Self-adaptive image processing using blind image quality assessment technique

Abstract

Image processing technique for filtering noise is a major challenge for DSP engineers. When the images are corrupted by noise whose characteristics cannot be evaluated a priori, processing systems need to be flexible and adaptable. General-purpose filters based on assumptions about image noise models fail to meet the quality and performance criteria in dealing with unmodelled noise. At the same time, evolutionary algorithms based adaptable filter architecture is proved to be successful in this regard. While existing evolutionary techniques based designs use uncorrupted reference image and compute mean absolute error for evolving a noise filter, the paper proposes a novel noise quality index based technique. The proposed entropy based scheme estimates noise content without any reference image and such a system is vital in situations where uncorrupted image reference is unavailable. Based on experimental results, the paper compares no-reference image noise assessment techniques with reference based technique and concludes that the proposed blind noise assessment method is accurate as referenced based techniques. From implementation point of view, the no-reference scheme is computationally intensive.