Cascading Transformations and Vector Quantization in Image Compression

摘要

From the perspective of reducing the bandwidth in digital image transmission or lowering the storage needs, image compression plays a key role. In this paper, a new approach for lossy image compression is proposed: the original grayscale image is decomposed by applying the discrete wavelet transform. The detail coefficients are then vector quantized by using SOM neural networks. The approximation coefficients are preprocessed before being compressed by using the JPEG method. The reconstruction of the image is achieved by performing the corresponding inverse transformations in reverse order.