Accelerated Compression Algorithm Based on Factoring Repeated Content

In this paper, we introduce the accelerated image compression method to reduce memory and bandwidth cost by factoring repeated content within images. Since the compression procedure costs much time to search for the similar regions, we adopt the feature descriptor Gray Split Rotate (GSR) to accelerate the self-similarity computation. The self-similarity computation is partly transformed into the comparison of the distances between feature descriptors. And we find that the computations of GSR distances of each feature descriptors are independent with each other. By the use of GPU parallel computing power, we filter out a huge amount of unmatched candidates. In our experiments we improve the speed of the compression process by one order of magnitude and meanwhile still preserve the quality of the compressed image by reducing unnecessary candidates.