Research On Color Images Coding Based On Litfing Wavelet

Intelligent video surveillance is an important research field of machine vision, in recentyears, in the field of surveillance video analysis and encoding, the region of interest encodingand the fast image transformation based on wavelet have become hot spots. In actualapplication process of video image information, people are often interested in some specificarea of the whole image, so we use different method to deal with the interested region andbackground region to improve the compression ratio. This paper further research investigatesthe moving target detection, the interested region coding and image transformation based onwavelet. We find the traditional motion target detection cannot adapt to the local mutation ofthe scene, and it needs a lot of computation and storage. In the current video surveillancesystems, surveillance cameras are generally get a fixed field of view, and the backgroundinformation is relatively stable. During the encoding process, the system only extracts theforeground motion information, and compresses the motion area. And for the samebackground information stored in each frames, we can save it only once within a certainperiod of time. If some changes come up in the background, the monitoring equipment canadaptively update the information. And if using traditional image video encoding, it is verydifficult to improve compression ratio. So this paper makes a series of studies from theinterested area generates to image compression.The main contribution of our work includes:(1)A new clustering background modeling based on human memory model is proposedin this paper. First analyze the traditional Gaussian background modeling algorithm, and listits shortcomings. To solve these problems, the algorithm sets up a cluster model for each pixel,and each cluster can be adaptively created, updated, and deleted according to backgroundchanges combined with human memory model. The algorithm makes accurate judgmentsaccording to human ultra-short-term memory, short-term memory and long-term memory,and the moving target detection results meet the judgment of the human sensory organs.(2)Anew encoding algorithm combines the lifting lazy wavelet and ROI coding isproposed. The moving target detection algorithm mentioned above can provide ROI (regionof interest) for each frame. First turn the image into YUV format, using4:2:0sampling, to getthe normalized quantization and normalize them to an amount of integers. The traditionalvisual sensor image segmentation technology is immature and the first generation waveletencoding has high computational complexity. In this paper, based on visual memory model,the video sequence is divided into background frame and ROI frame. And this algorithm canachieve integer arithmetic.(3)A new wavelet image coding algorithm base on embedded block coding withoptimized truncation (EBCOT) is proposed in this paper. Comparing several traditionalcoefficient encoding algorithms, and consider wavelet image coding for ROI and the energydistribution of the high frequency region in the wavelet decomposition image, we find theEBCOT is the best. Under the premise of having the mask of ROI, this algorithm givespriority to the region of interest to encode. This greatly improves the decoded picture’s visual effects.(4) For the proposed algorithm, several standard video sequences are used to take thecompression test and proof the advantage of the algorithm. Experimental results show thatthis method can reduce the complexity and the storage as well. It achieves a more effectiveimage reconstruction and gets a better reconstruction image. The peak signal to noise ratio(PSNR) and the visual effect is better than tradition method in the same compression ratio.