Noise-refined image enhancement using multi-objective optimization

Image enhancement plays a fundamentally important role in nearly all of the vision and image processing systems. This paper presents a novel scheme for the enhancement of images using stochastic resonance (SR) noise. In this scheme, a suitable dose of noise is added to the lower quality images such that the performance of a suboptimal image enhancer is improved without altering its parameters. In this paper, image enhancement is modeled as a constrained multi-objective optimization (MOO) problem, with similarity and some desired image enhancement characteristic being the two objective functions. The principle of SR noise-refined image enhancement is analyzed, and an image enhancement system is developed. A genetic algorithm-based MOO technique is employed to find the optimum parameters of the SR noise distribution. In addition, a novel image quality evaluation metric based on human visual system (HVS) is developed as one of the objective functions to guide the MOO search procedure. For illustration, four types of SR noises are employed in this paper to improve different enhancers. Encouraging results are obtained when applied to a variety of image distortion situations.