Analyzing image compression with efficient transforms & multistage vector quantization using radial basis function neural network

A fundamental goal of image compression is reduction in size of the image with maintaining acceptable image quality. It plays a vital role in storage and transfer of images over the web. Quality of image along with its size plays a vital role in various applications. Image compression in transform domain is one of the popular technique. This work presents use of 2-Dimensional Discrete Wavelet Transform (2D-DWT) along with a technique such as Multistage Vector Quantization (MSVQ) for compressing image. The codebook required in different stages of vector quantization (VQ) is generated using LBG algorithm. The neural network approach called Radial Basis Function (RBF) neural network has been employed to provide training to the indices generated from MSVQ stages. Finally arithmetic coding is used for further encoding. The method tries to achieve the advantages provided by these techniques. The method is also applied on Discrete Cosine Transform (DCT) and 2-Dimensional Discrete Cosine Transform (2D-DCT) for comparison. The experiment is performed on six images of size 128*128 each. The results proved the better quality of image in terms of PSNR and acceptable compression ratio compared to other mentioned transforms. First image is split into number of tiles then each tile is applied with DWT resulting in number of sub-bands according to one, two or three levels of decomposition.