Improved Bandelet with heuristic evolutionary optimization for image compression

As an efficient geometric analysis tool, Bandelet has exhibited enormous potential in image compression for its capability in capturing the geometrical structure in images. However, in available implementation of Bandelet, the optimal geometric flows are determined by a coarse and exhaustive search, which will degrade the performance of Bandelet in an accurate representation of images. In this paper, we advance a new image compression approach based on Laplacian Pyramid (LP) and improved Bandelet, where a Heuristic Memetic Algorithm (HMA) is proposed to locate accurate geometric flows. The chromosome is defined to represent geometric flows, and local and heuristic evolution operators are employed to make fast search possible. The improved Bandelet is used to compress the high-frequency band of the image decomposed by LP, and the low-frequency band is coded by Set Partitioning In Hierarchical Trees (SPIHT) algorithm. Some experiments are taken on some natural images and remote sensing images, and the results demonstrate that our proposed scheme outperforms SPIHT and the second generation Bandelet(2 G-Bandelet) in both PSNR and time consumption at low bit rate compression.