The Image Enhancement Algorithm Research Based On Retinex Model

For the weather, the shooting angle and the illumination condition, some problems willbe produced when the images are gained. Such as the dynamic range is compressed, contrastis low and so on. For the reason, it is significative that recovering the information from thedim images or enhancing the visual effect of low dynamic images. In1971, an colorconstancy theory called Retinex is proposed by the American physicist Edwin. Land, withwhich the problems above-mentioned is resolved effectively.The paper takes the Retinex model as the main research content. By analysing theexisting algorithms based on the Retinex, we point out their advantages and shortcomings.The methods include random path by Edwin. Land, iterative Retinex algorithm by Mccan,center surround Retinex and so on. Based on the foundation, this paper proposed an Retinexalgorithm combined with the non-subsampled contourlet transform and Bilateral filtering,which overcomes the two main problems in traditional methods that the noise is didn’trestrained when the image is enhanced and the illumination is not uniform in the overall area.Here, we give out the compared images and the objective evaluation indexs, the results showthe proposed method is superior to the traditional methods in index and visual effect. Themain contents of this paper are arranged as follow:Introduce the color constancy theory and analyse the advantages and shortcomings ofexisting algorithms based-on Retinex.Facing to the problem that the noise is didn’t restrained, propose an noise removedalgorithm by NSCT with its multi-directions and translation invariance characters.Based on the Bilateral filtering theory, propose the piecewise smooth illuminationestimation with local Bilateral filtering to resolve the illumination un-uniform...