A Research Of Color Image Enhaneement Technology Based On Retinex Theory

The main purpose of image enhancement is to highlight the useful information or inhibit the unwanted noises in the image doped in the process of transmission or acquisition, so that the image can be consistent with human visual system as much as possible. The result of image enhancement enormously increase it’s application value and follow-up processing power, which will be propitious to recognize and analyse by computer. Currently, image enhancement technology has been widely used in military reconnaissance, aeronautics and space, remote sensing, medical images, civilian shooting, underwater images etc. And many scholars pay close attention to it.The traditional image enhancement algorithms just have a single function, usually enhancing for the certain features of the images. when a color image is enhanced by the traditional algorithms, the histogram will be prone to distort, in other words, the color image appears color distortion. Retinex image enhancement algorithms are based on human visual system, comparing with the traditional algorithms that have lots of incomparable advantages, such as color constancy, image sharpening, high image fidelity etc. The color information of the image enhanced by Retinex image enhancement algorithms has been well restored and maintained. This paper has made the in depth theoretical and applied study based on these two aspects: Retinex algorithms theory and its development. Summarized as follows:(1) The Retinex theory and it’s mathematical models were researched, the developing process and research status of Retinex image enhancement algorithms were discussed.(2) Studied the Retinex algorithms based on iteration and center\surround, including: Frankle-Mc Cann, Mc Cann99 algorithms and SSR, MSR, MSRCR algorithms. One dimensional path comparison in iterative algorithms was modified to two dimensional regional comparison in center\surround algorithms. Compared with iterative algorithms,center\surround algorithms have greatly improved both in the processing efficiency or the enhancement result of the images.(3) Proposed a new Retinex algorithm based on single-channel processing in view of existing faultiness, including LB_MSR and MSRCP algorithms. We apply multi-scale Retinex on the intensity channel instead of each color channel of RGB color space. Then the results of MSR is stretched by a linear transformation. Finally, a color restoration step similar to MSRCR is applied to maintain the initial chromaticity. Experimental results show that, compared with traditional Retinex algorithms, the proposed algorithm not only can well protect the original image color, but also can effectively overcome the color distortion issue on the premise of ensuring color constancy.(4) In this paper, we tested and verified all the Retinex algorithms. And a lot of comparative work has been done on the results of various algorithms and the parameters of impact algorithms.