最新刊期
卷 14 ， 期 11 ， 2009
- 摘要：The expression generation is one of important research contents of intelligent human-computer interface.Because images are the research objects in the majority,the facial expression generation techniques based on the study method dont need to model for 3D faces and potentially can be of highly realistic expressions.These are of considerable advantages in reality improving and complexity reducing.Some methods can also be used to 3D models.The paper summarizes the published researches from the aspects of morphing,expression mapping,statistical method,2D meshes,facial color expression and performance-driven.The conceptions,principals and the methods of typical achievements are discussed.The features and existing problems of different algorithms as well as the development trend are analyzed.It provides some useful information and gives a reference to related works.关键词：expression generation;morphing;expression mapping;expression ratio image;facial color expression;performance-driven3051|271|0更新时间：2024-05-07
- 摘要：In order to reduce the computational complexity of fast motion estimation，and avoid the local minimum problem caused by small diamond search,a motion estimation algorithm based on adaptive search pattern was proposed.This algorithm chose search pattern adaptively depending on SAD changing speed and the temporal relativity in video sequence.Experimental results demonstrated that by using this algorithm,the encoded bit-rate changes within ±0.6% compared with diamond search and hexagon-based search;searching points were reduced to 72%~77% and 83%~86% compared with the diamond search and the hexagon-based search respectively.In the condition of reducing searching points this algorithm avoids local minimum problems.关键词：motion estimation;adaptive;search pattern3059|252|0更新时间：2024-05-07
- 摘要：A unified high-performance and memory-efficient architecture is proposed to perform 5/3 DWT and 9/7 DWT in JPEG2000 with a novel modified lifting algorithm.By applying the proposed lifting scheme,the transposing buffer between the row processor and the column processor is eliminated,resulting in a reduction of the internal memory requirement.For an N×N image,only 2N internal memory is required for 5/3 DWT and 4N is required for 9/7 DWT to perform 2D DWT with the critical path limited by one multiplier delay by employing the pipeline technique.Compared with the existing 2D DWT architectures,the proposed 2D DWT architecture has the advantage of regular structure,low memory requirement and high system performance.The proposed architecture was described with Verilog HDL language and verified to be correct in ModelSim.It was also synthesized,placed and routed on an Altera StratixⅡ FPGA EP2S60F1024C4 using Quartus Ⅱ version 5.0 toolset.The experimental result shows that 1 284 ALUTs are utilized with memory size of 4 K words for 1 024×1 024 image and the operating frequency can be up to 172.56 MHz.关键词：discrete wavelet transform (DWT);lifting;JPEG2003295|256|0更新时间：2024-05-07
- 摘要：The copyright protection of digital videos is becoming more important with the development of multimedia technology.A novel adaptive video watermarking scheme integrated of several techniques is proposed in this paper in order to obtain one practical watermarking.The scheme is robust against manifold attacks and the watermark can be retrieved blindly.First,the watermark was scrambled by Arnold scramble and Turbo coded to obtain the improved security.Secondly,the key frames were selected based on the video characteristics such as frequency sensitivity,chroma sensitivity,movement sensitivity,texture sensitivity and luminance sensitivity which were summarized in this paper.Then the key frames were segmented to three dimension blocks and the watermark was embedded into the middle-low frequency domain of the three dimension wavelet decomposed coefficients of the blocks.In addition,a new method was adopted which can extract the invariable information of rotated and zoomed videos by Zernike movement.The invariable information can be used to conquer rotating and zooming attacks.Finally,the watermark was retrieved blindly.The robustness was obtained with attacks and tests to the typical videos.Experimental results demonstrate that the scheme is accommodated to videos with various content characteristics and can conquer the present attacks.It has certain practical values.关键词：video watermark;watermarking attacks;Turbo code;Zernike moment;video segmentation3231|287|0更新时间：2024-05-07
- 摘要：To improve image de-nosing effect,in this paper,a new method based on discrete cosine transform (DCT) in Contourlet domain is proposed,which utilizes noise energy rather than relying on accurate estimation of noise variance to perform image de-noising.This new method not only overcomes the psuedo-Gibbs phenomenon of the hard-thresholding method but also can avoid signal distortion which is caused by the soft-thresholding.Experiments show that the performance of the proposed method is obviously superior to other methods both in vision and in PSNR and is of more practical use.关键词：image de-noising;wavelet transform;Contourlet transform;discrete cosine transform（DCT）3356|255|0更新时间：2024-05-07
- 摘要：In order to preserve image fine details while de-noising,a new noise detection and adaptive filter method,which is based on fuzzy reasoning technique,was proposed.At first,according to the local statistic information ROAD(rank-order absolute differences) and orientational laplacian differences,the possible noises was detected with fuzzy reasoning technique.Then possible noise was filtered with an adaptive method,which can preserve details to a great extent by keeping uncontaminated pixels unchanged.Lastly,some possible error judged noise caused by local high noise intensity and high noise image were corrected with an improved filtering approach，in which relatively possible noise was located and filtered iteratively,each time only the most possible noise was filtered.Experimental results show that with appropriate iterative the proposed method is efficient for different noise intensities.关键词：fuzzy reasoning;noise detection;adaptive;de-noising2845|291|0更新时间：2024-05-07
- 摘要：In order to improve the quality of the state-space exploration and the accuracy of visual tracking,in this paper a particle filter algorithm based on maximum close distance(MCD) and partial linear Gaussian models is presented.MCD avoids the problem that each pair of pixels in the image contribute to the matching result equally.The proposed method uses neighborhood between pixels as the matching similarity.The correlation curve obtained in this way is much sharper.So the image matching method has high matching precision.A direct consequence of using partial linear Gaussian models is that the optimal importance function is adopted.The combination of them will be the optimal particle filter.The stability of the algorithm has been improved due to the robustness of MCD.Two simulated experiments are finally carried out to confirm the validity of the improved algorithm.关键词：particle filter;optimal importance function;validation gate;maximum close distance(MCD)3045|275|0更新时间：2024-05-07
- 摘要：Aiming to detect interesting targets in polarimetric SAR imagery,the problem of target contrast enhancement is studied in this paper,and a novel scheme based on the strategy of sample selection is proposed for heterogeneous areas having complex polarimetric properties.Firstly,a preliminary sample selection is achieved by treating two categories of pixels with different scattering mechanisms as sample pixels of target and clutter.Secondly,a further selection is complemented by a test statistic of wishart distribution,which treats pixels with similar polarimetric statistics as the final qualified samples.The novel scheme can enhance the performance merely acquired merely by the rough selection of pixel samples with manual intervention,and its validity is indicated by experimental results with E-SAR polarimetric SAR data sets.关键词：polarization;synthetic aperture radar(SAR);target contrast enhancement;sample selection3208|252|0更新时间：2024-05-07
- 摘要：Video steganography has the advantage of large embedding capacity,but the security is usually neglected when obtain the capacity.In order to balance the two aspects,a secure steganographic algorithm for embedding secret information in compressed video is proposed utilizing the principle of steganography based on error-correcting code.The strategy of two-time embedding is designed to embed secret bits.In other words,the secret bits are firstly embedded into the error-correcting codeword.Then by modifying the level value of run-level pair corresponding to the DCT coefficient which is not differentially encoded,the stego error-correcting codeword obtained by first-time embedding is secondly embedded.At the same time,measures are taken for the purpose of preserving video stream size.Experimental results show that the algorithm has a good visual and statistical imperceptibility,and satisfies the Kerchhoff rule in cryptography.Furthermore,the algorithm has a considerable embedding capacity,and can keep the nearly invariant size of video stream.关键词：information hiding;steganography;compressed video;error-correcting code;security3713|222|0更新时间：2024-05-07
- 摘要：Buried small targets usually have a very low signal-to-noise ratio(SNR) in a single frame image obtained by the vehicle-mounted forward-looking ground-penetrating imaging radar(VFGPIR) system.This will lead to poor detection performance of the single image.One special property of a forward-looking detection system is that the same spatial location is usually observed for multiple times,each at different sand-off distances,as the vehicle moves ahead.Thus a non-coherent range multi-look accumulation method is presented by exploiting the sequence image obtained by VFGPIR in this paper.The range multi-look processing can be achieved on line by using the distance sub-band partition and iterative operation techniques.The relationship of detection probability,false alarm probability and SNR between the multi-look image and the single image is derived,and the formulation to determine the accumulative frame number is given.Experimental results of the real data show that the SNR of targets is obviously improved and the speckle is also well suppressed after the multi-look processing.Moreover,the proposed approach can obviously improve the quality of image and the detection performance while having a relatively small computational load.关键词：Vehicle-mounted forward-looking ground-penetrating imaging radar;range multi-look processing;distance sub-band partition;iterative operation;detection probability3176|226|0更新时间：2024-05-07
- 摘要：In this paper we introduce a new MAP-based super-resolution method,which can effectively preserve image edges by using an improved directional image prior model.The proposed image prior model,which we refer to as a weighted directional Markov image prior model,utilizes different weights for different directional smoothness measures of the edge pixels.Definitely,larger weights are chosen for smooth measures along the edge in order to penalize them to a larger extent and smaller weights are chosen for smooth measures across the edge for a less penalization.Thus,the edges of the reconstructed HR image can be effectively persevered.The proposed algorithm is tested on different series of images.The experimental results indicate that the proposed algorithm has considerable effectiveness in terms of both objective measurements and visual evaluation.关键词：super-resolution;MAP;edge-preserving;weighted Markov random field prior3154|259|0更新时间：2024-05-07
- 摘要：An error concealment algorithm for a totally destroyed low-frequency sub-band in Motion JPEG2000 coding is proposed in this paper.Firstly,it applies the sobel operator to a high-frequency image to outline the objects.According to the correlation between those video frames,it then utilizes the temporal information to judge the motion vector between the reference frame and the destroyed one.Under the consideration of motion vector,motion compensation or replacement is used to reconstruct the damaged low-frequency sub-band with the aid of the reference frame’s data.The proposal method is low complexity and thus suitable for real time application.It takes the human visual characteristics into consideration and verifies parts of the result of motion estimation by changing the block size.Simulations indicate that the reconstructed frames have high subjective visual performance.2918|455|0更新时间：2024-05-07
- 摘要：An improved double test algorithm in Hadamard transform(HT)domain is proposed in this paper.First,two kick-out conditions in original DTA method are rewritten properly in HT domain.Then,codewords and input vectors in HT domain are properly partitioned based on the result of the preliminary experiment.Finally,by using the better kick-out condition many unmatched codewords can be eliminated.The simulation shows that the search space of the improved algorithm in this paper is reduced to about 14%～17%,and the encoding time is reduced to about 35%～45% compared to the original algorithm.关键词：vector quantization;fast codeword search;Hadamard transform;single test algorithm;double test algorithm3006|222|0更新时间：2024-05-07
- 摘要：Due to high bandwidth resource requirement,how to support the digital video cassette recorder(VCR)functionality is a key problem in practical video on demand(VOD) systems.Based on the investigation of the impact of the frame rate on quality of perception(QoP),an effective algorithm which selects frames by cooperatively adjusting the frame rate and position is proposed.The proposed algorithm presents a VCR operation model,and adds two constraints conditions to the fast forward and fast backward operations: bandwidth requirement and time limit.Based on MPEG dual-stream,the algorithm meets two constraint conditions by adjusting the frame rate and the frame’s position which would be displayed.Simulation results show that the proposed algorithm can effectively reduce the network bandwidth requirement of the VCR operation,while keeping good QoP.The proposed algorithm outperforms the dual bit-stream algorithm,which only needs about 90 percent of the mean bit-rate bandwidth.The method was already successfully deployed in Shanghai national high-tech research and development plan(863)IPTV systems.关键词：video on demand(VOD);video cassette recorder(VCR);quality of perception(QoP);MPEG dual bit-stream3469|232|0更新时间：2024-05-07
- 摘要：The contrast reduction of remote sensing images is mainly because of the reduction of the modulation transfer function(MTF)of the optical system.And MTF represents the characteristic of the light energy distribution of the optical system.By analyzing the theory of contrast reduction,this paper presents a remote sensing image enhancement method based on energy distribution.Firstly,the image energy at the low frequency domain is decreased,and then the decreased image energy is allocated to the high frequency domain to obtain the enhancement result.The experimental results demonstrate that the proposed method is simple and effective,which is feasible for remote sensing image enhancement.关键词：image enhancement;energy distribution;modulation transfer function(MTF);remote sensing image3115|315|0更新时间：2024-05-07
- 摘要：Numeral strings segmentation plays a significant role in the OCR systems.In many applications,numeral strings are filled in preprinted form frames.This makes,the segmentation problem easier.Other wise,the segmentation is difficult.A new segment method for handwritten numeral strings is proposed.Principal curves are selected to extract strokes of characters.The strokes in the initial group are disposed of by the fuzzy features and grouped based on the confidence of the classifiers.On the database composed of 3 000 bank checks with touching digits in 363 checks,the proposed algorithm has been evaluated qualitatively and quantitatively with an the accurate rate of 89.68%.关键词：principal curve;fuzzy feature;segmentation of numeral strings;stroke grouping3610|263|0更新时间：2024-05-07
- 摘要：An ameliorated algorithm that combined support vector machine (SVM) with k nearest neighbour (kNN) is presented and it comes into being as a new classifier,based on the research that SVM classifies some tested samples in error nearby the optimal super-plane.In the class phase,the algorithm computes the distance from the tested sample to the optimal super-plane of SVM in the feature space.If the distance is greater than the given threshold,the tested sample will be classified on SVM,otherwise,the kNN algorithm will be used based on the best distance measurement.The numerical experiments show that the mixed algorithm improve the accuracy compared to the sole SVM.关键词：SVM(support vector machine);kNN(k nearest neighbour);generalization error;best distance measurement6109|284|0更新时间：2024-05-07
- 摘要：Given an input video sequence of one person who conducted a sequence of continuous actions,we consider the problem of jointly segmenting and recognizing actions.To recognize the activities in videos,we propose a discriminative approach to this problem within a semi-Markov model framework,where we are able to define a set of features over input-output space that captures the characteristics on boundary frames,action segments and neighboring action segments,respectively.A Viterbi-like algorithm is devised to help efficiently solve the induced optimization problem.Experiments on a variety of datasets demonstrate the effectiveness of the proposed method.关键词：activity recognition;semi-Markov;support vector machine(SVM)3188|264|0更新时间：2024-05-07
- 摘要：In order to develop linear local tangent space alignment to supervised learning algorithm,an algorithm called orthogonal discriminant linear local tangent space alignment is proposed.The algorithm makes use of class information of the data samples to compute the interclass scatter matrix.Then we modify the objective function of the original algorithm,constructing the new optimization problem.Moreover,on this basis,the algorithm orthogonalizes the subspace to obtain the orthogonal one.The effectiveness of the algorithm has been verified on two standard face databases.With local tangent space representing for local geometrical structure of the manifold of the data samples，the algorithm fuses discriminant information and orthogonal technique to preserve the local geometrical structure and discriminant structure,and the algorithm improves the recognition performance.关键词：face recognition;orthogonal discriminant linear local tangent space alignment;Manifold learning;subspace3263|244|0更新时间：2024-05-07
- 摘要：Mosaic pictures,stitched up from many small photographs with some special symbols or rich meanings,have already been broadly applied in advertisement,artistic designs and drumbeating.This paper presents a series of semiautomatic methods to generate this kind of pictures.This method includes a matching model,which can arrange small photographs into a mosaic with an optimization algorithm,a model modifying some less matchable photographs in the HIS color space and a series of artifices to improve the visual effect.It can be concluded from the experiments on several symbols of the 2008 Beijing olympic that the series of semiautomatic methods can make mosaic pictures well and provide the materials of a unique style,for advertisement and flacks.关键词：mosaic;image matching;HIS color space3175|267|0更新时间：2024-05-07
- 摘要：The registration of heterogeneous images is one of the most important steps of multi-resource image fusion.Heterogeneous image registration algorithms need to extract and match the invariant feature precisely,which can be difficult due to huge differences between optical and SAR imaging principles.This problem can be avoided by introducing implicit similarity existing in the same scenes.The algorithm in this paper constructs the implicit features by large scale pixels and then similarity criterion function is constructed using pixels migration.The registration results can be obtained by optimizing criterion function through genetic algorithm(GA)’s global optimum solution.Experimental results show the registration accuracy of this algorithm achieves pixel level.关键词：synthetic aperture radar(SAR);implicit similarity;genetic algorithm(GA);image registration3398|243|0更新时间：2024-05-07
- 摘要：In order to realize the accurate deformation of the parametric surface,a special vector extension function was established to realize the interpolation deformation for parametric surface,in the course of the deformation.First,some deformation auxiliary data on the surface is chosen to construct the extension vector function,which is used to make the surface deform based on vector addition and ensures the deformed surface passing through the given position data.The deformation results show that the method can not only control the deformation region accurately but also make the deformation domain and the undiformation domain of the surface connect to C2 continuity.关键词：parametric surface;interpolation deformation;vector extension function3185|209|0更新时间：2024-05-07
- 摘要：Focusing on the fuzzy C-means algorithm’s problem that the cluster quality is greatly affected by the data distribution and the stochastic initializing the centrals of cluster,a single-point approximation weighted fuzzy C-means algorithm is proposed by using the part of prior samples information.After the probability statistics of original data is conducted,the weights of data attribute are designed to adjust to the uniform distribution,and then are added in the process of cyclic iteration.What’s more,in order to significantly improve the convergence speed and the cluster precision,the proper initial cluster centers are chosen by the single adjustment algorithm,which can also overcome the selection influence of prior samples.In addition,combined with the characteristics of remote sensing data,the modified algorithm is updated for remote sensing image cluster.With the comparison experiment of the UCI data sets and the Zhalong wetland remote sensing data,the real validity of proposed algorithm is proved.关键词：cluster analysis;fuzzy C-means;initial centre of cluster;attribute of the weight3070|228|0更新时间：2024-05-07
- 摘要：To effectively identify and interpret ambiguous zones in handwritten Chinese image,a new method for handwritten stroke extraction based on ambiguous-zone detection is proposed.In the method,a candidate set of fork points are extracted using a thinning algorithm,and the ambiguous zones are detected according to the fork points and the contour beside them.Then,the sub-strokes and ambiguous zones can be modeled with a graph,and a Bayesian classifier is built to analyze the continuity of sub-stroke pairs.Finally,sequences of sub-stroke are achieved by searching paths in the graph,and thinned strokes can be retrieved by B-spline interpolation.Experimental results show that the proposed method is effective and accurate for both ambiguous-zone detection and stroke extraction compared to other methods and reduce the shape distortions in ambiguous zones to an acceptable level.关键词：stroke extraction;ambiguous zone;thinning algorithm;writing sequence3515|263|0更新时间：2024-05-07
- 摘要：Virtual human deformation is one of the most important research focuses,and remains problematic in the research field of 3D garment CAD system.In this paper,we present multi-factor driven deformation methods to deform virtual human models.The driven factors are classified into three classes: size,posture and local figure.Three deformation methods are proposed to deal with the three factors.First,we use a feature size driven method to change the virtual human size,and this method can be divided into cross-section ring based method and templates interpolation based method.Then we implement a posture driven deformation to achieve a specific human body posture,and establish postures library for rapid gesture matching.Finally,we apply a local figure driven deformation which is achieved by changing the size or shape of the feature cross-sections to obtain the virtual human with local feature.Some given examples show that this method is very flexible to have diverse virtual human models for garment CAD.关键词：garment CAD（GCAD）;human model deformation;size driven deformation;posture driven deformation;local figure driven deformation3168|248|0更新时间：2024-05-07
- 摘要：The traditional fractal method for terrain generation has an isotropic fractal feature.In order to generate anisotropic terrain,a new approach to generating anisotropic terrain based on fractal is proposed in this paper.This approach uses the combined fractal Brown surface(CFBS) model and incorporates two fractal Brown surface(FBS) with different fractal features.The fractal features of the corresponding terrain are anisotropic.Finally,some simulations based on CFBS are conducted and the final figures of the corresponding terrains are shown in this paper.The analyses of the simulation results show that the fractal feature is anisotropic,which is accordant to the actual terrain.The simulation results verify the validity of the CFBS model.关键词：fractal Brown surface(FBS);random midpoint displacement(MPD);isotropic;combine fractal Brown surface(CFBS)3442|224|0更新时间：2024-05-07
- 摘要：A class of blending function with shape parameter is presented in this paper.It is an extension to the degree′n Bernstein basis function.Based on this blending function,we define a class of adjustable quasi Bézier curve.The blending functions and quasi Bézier curves have the most properties of Bernstein basis and the Bézier curves.Moreover the shape of the quasi Bézier curves with the same control polygon can be adjusted by changing the shape parameter value.Using this method,quasi Bézier surfaces with parameters are constructed.Experiments show that the method given in this paper is intuitive,effective and easy to control.关键词：blending function;shape parameter;quasi Bézier curve2873|314|0更新时间：2024-05-07
- 摘要：A novel disparity matching algorithm is proposed.A color segmentation based adaptive weight algorithm is adopted to improve the accuracy of distinctiveness estimation,which has an important influence on the result of disparity matching.Besides,we take advantage of the disparity correspondence of the pixels in the same color region to reduce the range of disparity search and to decrease the computation cost.We use the standard test images on the Middlebury website to evaluate our algorithm.Experimental results show that,compared with the DSM method,our algorithm has a smaller error rate and faster computational speed.关键词：disparity matching;color segment;DSM3287|247|0更新时间：2024-05-07
- 摘要：Accurate sub-pixel image registration is a key problem in image super-resolution reconstruction.Optical flow methods based on pixel feature,which are widely used in image super-resolution reconstruction,are difficult to achieve registration of sub-pixel accuracy for large motion field.This paper considered a robust multi-frame image super-resolution reconstruction method based on SIFT.Firstly,SIFT operator was used to pick up keypoints and their descriptors of input low-resolution image pairs which are to be registered.Then the candidate keypoint pair was selected,outliers were wiped off through RANSAC,and images pair displacement was computed at the basis of assumed transitional geometry constraint model.Secondly,initial reference frame was selected from vision center frame or specified image frame.Lastly,super-resolution reconstruction was done through conventional super-resolution reconstruction framework.Experimental results show that the proposed image super-resolution reconstruction method based on SIFT is feasible,and the quality of super-resolution reconstructed images is better than those of classical methods by both subjective evaluation and objective standards.关键词：super-resolution;image registration;SIFT;keypoint;optical flow;reference frame3819|272|0更新时间：2024-05-07
- 摘要：For un-calibrated parallel cameras,a new algorithm using affine transformation is proposed in this paper.This algorithm uses Harris operator to obtain corners,and determines corresponding image points with characteristic tracking method,and segments affine transform theory,and then constructs the affine matrix to make disparity block estimation.It has been proved that the algorithm can improve the result of the original motion estimation.关键词：corner detection;homogeneous coordinates;affine-transformation;block match3449|238|0更新时间：2024-05-07
- 摘要：This paper proposes an improved belief propagation algorithm for stereo matching.Based on the assumption that the disparity field is continuous,traditional methods regard the disparity field as a Markovian network that transmits two-way information.But in the occluded area,disparity is not continuous.So we propose a new method.Firstly,we use the cross-check technology based on the initial disparity to detect the occluded area.Secondly,we regard the disparity map as a mixed network of Markovian field and Bayes filed.Then the occluded area does not transmit information to the non-occluded area so as to reduce the computational cost of disparity matching.We use the standard test images to evaluate our algorithm.The result shows that the proposed method achieves a high accuracy and efficiency.关键词：stereo matching;belief propagation;Markovian field;cross-check technology3143|266|0更新时间：2024-05-07
- 摘要：When creating a digital forgery,it is often necessary to combine several images which were originally of different JPEG compression qualities.This paper proposes a passive approach to detect digital forgeries by checking inconsistencies of blocking artifact.We calculate the second-order difference edge map by convolving the blocky image with a Laplacian kernel.Then compute the mean of the results along the horizontal (vertical) direction and obtain a vector.Discrete Fourier transform of this vector produces a normalized spectrum curve and builds a blocking artifact measure.Divide the composite image into overlapped blocks and calculate the measure of each block.The forged area can be effectively exposed by checking inconsistencies of blocking artifact.Simulation results show the efficiency of proposed algorithm.关键词：digital forensics;blocking artifact;tampering identification;passive authentication3502|276|0更新时间：2024-05-07
- 摘要：A passenger counting method is proposed in this paper based on binocular stereovision for collecting the real-time passenger data in public transport.This counting method involves three blocks: moving object detection,object labeling,object matching and counting.Object detection is the vital issue of the counting system.A row frame difference method is firstly adopted to detect possible moving objects.Then,the facticity of each possible object is judged based on disparity analysis.Experimental results show that the proposed method can obtain an on-spot counting accuracy of 92% tested on a large data set capturing from simulated scenarios in laboratory and on moving buses.This approach is effective to avoid illumination changes and shadow of passengers.关键词：passenger counting;binocular stereovision;moving object detection;row frame difference;disparity analysis3407|318|0更新时间：2024-05-07
- 摘要：This paper proposed a multiview video rate control scheme based on human stereo visual feature.First,the new quantitative measure for stereo video quality is introduced.Then,the bit allocation scheme is derived on the basis of this measure.Finally,the macroblock quantization parameters are modified according to the quantization parameter of the neighboring macroblock on the purpose of eliminating block effect in stereo video.The experiment results demonstrate that the proposed scheme can improve the stereo video quality and obtain quite stable rate.关键词：human stereo visual feature;rate control;multiview video3140|212|0更新时间：2024-05-07
- 摘要：Blotch is one of the most common artifacts in archived film.This paper proposed an improved MRF-based blotch detection algorithm named MDV.The MDV algorithm has 3 steps.In the first step,SDIp and ROD algorithms are combined to increase the Recall rate in blotch detection.The definition of the second MRF-based step is restricted within the former result,which reduces the computational burden.The improved MRF-based algorithm then uses the combination of original frame and motion-compensated frame and the new denoising factor to reduce the false alarm rate.At the last step,spatio-temperal based matching algorithm is introduced for further refinement.The experimental results and comparative study with existing algorithms confirm that the proposed MRF method not only has higher recall rate and fewer false alarms,but also improves the computational speed significantly.关键词：digital film restoration;blotch detection;Markov random field3025|230|0更新时间：2024-05-07
- 摘要：Image quality evaluation of the traditional method MSE and PSNR are base on full reference(FR) quality assessment.It requires a reference image,so it is not practical to be used.While the no reference(NR) just can deals with this problem,so it is of great significance.Aiming at the object digital image quality assessment,to the degree of detecting the distance of blur,this paper presents a no reference metric of blur estimation base on the edge detection.After the analysis and comparison of the result,of which the relevance to the subjective quality perception is better than some other blur estimation metric which only detects the edge or estimates the blur.Whats more,since the low complexity of algorithm，the needless for source of reference,and the proposed metric can be practically used.关键词：no reference;image quality assessment;blur estimation;edge detection3802|307|0更新时间：2024-05-07
- 摘要：AdaBoost is a very popular classification algorithm on machine leaning.By studying the characteristics of the weak classifier,this paper proposes two new methods to calculate the threshold and bias of the weak classifier.The two methods make the correct rate of weak classifier larger than 50%,assure the convergence of AdaBoost training when the weak classifier reach a certain number.Simulation experiments show when the error rate is in an acceptable range,the algorithms using fewer weak classifiers will be able to guarantee the strong classifier to maintain a high correct rate.关键词：weak classifier;Adaboost algorithm;stronger classifier;error rate4028|249|0更新时间：2024-05-07
- 摘要：Human motion analysis in an intelligence surveillance system is a hot research topic in computer vision,and temporal segmentation of human activity sequence is the most fundamental step in human motion analysis.In this paper,an unsupervised online temporal segmentation algorithm is presented,and then the segmentation result is recognized by HMM.Firstly,a robust shape encoding scheme is employed to produce a compact representation of human silhouette,and a new feature called contour point set is proposed.Secondly,the intrinsic dimensionality of feature sequence and the corresponding low-dimensional manifolds are determined using SVD,and the break of projecting error of activity sequence on the determinate manifolds is detected as the segmentation point of the activity sequence.Temporal segmentation results are recognized by HMM finally.Experiments on two public databases show the effectiveness of the segmentation and recognition algorithms in this paper.关键词：temporal segmentation of activity sequence;activity recognition;intrinsic dimensionality;SVD;HMM4466|281|0更新时间：2024-05-07
- 摘要：In recent years,the method of content based copy detection (CBCD) is increasingly attracting our attention as internet and wireless network develops quickly.Our paper presents a new fast and efficient system for video copy detection aiming at the particularity of IPTV.The system takes a laconic method to compare the similarity of video frames,clusters the shots into shot-clusters after selecting them from videos and makes up a tree of shot-clusters named K-means tree.The experiment results show that our system has a satisfying searching speed,as well as good rate of covering and precision.关键词：video;copy detection;internet protocol television(IPTV)2852|236|0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