Research On Key Techniques In Video Image Processing

Image sequence based moving object detection, matching, tracking and recognition is an emerging field of computer vision of forefront subject and receiving much attention, which combines computer science, machine vision, image engineering, pattern recognition, artificial intelligence and other advanced technology, widely used in various aspects of human-computer interaction, intelligent monitoring, machine vision navigation, industrial robotics. Video of intelligent information integrating analysis and understanding is a research hot in the field of machine vision. It is including detection of region of interest, target matching, target tracking, target identification, image reconstruction, etc. Although there are many researchers who conduct a lot of related research work, and many simulation experimental results are made. But these techniques are far from the practical requirement. Experiments were devised to test and verify the feasibility and effectiveness of the algorithm, and the specific research contents are listed as follows:(1) The optimization of image fusion is researched. Based on the properties of nonsubsampled contourlet transform(NSCT), shift invariance, multiscale and multidirectional expansion, the fusion parameters of the multiscale decompostion scheme is optimized. In order to meet the requirement of feedback optimization, a new image fusion quality metric of image quality index normalized edge association(SSIM-NEA) is built. A polynomial model is adopted to establish the relationship between the SSIM_NEA metric and several decomposition levels, guide the fusion process, improved the quality of the fusion image, lay a foundation for the subsequent moving object detection.(2) Extracting foreground moving objects from video sequences is an important task and also a hot topic in computer vision and image processing. Segmentation results can be used in many object-based video applications such as object-based video coding, content-based video retrieval, intelligent video surveillance and video-based human–computer interaction. In this paper, we present a novel moving object detection method based on improved VIBE and graph cut method from monocular video sequences. Firstly, perform moving object detection for the current frame based on improved VIBE method to extract the background and foreground information; then obtain the clusters of foreground and background respectively using mean shift clustering on the background and foreground information; Third, initialize the S/T Network with corresponding image pixels as nodes(except S/T node); calculate the data and smoothness term of graph; finally, use max flow/minimum cut to segmentation S/T network to extract the motion objects. Experimental results on indoor and outdoor videos demonstrate the efficiency of our proposed method.(3) Image matching is an important question in computer vision, however, due to the large viewpoint and similar regions, there exist false matches. A robust matching method-DelTri is proposed. First, improved the traditional scale invariant feature transform from four aspects.1) Preprocessing the input image by bilateral filter to preserve the edge;2) Enlarging search ranged of extrema of keypoints detection to diminish the number of keypoints;3) Calculate the magnitude and orientation of gauss smoothed image by sobel operator to smooth the noised;4) K-means cluster filter uncorrected matches. Based on the initial matching of improved Scale Invariant Feature Transform, the matched keypoints are respectively triangulated to create the triangulation net, which can express the overlapped physical structure of the objects. The matched triangles can lead to the final matches.(4) Visual tracking could be formulated as a state estimation problem of target representation based on observations in image sequences. To investigate the integration of rough models from multiple cues and to explore computationally efficient algorithms, this paper formulates the problem of multiple cue integration and tracking in particle filter framework based on interactive multiple model(IMM). IMM can estimate the state of a dynamic system with several behavioral modes that switch from one to another using mode likelihoods and mode transition probabilities. For the problem of visual tracking, the models of IMM are three observation model: Corrected Background-Weighted Histogram(CBWH), Completed Local Tenary Patterns(CLTP) and Histogram Of Orientation Gradients(HOG).The models probabilities corresponding to the weight of multiple cue. IMM then dynamically adjusts the weights of different feature. Compared with state of the art methods, Experimental results demonstrate that this algorithm can track the object accurately in conditions of rotation, abrupt shifts, as well as clutter and partial occlusions occurring to the tracking object with good robustness.(5)For the moving object recognition, taking the human action as an example, integrating motion temporal templates with spatio-temporal interest points based appearance descriptor for action recognition is proposed in this paper. First, we model the background in a scene using our improved VIBE model and extract the foreground objects and binary mask of region of interest in a scene. Then, for the binary mask of object, we constructed motion temporal templates using the motion history image(MHI) and motion energy image(MEI); for the foreground objects, we performed spatio-temporal interest points(STIPs) detector. Thirdly, human action representation is combined with three dimensional Scale Invariant Feature Transform descriptor(3D SIFT) on STIPs and Hu moments extracted from MHI and MEI. Support Vector Machine(SVM) is adopted to classification the actions. To validate the proposed descriptor, we have conducted extensive experiments on the KTH and Weizmann action datasets.(6) For the different problems of the key technologies in the video image processing such as moving target detection, target matching, target tracking and target recognition, base on integration of these techniques to analyze the compacts between different modules, so that each module can mutually link and operate on each other, in order to improve the whole system’s performance, finally achieve reference and use for the relevant departments, and provide valuable reference for the same field of study.