A new method for arbitrary shape ROI coding based on ISA-DWT

In this paper, we propose a new method called arbitrary shape region of interest (ROI) coding based on the integer-to-integer shape adaptive discrete wavelet transform (ISA-DWT). Firstly, the whole image is encoded to measure whether the reconstructed ROI PSNR is higher than the target. If it is true, it means that the estimated ROI PSNR is higher than the target and there is no need to go on to encode the ROI and the background (BG) separately; otherwise the ROI is encoded until the estimated ROI PSNR is higher than the target and then the BG is encoded. ISA-DWT is adopted to transform the arbitrary shape visual object. The well-known set partitioning in hierarchical trees (SPIHT) is modified to generate the compression bit streams. The proposed method has several novelties: 1) it ensures the PSNR of the ROI at first to overcome the unbalance of the ROI and the BG; 2) it does not import redundant wavelet coefficients like the lifting method; 3) it is fit for arbitrary shape ROI coding; 4) it achieves low complexity by reaching the target ROI PSNR.