Quadtree image compression using sub-band DCT features and Kohonen neural networks

Image compression is an essential task for image storage and transmission. This paper presents a compression scheme for digital still images using Kohonenpsilas self-organizing map (SOM) algorithm with sub-band discrete cosine transform (DCT) features as inputs. Quadtree decomposition was applied first as preprocessing. It is an efficient way to segment images. The method of DCT is then used to identify the image frequency information. The sub-band scheme is utilized to separate the DC and AC coefficients in order to reduce the learning complexity of Kohonen networks. SOM is used in this paper to generate codebook for vector quantization (VQ). It has the advantages of preserving the topological property that generate ordered codebook with substantial dimension reduction. The consequence makes image compression even more effective. Simulation results show that with our scheme, high compression ratio is obtained while good reconstruction quality is also maintained.