Wavelet Domain Markov Model-based Video Object Segmentation Algorithm

With the development of communications and information processing technologies, the video-driven application show a very large degree of flexibility and extensibility, telemedicine are emerging, Video object information is the important thing which is provided by the telemedicine system. These diversified applications and services with a large amount of data demand more advanced digital processing technologies for efficient memory and transmission, as well as accurate analysis and flexible manipulation. In a new generation of video coding standard MPEG-4,video frame is regarded as to be composed of a series of independent semantic video object, which was encoded independently. Therefore, MPEG-4 is describing encoding using video object, which is the first requirement is that video images are separated into different objects or move objects are separated from the background and then using different encoding method accordingly. This not only can greatly increase the compression ratio, and allows users to operate interactively on content for multimedia data. Video Object segmentation is the basis of object-based encoding, so video object segmentation technology has the important research value and the application significance. Video object segmentation involves in the analysis and the understanding of video content, and the complexity of object based video segmentation increases as the video content becomes complex, and the computation becomes more expensive with the increase of the segmentation accuracy. Therefore, effective video object segmentation is very valuable in theory research and application.In order to improve the accuracy of video segmentation, and to minimize the complexity of the algorithm and computation, and to meet the requirement of video processing and transportation in real time, this paper describes the development and current status of video segmentation, analyzes the performance and efficiency of different segmentation algorithms, and present the basic structure of video object segmentation and feature expression.The quality of the video image affects the result of segmentation greatly. To reduce the influence of noise, we lucubrate the existing algorithmic of denoising method based on wavelet threshold. Through simulation, and signal to noise ratio for the use of indicators, with the visual characteristics of the human eye, the various results based on threshold Denoising were compared, and then a new denoising method based on wavelet threshold is presented. The coefficients including low frequency and high frequency of wavelet transform are magnified properly before using image denoising method based on wavelet transform. Then the noise in image is eliminated by using soft threshold wavelet method. In the last , the coefficients of wavelet transform are reduced aptly and restructured. According to the results of experiment , the given algorithm is good at retaining the edge of the image while suppressing the noise effectively.By lucubrating the existing algorithm of segmentation, the paper presents a novel algorithm of video object extraction based on MRF model using wavelet transform. The algorighm introduced the variable weighting parameter when constructiong the corresponding energy function of MRF, making the model difficult to a local maximum.We adopt the method of window ICM when we solve the superior solution of energy function by Maximum A Posteriori, to attain to the balance of the speed and accuracy, and it will not affect the accuracy of analysis. The experimental results showed that the algorithm retains more of the image information, and gets more accurate video object, and reduces the calculation of the segmentation.Usually, the movements of object is slow, non-rigid and continuous, so the region of motion can be achieved by analyzing the motion preperties of series, combining with space feature of object, exact and complete object can be extracted.To sum up, the paper uses the method of combining wavelet threshold and coefficient magnifing to denoise before the segmentation of video image, and in the segmentation constructing the MRF model in wavelet domain.The paper uses artificial images and several video image sequence to take the simulation object, multiple tested confirms the proposed algorithm can restrain noise effectively and improve the signal to noise ratio , and increase the object boundary element quantity, and extract the moving object exactly.There are many algorithms in video-segmentation based on objects currently, but most of them have poor generality and large quantity of computing. The object-segmentation algorithm based on MRF model using wavelet transform can reduce the mount of computing in object-segmentation, so that fit the requirement of real-time and assure the accuracy and coherency. Although this method makes some efforts in aspect of universal, we still need to do more work in deeper research.