Relevance vector machine with adaptive wavelet kernels for efficient image coding

This paper presents a practical and effective image compression system based on wavelet decomposition and RVM regression for compressing still images. Support vector machine (SVM)-based approaches have been recently proposed for image compression and have raised important interest. In this paper, it is genuinely proposed to use an RVM-based approach for the compression of color images. Since RVMs performance depends to a large degree on the choice of a kernel and kernel parameters, RVM with adaptive wavelet kernels (Adaptive WRVM) is proposed to improve the compression performance of RVM with standard wavelet kernels (Standard WRVM) for image coding. Comparative study of adaptive wavelet kernels and Gaussian kernel is carried out and results showed that adaptive Mexican hat wavelet kernel achieves the best image quality at a given compression ratio. A performance comparison of proposed algorithm with Rki-1, SVM with wavelet kernels (WSVM) and JPEG2000 compression systems is done. It is found that proposed algorithm gives better image quality for a given compression rate in comparison to Rki-1, SVM with wavelet kernels (WSVM) and comparable to JPEG2000.