Low-quality Image Enhancement Based On Human Visual Neuron Models

In the image preprocessing stage,in order to improve the recognizability of the hu-man eye and provide convenience for subsequent detection,recognition,tracking and clas-sification,as well as optimize the image quality,image enhancement technology is often used.Its main content is to highlight the part of interest in the image and reduce or remove noise and other artifacts,thereby making it easier for the human eye and the computer to identify key information in the image.There are many image enhancement methods,but relatively speaking,the Retinex theory based on human visual neuron mechanism is ideal for processing low-quality images.The theory believes that the color of an object is deter-mined by the object's ability to reflect light instead of the effect of uniform light.Based on this theory,researchers have proposed a variety of algorithms based on the Retinex model.This paper focuses on the theory of Retinex based on the mechanism of visual neurons.In order to release the serious halo artifacts,overexposure and details loss which caused by inaccurate illumination estimation,two novel Retinex image enhancement al-gorithm based on human visual neurons system are proposed.The main research of this paper is as follows:（1）The mechanism of visual neurons was studied.The neuron-based Retinex theory,the none classical receptive field suppression mechanism,and the prior conditions of illu-minace and reflection images are studied.Provides visual mechanism theoretical support for the proposed algorithms.（2）The Retinex image enhancement method based on Non-CRF suppression mecha-nism is proposed.This algorithm combines the non-CRF suppression properties with the variational function,making the enhancement results more consistent with the visual prop-erties.The weight matrix and transformation function that may influence results of image enhancement are analyzed,which proves the significance of constructing a weighted ob-jective function.At the same time,the numerical solution of this method is analyzed and optimized,which improves the efficiency of the algorithm.（3）The improved Retinex image enhancement method based on L₀is proposed.The characteristics of the L₀norm are analyzed.It has better sparsity and invariance.It can constrain the variational function more strongly,so as to rebtain more structured informa-tion in illuminance image and reduce the probability of halo phenomenon.At the same time,a denoising algorithm is proposed.The denoising operation is performed on the enhanced image,which effectively suppresses the noise amplified due to the brightness enhancement in dark region.（4）The index to evaluate the image quality was studied.Objectives such as LOE,IEM,PCQI,NIQE,and PSNR were used to objectively evaluate the image's light order,contrast enhancement effect,and denoising effect.Combined with subjective evaluation,the enhancement effect of the image is analyzed.