Disparity Compensated View Filtering Wavelet Based Multiview Image Code Using Lagrangian Optimization

This paper presents a disparity compensated view filtering wavelet based multiview image coding scheme. The proposed codec decorrelates the input views into their frequency subbands using a disparity compensated lifting based wavelet transform. The codec then applies a 2D wavelet transform on baseband-image transforming it into a number of subbands. The general form of the Lagrangian rate distortion optimization algorithm is then modified and used along with the steepest descent algorithm to assign bits among the different subbands in such a way that it minimizes the PSNR variance of the decoded images. Two sets of experimental results are generated using three sets of multiview test images. In the first set of experiments the effect of using a weighted lambda for high frequency images on the PSNR variance of the decoded images is investigated. In the second set of experiments, performance of the proposed codec in reducing the variance of the PSNR of the decoded images is investigated. Results indicate that the proposed codec gives lower PSNR variance among the decoded images compared to the basic form of the codec at all bitrates.