Directional Dct - Based Stereo Residual Compression

A novel stereo image compression method based on disparitycompensated residual image structure analysis is presented in this paper. The residual image resulting from the subtraction of the target image by the predicted one is exploited in this scheme. Unlike natural images, for which 2D-orthogonal transforms such as 2D-DCT provides good energy compaction in the frequency domain, stereo residual images exhibit directional and sparse components making them difficult to compress with classical methods. We propose to exploit the directional properties of the residual image, to compress it more efficiently, by using 1D Directional Discrete Cosine Transform (1D-DDCT). This 1D transform is applied to non-overlapping blocks of the residual image, along four directions: i.e. horizontal, vertical, positive diagonal and negative diagonal. The experimental results show that the proposed method, on some test images, outperforms the considered state-of-the-art methods.