最新刊期
卷 20 ， 期 6 ， 2015
- 摘要：A plenoptic camera captures 4D light field information in a single photographic exposure, i.e., 2D spatial and 2D angular information. Nonetheless, the spatial resolutions of images rendered from light fields are significantly low. In addition, angular resolutions cannot meet application requirements given the limited number of viewpoints. In this paper, we present a new edge-preserved super-resolution method based on the weighted BDTV model.We apply the function of the residual errors in the edges of reconstructed images to weigh the data term in addition to different low-resolution viewpoint images. We also employ the previous BDTV model for denoising and edge preservation.Experimental results demonstrate that in terms of visual effects, the reconstruction and edge-preserving effects of the proposed method is superior to those of state-of-the-art methods. Peak signal-to-noise ratio rose by 1 dB while the measures of structural similarity in images increased by roughly 0.01.Our approach not only improves the spatial and angular resolutions of rendered images but also preserves edge information. Although the depth estimate is inaccurate, the reconstruction and preservation effects of our method remain outstanding.关键词：super-resolution;light field image;edge preserve;BDTV prior model;spatial and angular resolution3696|415|4更新时间：2024-05-07
- 摘要：With the rapid development and popularization of computer networks and information storage technologies, networks have become a mainstream tool for information transmission. This medium is used to transmit information, including text messages, audio, images, and other multimedia files. Thus, networks intuitively play an irreplaceable media-related role in the transmission of information flow security as the main carrier of network information. The security problem in the network transmission of image data must be solved effectively; therefore, an encryption algorithm of image blocking and double adaptive diffusion using Arnold maps (BDAM) was proposed in this paper to address image security issues. This algorithm can improve the efficiency and security of image encryption.BDAM first uses Logistic, Tent, and Sine maps to build three types of novel 1D chaotic maps according to the scheme of pairwise combination. This algorithm then extracts the initial chaotic sequence. Then, BDAM defines an initial encryption image matrix that is similar in size to the plain image matrix. The latter is divided into small blocks. The initial encryption image matrix pretreats the parameters of the Arnold map. The pixels in the random position of the plain image matrix are deposited at random positions within the random block of the initial encryption image matrix when ordinary and reverse maps are joined. At the same time, adaptive diffusion occurs among intra-block pixels and inter-blocks until the initial encryption matrix is filled. Scrambling and diffusion synchronization are initiated by combining forward and reverse Arnold mapping processes. The BDAM algorithm can equalize the access sequence of each pixel and the position of each storage sequence, enhance scrambling randomness, and overcome the drawbacks of conventional Arnold mapping from rules to the defects of random access. In addition, this algorithm can apply block and double adaptive diffusion in the scrambling process. The interactions among the ciphertext pixels and ciphertext pixels, plaintext pixels, and chaos random sequence spread the influence of each pixel to the image matrix of the entire site nonlinearly. Furthermore, these interactions continuously disturb chaotic systems during diffusion, initiate adaptive diffusion, and enhance the sensitivity of the encryption key, the ciphertext, and plaintext. The results of simulation and of performance comparison analysis show that the BDAM algorithm is superior to other encryption algorithms for many types of gray images in terms of information entropy, key space, correlation, and sensitivity, among other aspects. Scrambling randomness is favorable as well. Moreover, the algorithm displayed high sensitivity to the key and to the plaintext, thus improving encryption results. Random scrambling that is combined with blocking and double nonlinear adaptive diffusion through the BDAM algorithm can resist various attacks effectively and exhibit high security. Therefore, this method is suitable for all types of gray-scale images and may be applied in this field.关键词：Arnold mapping;blocking;double adaptive diffusion;image encryption3547|284|9更新时间：2024-05-07
- 摘要：The edge information of images is important in image quality assessment. Nonetheless, the image quality evaluation method of feature similarity (FSIM) based on low-level features is not ideal for edge information detection although this algorithm considers the significance of low-level features. On the basis of the information provided above, this study combines the FSIM algorithm with the grid scheduling simulator (GSSIM) algorithm, which is sensitive to edge information, to generate the new image quality assessment method FGSIM. The new method is not only consistent with the characteristics of human visual systems, but it can also identify image edges effectively.The algorithm combines the part of FSIM that represents phase consistency with the component of GSSIM algorithm that can extract image information to generate the new image-quality assessment method FGSIM. The use of phase congruency represents image features, the part of phase consistency that can be used to keep the algorithm close to the human visual system, and the part of the GSSIM algorithm that can extract image information realized by Gradient. This part can be employed to identify image edges.FSIM, GSSIM, and FGSIM algorithms were used to evaluate images containing different motion blurs, and graphs were constructed to represent the obtained data. In the motion blur experiments, the numerical value of the FGSIM algorithm declines from 0.8943 to 0.3443 with the increase in image blur. Changes are significant, and the motion blur is highly sensitive. In the Gaussian blur and Gaussian noise experiments, the changes in the numerical degree value of the FGSIM algorithm are superior to those in the FSIM algorithm to some extent, although the former is inferior to the GSSIM algorithm. Experimental results on public image quality databases show that the scatter diagram of the FGSIM algorithm is slightly inferior to that of the FSIM algorithm. However, the former is significantly better than that of the GSSIM algorithm. Furthermore, the scatter diagram of the FGSIM algorithm is more concentrated than that of the GSSIM algorithm. The FGSIM algorithm also performs better than the GSSIM algorithm in terms of Pearson correlation coefficient, Spearman rank-order correlation coefficient, Kendall rank-order correlation coefficient, and root mean square error. These factors are commonly used to measure performance.Experimental results indicate that the FGSIM algorithm is a new image-quality assessment method that is not only consistent with human visual system characteristics but can also identify image edges effectively. Thus, this algorithm can identify edge information well and is sensitive to variations in image quality.关键词：image quality assessment;feature similarity;gradient feature;structural similarity;edge information4623|318|10更新时间：2024-05-07
- 摘要：Background subtraction is an important step in object detection for many computer vision applications, including intelligent surveillance and human detection. The purpose of this process is to segment moving objects from complex scenes. Performance mainly depends on the background modeling algorithm; however, the background is a complex environment that usually includes distracting motions. Thus, background subtraction is complicated, and an adaptive method is proposed to address this problem.The method is based on the Gaussian mixture model. In their approach, each pixel is modeled by a mixture of K Gaussian distributions. An online learning technique is employed to update background models. In their approach, online K-means is applied to initialize the parameters of the Gaussian mixture model. The number of Gaussian distributions cannot be changed in the process of detection. The initialization of the model parameters significantly influences foreground detection, and the fixed Gaussian distribution cannot accommodate the changing background. In this study, we initialize the parameters of the Gaussian mixture model by combining the online K-means and the online expectation-maximization (EM) algorithms. The outcome of online K-means algorithm is the input of the online EM algorithm. The former rapidly generates the parameter values that are close to the reasonable value, whereas the online EM rapidly and accurately astringes the result that is obtained through online K-means to a reasonable range. In addition, this paper also presents a gray-value classification algorithm to adjust the number of Gaussians to adapt to the dynamic environment. Recent statistics regarding gray value are obtained for each pixel. Then, this algorithm classifies these gray values into different categories. Finally, this method updates the number of Gaussian distributions on the basis of the number of categories. In this paper, we conduct several experiments with four video datasets to evaluate the proposed background subtraction algorithm. Three of these datasets are standard test videos that are widely used in the video monitoring field. The videos entitled “Waving Trees” and “Bootstrapping” are derived from the Wallflower dataset. The pedestrian video is selected from the Change detection dataset. Another video is captured by a local bus station downtown. Moreover, a quantitative analysis is conducted with the general evaluation criteria of Precision and Recall.In the conventional Gaussian mixture model, the number of Gaussian distributions is fixed for each pixel and the system cannot adjust to the changing background. Some static regions apply only one or two Gaussian distributions, whereas moving regions need more Gaussian distributions to maintain the model. However, maintaining additional distributions consumes resources, whereas a lack of distributions may result in false detections.Test results show that the proposed method performs better than the conventional Gaussian mixture model. Experimental findings also suggest that the proposed approach adapts more effectively to complex scenes than those presented in reviewed literature. The background can be segmented effectively and rapidly, and the results for Precision-to-Recall ratio demonstrate the superiority of the method to analogous algorithms. The proposed approach provides a new direction for research on background subtraction. In future work, we will attempt to control learning rate with effective strategies and improve Precision-to-Recall ratio.关键词：background subtraction;Gaussian mixture model;online K-means;online EM;gray value4536|285|7更新时间：2024-05-07
- 摘要：This paper proposes an interactive method of multi-phase image segmentation that maximizes the boundary information in the superpixels of an image. This new method is adequately fast as a real-time, interactive segmentation tool. The new approach first constructs a multi-layer graph that employs the superpixels of an image as graph nodes. The weights of the graph edges are assigned specifically by applying the GraphCut algorithm that appropriately segments the input image. We also propose an interface through which new indicating strokes can be added interactively to improve segmentation quality. A number of examples demonstrate the capability of the new approach to facilitate accurate multi-phase segmentation at low computational cost. In fact, a satisfactory segmentation result is obtained in less than one second for an image with adimension of 449×275 pixels. The computation time of the GraphCut algorithm increases logarithmically as the number of superpixels increases. By contrast, super-pixel computation time increases linearly. Thus, our new method is advantageous in that it uses superpixels as graph nodes instead of employing pixels, as in previous methods. The utilization of pixels considerably reduces graph dimension.关键词：image segmentation;multiphase segmentation;superpixel;GraphCut;network flows3542|495|4更新时间：2024-05-07
- 摘要：Image-based smoke detection methods have many advantages over traditional point-based smoke sensors, including their fast response and lack of contact. Nonetheless, existing methods remain challenged in terms of accurately detecting smoke in images due to significant variances in smoke shape, color, and texture. To improve recognition accuracy, we extract the features of pyramidal textures and edges to propose a novel image-based smoke detection method. We first decompose an image into an image pyramid and then extract the local binary patterns (LBPs) and edge orientation histograms (EOHs) from each layer of this pyramid. These patterns and histograms are called pyramidal LBPs (PLBPs) and pyramidal EOHs (PEOHs), respectively. We also adopt different pooling schemes to generate sequential PLBP and PEOH histograms that represent smoke textures and edges. Finally, we concatenate these histograms to form smoke feature vectors and use support vector machines for training and classification. Image pyramids contain scale information; thus, our pyramidal texture and edge features display certain scale-invariance. Experimental results show that the method reports detection rates of above 94% and false alarm rates of less than 3% given our large image datasets. The texture and edge features extracted with our method exhibit certain illumination and scale invariances. Experiments indicate that these features discriminate and generalize effectively in terms of smoke detection.关键词：edge orientation histogram;local binary pattern;support vector machine;image smoke detection3464|316|8更新时间：2024-05-07
- 摘要：Hand gestures have recently been regarded as an important communication modality in human-computer interaction. These gestures are considerably advantageous over traditional interfaces such as the keyboard and mouse given their convenience and naturalness. Hand gesture interfaces have been successfully employed in many areas over the past 20 years, including robotics, sign language communication, clinical medicine, and the entertainment industry. Recognition is the key technique in most interaction systems based on hand gestures. Such recognition can be grouped into two categories: dynamic and static hand gesture recognition. Most existing methods for static hand gesture recognition usually perform poorly given the large number of hand gesture categories. In this study, we propose a hierarchical method of static hand gesture recognition that combines finger detection and histogram of oriented gradient (HOG) features to overcome this limitation. A hierarchical strategy is designed based on the number of outstretched fingers to solve the “large-category” problem described above. The basic idea is as follows: Entire categories of gestures are divided into several subsets according to the number of outstretched fingers. This number is first recognized by a certain finger detection method as an input gesture. Then, we derive the appropriate gesture from the subset that contains the gestures featuring the given number of fingers via a certain feature extraction approach. Thus, the recognition problem is solved in two stages. Following the first finger-based stage, the number of categories that must be addressed by the second layer decreases significantly. To achieve this target, we propose a new finger detection method based on morphological operations. The proposed method consists of the following steps. First, the hand region is detected with a skin color model. Then, the number of outstretched fingers is identified by the finger detection method. Accordingly, a support vector machine classifier is selected from a set of pre-classified classifiers. Finally, we calculate the HOG feature of the input gesture, which is then inputted into the selected classifier to obtain a result. In our experiments, 25 popular static gestures are tested to verify the effectiveness of the proposed recognition method. The one-stage HOG-based recognition method is mainly used for comparison. Experimental results demonstrate that the proposed method can improve the recognition accuracy of the conventional HOG-based method by approximately 20%.This study makes three main contributions to literature. First, we present a new finger detection method that can accurately determine finger position and identify the number of outstretched fingers. Second, we design a finger-based hierarchical recognition strategy that can effectively overcome the “large-category” limitation for conventional one-stage recognition methods. Third, we propose a two-stage method of static hand gesture recognition that combines finger detection and HOG features. When hand regions can be detected effectively, the proposed method can perform satisfactorily with real-time efficiency. In the future, we will design a practical system for hand gesture recognition that employs a robust hand detection approach. Furthermore, the presented finger detection method can be used to normalize rotations as necessary.关键词：human-computer interaction;static hand gesture recognition;hierarchical strategy;HOG features;support vector machine4626|310|12更新时间：2024-05-07
- 摘要：The fast recognition and classification of objects with rectangular shapes are an essential component in the automatic detection of printed circuit board (PCB) defects, such as through automatic optical inspection or automatic X-ray inspection. The image obtained generally covers only a certain part of a PCB that contains rectangular shapes. The amount, positions, orientations, and exact shapes are almost undetermined. Most traditional algorithms for rectangular region detection are based on the Hough transform or on its derivations. In these approaches, each potential region must be voted on in a 2D or 3D domain. This process increases time complexity and lowers computation speed. By contrast, shape angle theory is an effective method for detecting rectangular regions in contour inspection. Shape angle theory can avoid the inaccuracy and computational inefficiency induced by Hough transform. The accuracy and time complexity of shape angle calculation depend on the efficiency of the tangent calculation at discrete points on the contour. This calculation is currently performed using either Vialard's algorithm or its derived methods. These procedures can be complicated and inaccurate when applied to rectangles with relatively smooth edges.A new method based on Fourier fitting has been proposed in this paper to solve this problem. A polar coordinate system transform is first applied to the discrete points. Then, Fourier series are used to fit envelopes and to calculate the derivative of the envelope function to obtain the tangent. The integrated chip FUSION 878A is derived from a gray-scale image to verify the accuracy and reliability of this method. The average computing time of the proposed method is 1.5775 s. By contrast, the traditional approach reports an average computing time of 156.155 s. The proposed method avoids iterations in handling rectangle contours. Therefore, time complexity is significantly reduced by two orders of magnitude. Moreover, the proposed method generated a precise outcome. The improved approach is applied to rectangle region detection problems. Experimental results indicate the acceptable accuracy and reliability of this algorithm. The following conclusions are drawn: 1) the proposed method exhibits little time complexity when calculating the tangent vector of the rectangle contour. The findings of both experiments and theoretical analysis suggest that specifically, time complexity is reduced from (n) to (n). 2) We examine the effects of breadth-to-length ratio and rectangle dimensions on shape angle. Understanding these effects can provide reference information for template selection and data matching in industrial applications. 3) This approach improves on the shape angle method and can effectively classify rectangles in images by implementing rectangle detection based on shape angle theory.关键词：rectangular object inspection;shape angle;tangent;Fourier fitting3011|220|0更新时间：2024-05-07
- 摘要：Object tracking process is a key step in intelligent surveillance. A new object tracking algorithm based on perception hash is proposed to solve the problem of losing a target caused by mutual occlusion and relatively large scale change of tracking a target, as well as reduce the computational complexity of the traditional tracking algorithms based on template matching. The proposed algorithm introduces perception hash to help in the tracking. Perception hash is a one-way map from a multimedia presentation to a perceptual hash code, which means that the multimedia data that contain similar perceptual contents will result in the same hash code and that multimedia data with different perceptual content will result in different hash codes. Perceptual robustness and security make the perception hash reliable for image identification, retrieval, and authentication. Perception hash is used to generate the hash codes of template images and of foreground target images. Hash code is used for matching. In our algorithm, the hash code is abstracted from the DCT (discrete cosine transform) coefficients of an image and is a binary string. In this paper, we use hamming distance between the hash codes of template images and of foreground target images to distinguish similarities. When tracking, the new algorithm searches for the most similar object as the optimal match template for each moving target in the upcoming frame, and the objects that marched are the foreground regions detected by VIBE. The tracking process is called matching strategy. The optimal match template found in the matching strategy records the accurate position and dimension information of a movingtorger, but fails to track the targets with similar perceptual content. To overcome the weakness of the matching strategy, a searching strategy is designed for further tracking. The searching strategy searches the most similar image region surrounding the moving target to be tracked with the use of diamond search. Subsequently,we divide a rectangular area five times as big as the former one, and then carry out the matching strategy again to get an optimal match template. The accuracy of the optimal match template obtained in each frame is verified by the template evaluation function designed in the proposed algorithm. In addition, an adaptive template update strategy is designed for continuous tracking. The template evaluation function and update strategy ensure good adaptability against occlusion and variation of targets when tracking.Compared with the NCC, mean-shift, and the compressive tracking algorithm, the proposed algorithm is more robust when the target is occluded or has large scale change. In addition, the proposed algorithm has lower computational complexity and time cost is reduced by 6.2%, 6.3%, and 9.3%, respectively, compared with the three algorithms previonsly mentioned. The proposed object tracking algorithm features better adaptability against occlusion and variation of targets when tracking and has lower time cost. The algorithm can help to build a real-time tracking system.关键词：target tracking;mutual occlusion;large scale change;perception hash;adaptive template update;template matching3532|471|4更新时间：2024-05-07
- 摘要：Complicated tracking environments and changes in object appearance are the major causes of failure in visual tracking. The regression tracking algorithm facilitates tracking by establishing a regression model on the basis of information regarding an object's appearance. However, this algorithm displays low tracking efficiency. The use of a tracking-by-detection algorithm that is based on circular structures can improve this efficiency, but such an algorithm is unsuited for handling changes in the scale of an object. Therefore, a scale-adaptive regression tracking algorithm based on fast Fourier transform is proposed to address these problems. First, the algorithm determines the center position of the object in the search region using the regression model. The algorithm then estimates the ideal scale by considering the weight image of all pixels in the candidate region. Comparing with the popular algorithm such as CBWH、IVT and so on, the results from six experiments indicate that the proposed algorithm can not only adapt to changes in background, object scale, and pose, but that the running time of each frame is short as well.This paper presents a scale-adaptive regression tracking algorithm based on fast Fourier transform. It owns good robustness and efficiency for tracking target with the changing of background, object's scale and pose.关键词：visual tracking;scale adaptation;kernel ridge regression;fast Fourier transform2858|275|0更新时间：2024-05-07
- 摘要：Color and intensity which are attributes of a point cloud could be effectively used in the feature extraction of a point cloud. Base on the geometrical and spectral characteristics of the flat target, this paper presents a fast flat target extraction algorithm to determine the coordinate of the target center. First, spectral Euclidean distance function is adopted to classify the point cloud based on color information and then a method of color resampling is used to improve the color consistency in the same category. The extraction of the target area is implemented by the values of RGB in terms of the spectral property of the flat target. In addition, the target's validity is confirmed by measuring the area of the extracted circle field. Finally, according to the intensity values of the points, the center of the target area is determined. The feasibility and accuracy of the proposed method are demonstrated by two groups of experiments. The first experiment was conducted by fixing the distance to analyze the extraction accuracy of the center of the target. Comparing with the mean square error of 11.78 mm of the manual extraction method, the mean square error of automatic extraction of the proposed method can reach to 3.31 mm, which is obviously more accurate than the method of manual extraction; and the second experiment involveal an analysis by arranging the targets in different places to get the center coordinates. Through the analysis the following results can be obtained that the extraction accuracy could be better than 2 mm at the distance of 5 m, and be better than 4 mm at the distance of 10 as well as 5 mm at the distance of 15 m, respectively. The feasibility and accuracy of the proposed method are also demonstrated by the experiments conducted in this paper as well.关键词：attribute information;feature extraction;flat target;point cloud;classification3043|467|1更新时间：2024-05-07
- 摘要：Sparse representation is a popular topic in computer vision and pattern recognition. This process first expresses the test sample as a linear combination of training samples. Sparse representation also determines the class that minimizes deviation for classification purposes. Recent advances in representation-based classification methods show that collaborative representation, rather than sparsity, improves face recognition accuracy. Unlike sparse representation, collaborative representation is computationally efficient and recognizes faces well. However, this performance degrades sharply if the training samples are corrupted by noise. This noise is very common in practical applications and causes side effects that destabilize classification results. Collaborative representation also ignores data locality, which is important. To address these two problems, a new algorithm for locality-constrained collaborative representation classification is proposed in this paper for robust categorization. Singular value decomposition is performed to remove noise in the training samples, and the training samples are approximated as the clean training samples. Local similarity is employed to maintain the similarity between the test sample and its adjacent training samples. Locality is an important characteristic in the reduction and classification of dimensionality because locality results in sparsity but not vice versa. The proposed algorithm can obtain a closed-form solution through collaborative representation and avoid the dilemma of expensive computations in sparse representation. This algorithm also considers local similarity. Experiments are conducted on ORL, Extended YALEB, and PIE face databases. Results demonstrated that the obtained coefficients display much discriminative power and that the proposed algorithm performs well. Moreover, recognition rates reach 91.4%, 93.8%, and 93.2% respectively. The proposed algorithm simply and feasibly reduces the effect of noise in the training samples and obtains “clean” samples for representation-based classification. The proposed method measures dissimilarity by applying locality. This technique also suppresses the corresponding weight coefficient for distant samples while emphasizing the role of samples that are similar to the test sample in the training set. The test samples are represented as well. Experimental results demonstrate the feasibility and effectiveness of the algorithm. Thus, this method is a new technique for representation-based classification, such as face recognition, in that the side effects of noise on classification can be eliminated. This elimination is almost impossible for conventional methods.关键词：collaborative representation;sparse representation;locality;robustness;face recognition2882|253|2更新时间：2024-05-07
- 摘要：Human modeling has long been researched extensively and has many potential applications, such as in 3D games, animation, and education. Therefore, simple and efficient methods of generating 3D models of human bodies, especially geometric models, has been a significant research theme in the computer graphics field. Researchers have recently recognized the intuitiveness and importance of sketching as a tool for 3D modeling and design. In this paper, we present a novel sketch-based method for 3D human modeling through template deformation. We develop a sketch-based method for human body modeling that enables users to generate various models by sketching human contours. First, we propose a method of localizing human body joints to recognize human joint locations from sketches by users. Second, we use a skeleton recognition method to infer the human skeleton structure on the basis of the constraints on the human body. Then, the physiques and shape parameters of the body are interpreted according to the inputted human sketch. Finally, we apply a template deformation method to construct 3D human bodies. This method can incorporate the human body parameters extracted from the sketch into a plausible 3D human model. Specifically, this model is a parameter-driven 3D human body model that consists of the skeleton and surface models.Experimental results demonstrated that the proposed method for 3D human body modeling can automatically infer the intended pose and shape of the body from the sketch. This method can also transfer this information to the parameterized template of the human body. As such, the constructed 3D human model can be used effectively in human animation design.In this paper, we propose a sketch-based method for 3D human modeling through template deformation. This method can assist animators in 3D human modelling through sketching and is fast, easy, and low-cost. Moreover, this technique may benefit various users. Thus, this method can be used to generate and easily animate a 3D human model.关键词：sketching;sketch-based modeling;3D human modeling;human animation3205|312|0更新时间：2024-05-07
- 摘要：The algorithm for spectral angle mapping classification is insensitive to the local characteristics of the spectral curves of hyperspectral image pixels, as well as to its radiation intensity. This algorithm is easily affected by noise and dimension disasters as well, thus lowering classification efficiency and precision. This study presents a model based on harmonic analysis and spectral angle mapping (HA-SAM) to classify hyperspectral imagery.HA technology was used to convert hyperspectral imagery from the spectral dimension to the feature dimension of the energy spectrum. The low-order harmonic component and its characteristic coefficients (harmonic remainder, phase, and amplitude) are extracted, and the spectral vector is replaced with the energy spectrum feature vector to classify hyperspectral imagery with SAM.SAM and HA-SAM are applied in EO-1 Hyperion hyperspectral image classification. The superiority of HA-SAM is verified through contrasts and analysis. This model also exhibits strong universal applicability on the basis of AVIRIS(airborne visible infrared imaging spectrometer) hyperspectral images.HA-SAM not only improves the efficiency and precision of traditional hyperspectral image classification through SAM but also displays strong applicability and application prospects.关键词：hyperspectral remote sensing;hyperspectral imagery classification;spectral angle mapping;harmonic analysis;energy spectrum3965|391|9更新时间：2024-05-07
- 摘要：This research applies the semi-global matching (SGM) algorithm to generate high-quality urban digital surface models (UDSMs) with airborne oblique images. The oblique photogrammetry system (OPS) was influenced by and significantly affected photogrammetry. Thus, OPS can act as a good image source for generating high-quality UDSMs. UDSMs are more appealing than traditional DSMs to users due to the following advantages: 1) UDSMs comprehensively describe 3D urban surfaces. Therefore, inclined oblique images that are originally designed to determine the textures of overall buildings textures effectively solve the problem of the current DSM product involving space information loss at urban building facades. 2) Image matching and space interpretation are highly reliable in UDSMs. Similar targets, such as building roofs and floors, are often observed in multiple oblique stereo images with different title angles and baseline conditions. Thus, strong constraints based on redundant information can be used to either filter UDSM “noise” or remove false image matches effectively. 3) In UDSMs, the aerial solution to image occlusion is low-cost. Severe collisions usually occur among buildings in the aerial images of urban regions. The traditional solution to aerial photography occlusion involves increasing the degree of overlap among images at the cost of dense flights. This process ensures that the occlusion region is “visualized” in different images. The OPS itself considers the omnibearing observation of the urban scene; therefore, the occluded urban regions can be determined at minimal cost with this system. However, oblique image processing is challenging in terms of automation and quality. For instance, the stereo matching of oblique images is subject to various obstacles, such as significant illumination differences, severe occlusions, discontinuous object boundaries, and low or repetitive textures. To address such problems, a new approach is proposed based on the improved SGM algorithm and perspective projection-induced, multi-view UDSM fusion to generate UDSMs from airborne oblique images. The proposed approach is composed of two stages: 1) path-constrained SGM pyramid matching, in which the SGM algorithm is applied to oblique images that are subject to severe perspective distortion and long-range disparity search. To this end, reliable tie points from photogrammetry automatic triangulation (AAT) are used as unchanged “anchors” to block the propagation of SGM mistakes along the cost path. An effective initial disparity map is generated from these tie points through thin-plate spline transformation for implementing the pyramid version of SGM.2) Multi-view UDSM merging based on perspective occlusion judgment, where UDSM point clouds from the titular oblique view are projected onto the nadir image for occlusion judgment and overlap test. As a result, projected UDSMs are classified into occlusion, redundant, and non-redundant collections. A UDSM is obtained with complete space information by removing redundant point clouds from the titular oblique UDSMs and by merging the remaining point clouds with nadir UDSMs.Five airborne oblique stereo images are derived from OPS, and cameras are positioned in the configuration of a Maltese cross. The selected images are tested using the proposed approach to generate UDSMs for local urban regions. The proposed approach not only generated a dense disparity map for each oblique stereo image more effectively than the original SGM algorithm did, but the method also produced high-quality UDSMs that describe 3D urban surfaces more completely than traditional UDSMs.The following results are obtained: (1) the incorporation of constraints from reliable AAT tie points and the implementation of pyramid matching improve SGM performance with respect to the dense matching of airborne oblique images. In particular, this improvement is observed with regard to blocking the propagation of SGM mistakes along the cost path, reducing search range disparities, and saving on-line memory requirements. As a result, a high-quality disparity map can be produced at low computational cost. (2) As per the analysis of perspective projection, most building facades and occluded ground surfaces can be extracted automatically from four titled UDSMs. This process compensates for the deficiencies of nadir UDSMs while removing redundant point clouds simultaneously. As a result, high-quality UDSMs can be generated that describe 3D urban surfaces completely. These findings act as a foundation for actual orthoimage generation and 3D city modeling.关键词：DSM (digital surface model);oblique photogrammetry;dense matching;SGM (semi-global matching)3485|657|4更新时间：2024-05-07
- Postal code：100190
- Tel：010-58887035/58887030/58887418 Email：jig@aircas.ac.cn
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备05080539号-4
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0