Layer based multi-view image compression

We propose a compression algorithm for an array of multiview images. First, we apply a segmentation algorithm to partition the data into coherent layers and significantly reduce the number of images required for artifact-free rendering. Then, we exploit the coherence in each layer by applying a 1D disparity compensated wavelet transform across the views followed by a 2D SA-DWT on each of the spatial subbands. Finally, the data is entropy coded using a modified version of EBCOT. Experimental results show that our coder outperforms state-of-the-art H.264/AVC at low bitrates and intra-image JPEG-2000 over the complete range of bit-rates. Furthermore, unlike other multi-view image compression techniques, our implementation does not rely on estimating a 3D geometric model of the scene.