Image compression by redundancy reduction

Abstract

Image compression is achieved by reducing redundancy between neighboring pixels but preserving features such as edges and contours of the original image. Deterministic and statistical models are usually employed to reduce redundancy. Compression methods that use statistics have heavily been influenced by neuroscience research. In this work, we propose an image compression system based on the efficient coding concept derived from neural information processing models. The system performance is compared with principal component analysis (PCA) and the discrete cosine transform (DCT) at several compression ratios (CR). Evaluation through both visual inspection and objective measurements showed that the proposed system is more robust to distortions such as ringing and block artifacts than PCA and DCT. © Springer-Verlag Berlin Heidelberg 2007.