Study On Rate Control And Region Of Interest In Image Coding

This thesis focuses on the rate control and region of interest coding technologies in image coding. We lucubrated the rate control of JPEG2000 and H.264 and proposed new methods for it. Arbitrary shape region of interest lossily and losslessly coding technology based on integer-to-integer shape adaptive discrete wavelet transform has also been studied.The major contribution of this dissertation is specifically stated as follows:1) At low bit rate, most computation in Tier-1 of JPEG2000 is redundant. A novel Slope Threshold Descending with Minimal Slope Discarding (STDMSD ) rate control method is proposed to alleviate this problem.2) The statistical result based on extensive experiments shows that ρ(ρ denotes the percentage of the zeros among the quantized transform coefficients)is still linear with the bit rate of a macroblock row in H.264. Based on the above conclusion, we have successfully developed a macroblock row layer rate control algorithm for H.264 via p -domain source modeling. The bit allocation of the proposed method is achieved according to the percentage of bit count of each macroblock slice in the first RDO. Considering the vast computation of the first RDO in the macroblock row layer rate control algorithm for H.264 via p -domain source modeling, we have presented a basic unit layer rate control algorithm for H.264 via p -domain source modeling, which obtains the statistical feature by the pre-analysis process with only using INTER₁6×16 mode and uses the linear rate model of the whole frame. Experimental results show that, compared to the rate control in JM8.4, the macroblock row layer and basic unit layer rate control algorithms for H.264 via p -domain source modeling can keep the same reconstructed image quality with less coded bit stream fluctuation and, at the same time, track the target bit rate better.3) A new arbitrary shape region of interest coding algorithm based on integer-to-integer shape adaptive discrete wavelet transform(ISA-DWT) is proposed. According to Parseval's theorem of energy preservation, the fidelity in the image domain can be estimated by that in its wavelet domain. The fidelity unbalance between the region of interest and background, in the case of guaranteeing thereconstructed quality, can be effectively decreased. The above algorithm is only used to code a single arbitrary shape region of interest. On the basis of the above algorithm, a lossy and lossless coding framework based on ISA-DWT for a multiple, arbitrary shape region of interst is presented. The lossy and lossless coding framework can not only code multiple, arbitrary shape regions of interest but also preferentially guarantee the reconstructed fidelity of regions of interest with high priorities. The above two algorithms are based on the presumption that the shape of region of interest is known, but the shape of arbitrary region of interest needs to be extracted at first in many applications. In this thesis, setting the MRI as an example, we use the morphological method to extract the shape of foreground(region of interest) and then introduce the ISA-DWT into coding the MRI. This method can not only guarantee the diagnostic information losslessly, but also achieve good coding performance.