Xin Zhan, Rong Zhang, Dong Yin, and Kai Chen
Compression, image coding, remote sensing, sparse representation
This paper presents a new compression scheme for remote sensing image, the texture of which is much richer than that of natural image. The last decade has seen a growing interest in the study of dictionary learning and sparse representation, which have been proven to perform well on image compression. Compared with DCT or DWT for compression, the advantage of sparse representation is that the coefficient matrix is sparser with less non-zero elements. However, the non-zero positions in the sparse coefficient matrix are almost random and thus not conducive to entropy coding. As the content of remote sensing image is usually very complex and the position indices cannot be quantized, the coding cost of these non-zero positions becomes the bottleneck of compression. In this paper, an entropy-constrained dictionary learning algorithm is introduced to make the coefficient matrix more structured and adapted for entropy coding. The experimental results reveal that the proposed method reduces the coding cost of the position indices significantly and achieves better results than the JPEG2000 standard.
Important Links:
Go Back
