Region of interest (ROI) based scalable multiple description image coding using MDSQ-SR

In this paper a novel region of interest (ROI) based scalable multiple description image coding technique using multiple description scalar quantizer with successive refinement (MDSQ-SR) is proposed. Visual attention model using graph based visual saliency is used to find the region of interest in the image. ROI based scalable multiple descriptions are then generated by selecting different refinement layers of the side quantizer of MDSQ-SR for the interested and uninterested regions of the image. More number of refinement layers are selected for the interested part as compared to the uninterested part However, the different refinement layer selection is performed in such a way that the rate distortion constraints of the multiple description coding are satisfied i.e. the joint decoding quality is better than the side decoding quality at particular rate. The proposed scheme is evaluated under lossless and lossy channel conditions. In lossless channels conditions, PSNR of the decoded image using region of interest based scalable multiple description image coding is increased by 0.6 dB when compared with scalable multiple description image coding using MDSQ-SR. The average PSNR improvement of the proposed scheme is almost 0.7dB under lossy channel conditions.