Multi-Resolution Contextual Image Fusion and Compression Using PCA Features, SVM Classification, and Hybrid Entropy Encoding

Abstract

With the rapid growth of digital imaging applications, efficient image compression has become increasingly important for reducing storage requirements and enabling faster transmission, particularly in bandwidth-constrained environments. Healthcare, remote sensing, multimedia, and scientific analysis digital images must be highly compressed, with important visual and structural information retained. In this work, a new system of multi-resolution contextual image fusion and compression is introduced with the use of Principal Component Analysis (PCA), Support Vector Machine (SVM)-based adaptive patch classification and hybrid entropy encoding. The approach proposed uses a patch-based representation in overlapping patches to maintain continuity and consistency of context across space. The aim of PCA is to reduce the dimensions of data and preserve the main features of the image, which allows compact and informative representation of patches. To enhance adaptive compression, image patches are categorized into important and non-important ones by an SVM trained on features based on variance. Critical areas are more faithfully recreated, whilst less crucial areas are more severely compressed to achieve better storage performance. A further refinement of the reconstructed image is done with the help of Gaussian filtering and the Haar wavelet-based multi-resolution fusion to get a higher quality of preservation of edges, and also a better perceptual quality. The fused image is coded with a hybrid entropy coding scheme, two-channel bit-plane coding and Huffman encoding. It has been experimentally shown that the proposed method always performs better on the compressed size, compression ratio, saving percentage, compression gain, bits per pixel and coding efficiency compared to the traditional BMP, TIFF, LZW and baseline compression methods. The framework has compression ratios of 2.655, storage savings of 62.34 and reduced bit rates with high reconstruction quality. These findings prove the usefulness of the suggested framework to the efficient image storage and transmission applications demanding high-quality reconstruction as well as better compression performance.