Structural Similarity Based Image Quality Assessment: Pooling Strategies And Applications To Image Compression And Digit Recognition

Abstract

This thesis studies a recently proposed perceptual image similarity measure: the structural similarity (SSIM) index. Although still in its early stage, the SSIM index has demonstrated superior performance in a large number of tests as compared to the currently most widely used image distortion/quality measures, the mean squared error (MSE) and the peak signal-to-noise-ratio (PSNR). This motivates us to further investigate the SSIM method and extend it to other image processing and pattern recognition applications. Specifically, three topics have been studied in this thesis: spatial pooling strategies for perceptual image quality assessment, structural similarity-guided perceptual image compression, and handwritten digit recognition using complex wavelet structural similarity index.