Video Dynamic Texture Feature Extraction And Segmentation Technology Research And Implementation

With the advent of the information age, more and more digital media and digital terminals are used to record, send-receive information, the digital image is one important way of recording information. The video is multi-frame image composition, so this gives a great challenge for dynamic feature of the video, as a class of complex dynamic texture movement which attracted the attention of many scholars. To express the dynamic texture recognition, segmentation and expand applied research will help us analyze the complex motion models, to realize the potential applications in video surveillance, video retrieval, motion recognition and other areas.Firstly, the definition and the related significance textures were introduced, and then describing the common dynamic texture model to analyze the texture of each model in a dynamic field of study treatment, as well as practical significance, for example, the dynamic texture detection technology, classification technology, segmentation technology. This thesis then describes the common texture characteristics, and the usual four GLCM characteristics (energy, contrast, correlation, entropy) conducted a comparative experiment to analyze the texture image corresponding to each characteristic texture feature. Another commonly described in detailed texture feature (LBP feature) of the core ideas, Meanwhile, this thesis focuses on the dynamic feature for dynamic texture description of video and introduces to describe the dynamic feature of the video features the essential attribute of strength, which can conduct video scene with similar classification and classification good video episode, according to video the dynamic feature of strength to the video sort order. LBP texture feature based on dynamic as well as the improved algorithm traditional VLBP, LBP-TOP and other feature. Both EOH and BGC3features are improved, and expanded space-time domain to the video. Make the EOH-TOP feature and BGC3-TOP feature successfully apply to dynamic texture video classification. In the experiment, though the feature is very large of dimension VLBP, this thesis to do only the characteristics of LBP-TOP, EOH-TOP and BGC3-TOP. Experimental results show that the improved feature algorithms named as EOH-TOP and BGC3-TOP have improved the recognition rate.This thesis also describes several current dynamic texture segmentation algorithms which are used in the field of video segmentation algorithm, and describes this almost on the latest algorithm in detail, and for the corresponding experiment. Then, on the basis of traditional optical flow method, we propose a new concept-the residual optical flow mapping. Successfully applied to a scene containing the complex dynamic scenes for the foreground objects. Presented a video foreground object segmentation algorithm under the complex dynamic scenes. To compare the segmentation results between our algorithm and the several common segmentation algorithms, we have given the subjective and objective evaluation, experimental results show that our algorithm has more advantages.