Research Of Image Enhancement Algorithms Based On Retinex And Implementation

It is a difficult task to process shadow images, images under weak illumination and gloomy images. It is theoretically important for solving the previous difficulties while there are vast demands in the fields of application. For example, we often need to process images of aerial shoot , remote sensing images and clouded images in the video surveillance. And the Retinex theory that is the main object of this article could be applied to those images and have relatively better effects.The essence of Retinex theory is to estimate the illumination from the original image and get the real visage. Retinex algorithm could increase the dynamic range compression and keep better color constancy and tonal rendition. There are single scale Retinex (SSR), multi-scale Retinex (MSR) and multi-scale Retinex with color restoration (MSRCR) one after another. SSR can enhance the contrast of image, but it will cause color distortion. MSR has a better ability to increase the dynamic range compression than SSR, but it is not satisfactory enough for its ability to keep color constancy. MSRCR can solve the previous problems, but it has high computational complexity. So, it is not fit for real-time system.In the light of idea of multi-scale Retinex and on the basis of in-depth research on Retinex theory and theory of color constancy, four-scale Retinex algorithm was proposed combined with the original image. It is an important work to choose scale parameter in Retienx algorithm. Small scale could keep details of image, but it can cause serious distortion of color. Vast scale can achieve better color restoration, but it could cause details of image blurred. Four-scale Retinex adds a Gauss kernel with a medium scale. Medium scale can balance details of image and color restoration. The result proves that the algorithm of this paper not only expands dynamic range of image and keeps details, but also has better color constancy and tonal rendition. And the speed of this algorithm is faster than multi-scale Retinex with color restoration.While enhancing image, this paper has compared the objective index of image evaluation and analyzes the data. Observing the data and the histogram, we could find that influence on image enhancement is very little with tiny variation of scale parameter. And for some images, the difference of enhancement is also little between SSR and MSR. We also find that images enhanced by the algorithm of this paper have bigger entropy. This shows that those images keep good details and have good definition.In the light of the low contrast of remote sensing image, this paper has enhanced the remmote sensing image. The results shows that image contrast was increased and entropy was also larger after prosessing. It means that hue information is richer and details are kept better. Retinex algorithm may enlarge the noise. This paper adds a step for noise removal and combines the bilateral filtering with the algorithm of this paper. The result shows that the the contrast and entropy of enhanced image after noise removal are becomes larger than non-noise removal. This proves that it has a better result after noise removal.