Research On Non-uniform Illumination Image Enhancement Based On Retinex Theory

Image enhancement refers to highlighting useful information in an image to remove or weaken useless information according to a specific need.Due to the influence of ambient light or the reflection of the surface of the object,the captured image may have uneven illumination,such as over-dark areas and high-exposure areas.This uneven illumination changes the original appearance of the image to a certain extent,which not only brings poor visual effects,but also is not conducive to subsequent image processing.Image enhancement can make the part of interest more prominent,making the processed image more suitable for viewing or easy for other engineering applications.With the development of several decades,many image enhancement algorithms have been proposed,such as histogram equalization based algorithms,unsharp masking algorithms,wavelet transform algorithms,Retinex based algorithms,and so on.Among various image enhancement algorithms,Retinex-based algorithms can effectively enhance the details and are widely used.Based on Retinex theory,an image is the product of the illuminance component and the reflectance,where both the illumination and the reflectance are hypothetical unknowns.In recent years,the focus of Retinex's image enhancement algorithm is to estimate the illumination based on various a priori assumptions,and then to separate the reflectance from the original image.Since illuminance estimation is an indeterminate problem,it is difficult to accurately estimate the illumination of a non-uniform illuminance image.Some algorithms attempt to directly use the reflectance as the output image by removing the illumination of the original image,but this method will make the image lose its naturalness and the visual effect is poor.In addition,the illumination cannot be completely removed due to the intense illumination changes in the entire image.Therefore,in order to reasonably estimate the illumination of the non-uniform illumination image and thus effectively enhance the non-uniform illumination image,the main work and innovations of this paper are as follows:1.The development of image enhancement methods based on Retinex theory in recent years is introduced.The key problem of Retinex image enhancement algorithm,illuminationestimation,is summarized and analyzed.Through analysis and comparison,this paper summarizes the three main constraints that need to be considered for illumination estimation:spatial smoothness,overall edge structure and size range.At the same time,this paper analyzes and compares several traditional Retinex image enhancement algorithms,and summarizes the importance of illumination components for image retention and good visual effects.2.A joint filter illumination estimation algorithm based on step function control is proposed.On the one hand,unlike other rough estimation illumination methods,this paper uses the local maximum channel method to calculate the rough estimated illuminance,which not only conforms to the physical significance of spatial smoothing of illumination,but also to the nature of non-uniform illumination images.On the other hand,due to the filtering method,the algorithm is able to maintain the overall edge of the illumination and avoid excessive smoothing.In addition,by controlling the magnitude range of the illumination by the step function,the reflectance can be limited to [0,1].The method of this paper fully considers the constraints of illumination estimation,subjectively and objectively evaluates the separated reflectance through experiments,and compares and analyzes the effect of estimated illumination on image enhancement.3.A histogram specification method combining logarithmic transformation is proposed to correct the illumination image.In order to enhance the image detail while maintaining a good visual effect,the estimated non-uniform illuminance is multiplied by the reflectance after being corrected to obtain a final image.The experiment compares commonly used correction methods,such as the Gamma transformation.At the same time,the experiment also compared several commonly used Retinex image enhancement algorithms,and carried out a comparative analysis of subjective and objective aspects of non-uniform image enhancement.