further improving the segmentation accuracy by 3%.Results show that contrast-enhanced and non-contrast-enhanced images can be accurately segmented.Conclusion Deep learning model trained with huge data set of natural images can extract hierarchical features and can be introduced in medical images segmentation by transfer learning.Missegmentation is reduced by the proposed network and can be corrected effectively.The experimental results indicate that the proposed method is more suitable for kidney images from different modalities and outperforms typical methods.The proposed method can provide a reliable basis for clinical diagnosis because medical images from normal and abnormal kidneys both can be segmented accurately.The furture work includes the following tasks:the accuracy of non-contrast-enhanced image segmentation will be further improved by optimizing algorithm;the proposed network will be used for segmentation of other organs by fine-tuning the parameters of network.GrowCut will be plugged in the network as a new layer to perform end-to-end training.
- Hidden information extraction from the ancient painting using hyperspectral imaging technology Guo Xinlei, Zhang Lifu, Wu Taixia, Zhang Hongming and Luo Xudongdoi:10.11834/jig.170093 23-09-2017 146 17 Abstract：Objective The hidden information of ancient painting is a kind of weak information contained in the culture relics,such as the smearing information,hidden text information,and post-repair information.It may reflect the process of painting,and then we can explore the deep information of the ancient paintings,which is very helpful for the restoration and identification of cultural relics.The hidden information can also provide a new way for us to know more about the culture relics.Traditional identification work for ancient paintings mainly rely on experienced cultural relics identification personnel.Based on their experience,they usually make a judgment to the ancient painting according to the material,creative style,age,and other information to identify.It is with high requirement for staff technical requirements in traditional cultural relics identification methods.Beyond that,this method is inefficient and prone to errors.Thus,quick and effective extraction of the weak hidden information in ancient paintings in a non-destructive method is a desirable technique problem for research and application.Hyperspectral imaging technology takes full advantage of its spectral characteristics,so its extraction accuracy is higher than other imaging method.This paper aims to study how to use hyperspectral imaging technology quickly and intuitively extract the hidden information of ancient paintings.Method For the difficulty of ancient paintings' hidden information extraction,only a few scientific instruments are available for the extraction of ancient painting information.We use the hyperspectral imaging spectrometer to scan the painting to obtain the short-wave infrared hyperspectral data(1 000~2 500 nm).Then,spectral correlation coefficient matching algorithm(SCM),spectral information divergence(SID),and matching algorithm based on spectral information divergence and spectral angle mixing method(SID_SA) is used according to the spectral angle matching algorithm(SAM) to analyze the crown's pigment.The modified information of the crowns are extracted by principal component analysis.Result To determine the composition of the brown pigment in the crowns,the spectrum of the brown pigment was analyzed by a variety of spectral matching algorithms with the established standard spectrum of the ancient pigments.The brown pigment contains the ocher component and the distributions of the carbon black pigment and the ocher pigment vary with the different proportions of the position.The reason for this may be related to the deployment of the pigment when it has been painting,so the proportion of the pigment in the different painting areas is different,resulting in different crowns for spectral mixing phenomenon.The pigments are analyzed by a variety of spectral matching algorithms to determine the types of the pigments.The matching algorithms are compared,then we find the mixed pigments.After the main pigment information is peeled off,the modified information is enhanced to extract effectively.Results show that according to the spectral feature of the pigments,spectral matching algorithms can effectively identify the end element of the mixed pigment in the case of pigments mixing.Compared with other algorithms,the matching algorithm of spectral information divergence and the matching accuracy of the spectral information divergence and the spectral angle SID_SA can reach 0.096,compared with other matching algorithms;thus,its effect is the best.After the main background information is separated by the principal component analysis technique,the smearing information around the crowns,which are located at the bottom of the painting,can be enhanced.This method can effectively distinguish the background pigments information and smear information,and extract the hidden information.Conclusion Hyperspectral analysis technology can get a good recognition result on the extraction of hidden information of ancient paintings,and can accurately extract the types of paint pigments and find the traces of painting.It is especially suitable for the identification of mixed pigments and the underlying information of paintings.Certain pigment colors and hidden information can be enhanced for effective extraction using hyperspectral imaging processing techniques.The shortwave infrared imaging spectrometer may play an important role in the production of digital archives and in the analysis and restoration of cultural relics,thus promoting the development of digital work associated with cultural relics.This method can provide support for the restoration and identification of heritage conservation.
Remote Sensing Image Processing
- Airborne LIDAR building detection based on voxel data structure Wang Liying, Wang Sheng, Xu Yan and Li Yudoi:10.11834/jig.170159 23-09-2017 109 10 Abstract：Objective Automatic building detection is important for 3D city modeling.Airborne light detection and ranging(LIDAR) point cloud data are dense,georeferenced as well as 3D,they are the natural choice for 3D object detection and extraction,e.g.,building.Point cloud,raster grid,and triangulated irregular network(TIN),which are the commonly used methods to represent scattered LIDAR point cloud data for building detection,have defects;for example,their model representations are complex,and thus using the data processing algorithm is difficult,and the results are not accurate and unable to represent multiple returns LIDAR data.To overcome the restrictions of existing point-,grid-,and TIN-based approaches,this paper focuses on "establishing voxel structure model for airborne LIDAR point cloud data and developing a new building detection algorithm based on the constructed voxel model" and proposes a Voxel-based building detection(VBD) algorithm for separating buildings from non-buildings.Method The proposed VBD algorithm consists of three steps.First,LIDAR point clouds is regularized into binary 3D voxel structure,which can be obtained by dividing the entire scene volume into a 3D regular grid(the 3D sub-volumes,called voxels),remapping the LIDAR points to 3D voxels and assigning the voxel value 1 or 0 when a voxel contains LIDAR points or not.Second,the voxels with voxel value 1 are separated into ground and unground voxels utilizing a voxel-based 3D filtering algorithm.Third,a group of non-ground voxels with almost straight line and jump features are selected as building edge seed voxels and then their 3D connected set are labeled as building voxels.The proposed algorithm is based on the idea of 3D connectivity construct and is designed based on a binary voxel structure which is a simpler 3D structure,in which topological and adjacent relations between voxels can be established much easier.The advantage of the proposed algorithm lies in utilizing connectivity and hidden elevation information between voxels.Result ISPRS urban LIDAR datasets,which are representative of buildings of diverse types,are used to analyze the sensitivity of "adjacency size" parameter in the model and assess the accuracy of VBD algorithm quantitatively.The quantitative evaluation results indicate that:1) the 56-adjacency is the optimal adjacent size;2) an over 95% average quality of building-detection,achieving 95.61% average completeness and 95.97% average correctness,which report promising performance.The qualitative evaluation results indicate that large,dense,and irregularly shaped buildings or buildings with eccentric roofs are all successfully captured.The outstanding merit of the VBD algorithm based on the idea of 3D connected set is that some roof elements(e.g.,small dormers,chimneys) can also be detected.Conclusion The proposed building detection algorithm of 3D connected-component label can effectively detect all kinds of buildings,especially urban buildings.The detected building results can directly serve as building model that is a new form of 3D voxel model with certain accuracy and are easy to build a 3D building model.However,the proposed algorithm suffers from separating trees and buildings if both are adjacent to each other,thereby forming a 3D connected set.The reason for this is that the regularized binary 3D voxel data cannot integrate with multispectral,hyperspectral,or other optical data source.Optical images can provide a variety of information,such as intensity,colours and textures,which can provide detailed information for preventing the spreading of the building-regions throughout the neighboring objects,and thus further improving the accuracy of building detection.Nevertheless,we could conclude that the proposed algorithm is promising for automatic building detection.
- Onboard ship saliency detection algorithm based on multi-scale fractal dimension Li Wenjuan, Zhao Heping and Shang Shunandoi:10.11834/jig.160529 23-09-2017 115 9 Abstract：Objective Detection of ship targets in seas is an important research field in remote sensing image target detection.Onboard ship detectors need to detect targets rapidly under limited resources and time-constraint without the prior information about the type and size of the ship targets as guidance.In detection of vessel formation,different sizes of ships in a scene are normally present.Thus,an onboard ship detection method needs adapt to changeable detection scene.Method To solve this problem,a novel multi-scale fractal dimension based onboard ship saliency detection algorithm is proposed,which can detect ship targets of different sizes.Analysis of natural texture images showed that the images of the natural scenes fit the fractal Brown random field.The study of images,such as sea waves,clouds,and other natural objects,shows that the natural objects all have fractal characteristics,whereas ships,aircrafts,vehicles,and other man-made objects hardly show any fractal characteristics.Therefore,the fractal difference between natural background and man-made targets can be used for target detection and recognition of the targets.The key problem is how to estimate the fractal dimension accurately.An improved algorithm is proposed considering that the fractal dimension calculation accuracy is affected by the constrained size of boxes of the differential box counting(DBC) algorithm.In the improved algorithm,the numbers of pairs of points of the fitting line are increased,and the fitting error to eliminate the error pairs of points is introduced.Thus,the calculation accuracy of the fractal dimension features is improved.Experimental results show that the improved algorithm is more accurate in the fractal dimension calculation of small images than the classic DBC algorithm.Based on the accuracy improved fractal dimension,a center-surround(c-s) operator based on the detection principle of Itti model is used in the new algorithm to eliminate the natural background and highlight the ship targets simultaneously given that the natural objects show the self-similarity at different scales,which is different from the man-made objects.The two-parameter CFAR detection method is a classical algorithm for ship target detection commonly used in optical remote sensing images.This method is suitable for the detection of complex image objects with local background changes.Result The proposed algorithm is compared to the two parameter CFAR algorithm.For the different sizes of ship target detection in one scene,the more obvious ship targets are highlighted by the two-parameter CFAR algorithm,and the targets that are considerably different from the obvious ones in size are easily missed.Thus,the detection rate of target detection is reduced.Moreover,ship targets of different sizes are highlighted through the multi-scale approach of the new algorithm and the background is weakened,which is conducive for ship target detection with different sizes.In the experiment,25 remote sensing images are selected.The total number of ship targets in the selected images is 102,and the size of the ship targets varies in a scene,in which the two algorithms are used to detect the ship targets.Compared with the two-parameter CFAR algorithm,the ship detection method based on the multi-scale fractal dimension has a higher detection rate and lower false alarm rate.Conclusion The proposed detection algorithm based on the multi-scale fractal dimension can realize the detection of differently-sized ship targets in a scene of remote sensing images,thus effectively reducing the false alarm rate of target detection while ensuring a certain detection rate.To cope with the changeable and complex detection scenarios of onboard imaging,the new algorithm,which can adapt to different sizes of ship target detection in one scene,is more flexible and has strong adaptability.
Geoinformatics
- Progressive line simplification approach based on the double-oblique-dividing-curve method Du Jiawei, Wu Fang, Gong Xianyong, Li Jinghan and Xing Ruixingdoi:10.11834/jig.170238 23-09-2017 106 9 Abstract：Objective Line simplification is not only one of the most classic problems but also an important content in cartographic generalization.Most custom line simplification methods focus on the vertexes compression of line features.In cartographic generalization,line simplification is a process of deleting subordinate bends and preserving major bends.Therefore,a new line simplified method based on bend units is proposed.Method Line features are simplified by the processes of curve bends division and bends deletion.First,after analyzing the characteristics of the oblique-dividing-curve method,the double-oblique-dividing-curve method is proposed to cover the shortage of the former.Bends divided by this method are more accurate and complete.Second,the relationships of bends divided by the double-oblique-dividing-curve method are detected.All bends are classified into three types according to the relationships detected before.In addition,line features are divided into two types of bend-groups,which are distinguished according to different types of bends.Third,different bend deletion methods are used in different types of bend-groups to simplify curve bends gradually and to avoid simplifying excessively.Finally,processes of bend division,relationship detection,and bend deletion are not repeated until all bends divided by the double-oblique-dividing-curve method can be distinguished in the simplified scale.Line features are simplified progressively to reach the visual requirements of the target scale.Result Line features of hydrographic net someplace at the scale of 1:250 000 are used in the following three experiments.In the first experiment,16 line features and part rivers of the hydrographic net are divided into several bends used the double-oblique-dividing-curve method and the oblique-dividing-curve method.Apparently,each bend divided by the double-oblique-dividing-curve method is more suitable for cognition.The result of the double-oblique-dividing-curve method is more completed than the oblique-dividing-curve method because of avoiding bends omission of the oblique-dividing-curve method.In the second experiment,these 16 line features,at the scale of 1:250 000,are simplified to the scale of 1:500 000,1:1 000 000,and 1:2 000 000 used D-P method,the simplification method based on three-element-bend group,and the proposed method.In terms of the cognitive level,the geometric level and the geographical level,each simplification result used the proposed method of the three target scales is more perfect than the other simplification results used in the other two methods of the same target scale.Moreover,the number of test line features are extended in the third experiment.The hydrographic net,with 965 line features in various shapes,at the scale of 1:250 000,is simplified to the scale of 1:500 000,1:1 000 000,and 1:2 000 000 used the same method as the second experiment.In addition,the location error of each simplified line feature is calculated to evaluate the simplification in quantification,which is used to measure the average displacement resulting from the simplification method.To show clarity,the mean value,the weighted value,and the median value are counted from all location errors of line features in the result,which is simplified to one of the three target scales by one of the three simplification methods.Mean values,weighted values,and median values of results simplified by the proposed method are smaller than the other results simplified by the other two methods at all the three target scales.Conclusion Curve bends divided by the double-oblique-dividing-curve method is more complete and more accurate according to cognition.Line simplified method based on the double-oblique-dividing-curve method is suitable to the process of manual simplification and cognition.Additionally,it has some advantages in preserving the similarity of geometric features and the integrity of geographic features.Although the proposed simplification method is suitable to simplify line features of various shapes.The proposed method has a significant advantage in simplifying complicated line features,which are composed of numerous crowded bends.Importantly,the proposed simplification method based on the double-oblique-dividing-curve method is suitable for practical simplification of cartographic generalization.