Region-of-interest Video Coding Based On Wavelet Transform

Region-of-interest (ROI) image coding technique means to compress interestingregions in an image without loss or with little loss, and to compress uninteresting(background) regions with much loss. Based on this idea a high compression ratio canbe obtained and the important information can be preserved in the image. ROI codingtechnique provides a new way to solve the contradiction between the compressionratio and the image quality. Therefore, it is of great significance in theory andapplications.Based on the correlative research at home and abroad, this dissertation discussedsome key techniques related to ROI coding of sequence image, which include ROIbitplanes shift method, motion estimation algorithm, ROI tracking algorithm, and soon. All the work of this dissertation is focused on the realization of ROI video codingsystem by software for such applications as remote sensing image analysis, remotediagnosis and video communication.The major innovations achieved in the dissertation are as follows:1. A new ROI coding method called generalized partial bitplanes shift (GPBShift)is presented. To control the relative importance between ROI and background, themethod divides the bitplanes of ROI and background coefficients into two parts byusing scaling values S1 and S2, respectively. Instead of shifting the bitplanes all atonce by the same scaling value S in the standard methods, GPBShift shifts part ofthem on the basis of the bitplane shifting scheme. The Maxshift, generalizedbitplane-by-bitplane shift (GBbBShift) and partial significant bitplanes shift(PSBShift) methods are special cases of the GPBShift method, while GPBShiftprovides more flexibility for "degree-of-interest" adjustment of the ROI. TheGPBShift method not only is able to code arbitrarily shaped ROI without explicitlytransmitting any shape information to the decoder, but also flexibly select the scalingvalues to adjust relative compression quality in ROI and background. Additionally,the method may efficiently code multiple ROI with different priorities in an image.2. A predictive adaptive cross-quasi-diamond search algorithm (PACQDSA) ispresented. Based on the cross-center-biased motion vector distribution characteristicof the real-world sequences and high space-time correlation of adjacent blocks'motion vectors, a new cross-quasi-diamond search pattern is designed with sucheffective techniques as prediction of initial search point, half-stop criteria andadaptive search modes. Experiments show that the algorithm is able to fit all types ofvideo sequences adaptively in spite of the degree of the motions. And it is better thanthe traditional fast motion estimation algorithms such as NTSS, FSS and DS, in termsof both speed and PSNR, especially for the sequences with large motion.3. An ROI tracking algorithm based on Hausdorff distance is presented. Thealgorithm detects edge pixels in an ROI model and an image frame as the recognitionfeatures, then using them matches the ROI in subsequent frames by using theHausdorff distance rule. The ROI model is updated in real time using a new methodbased on the concept of the moving connected components. Finally, the ROIs areextracted from the binary model sequence.4. A new method for selection and evaluation of wavelet bases in imagecompression is proposed. By analyzing mathematical characteristics of wavelet base,several criteria are presented which include entropy, coding gain, product ofpeak-to-peak ratio (PPR) and coding gain, PSNR of the reconstructed image, energycollection characteristic and energy distribution characteristic. The experimentalresults show that the method is effective on selection and evaluation of wavelet bases.