DEMD-based Image Compression Scheme in a Compressive Sensing Framework

Efficient representation of the background texture in video image frames, motivates compression strategies based on good perceptual reconstruction quality, instead of just bit-accurate reconstruction. This is especially true for video image frames in applications such as videos with structural patterns, and Bi-Directional Reflectance Distribution Function (BRDF) image frames of an object, where different images of an object in a single pose are taken in different illumination conditions. This paper investigates a new approach for an efficient representation of a class of images from textured videos and different BRDF images of an object, using sparse representation of the Directional Empirical Mode Decomposition (DEMD) residue of the frame. The efficient representation of the DEMD residue is achieved as a sparse coding solution based on a Discrete Wavelet Transform (DWT)-based sparsification. Experimental results demonstrate the effectiveness of the algorithm showing higher compression as compared to standard wavelet-based image compression schemes in a Compressive Sensing (CS) framework and JPEG2000, at similar perceptual reconstruction quality.