Image compression via multiple learned geometric dictionaries

Abstract

In this paper, we present a novel image codec by leveraging sparse representation strategy for geometric pattern encoding. Specifically, we propose a Multiple Learned Geometric Dictionaries (MLGD) solution to explore various texture patterns of images, and use different dictionaries to encode homogenous smooth components and heterogeneous directional components. Profiting from model proficiency, our approach better preserves subtle details with high expressiveness and eliminates artifacts such as ringing and blurring. Experimental results indicate the proposed codec outperforms state-of-the-art in both numerical measures and visual fidelity, lifting the quality of experience (QoE) at low bit-rate.