Reseach Of The Color Image Enhancement Based On Retinex Algorithms

The purpose of image enhancement is to improve the visual effect of the image, for agiven image applications, to emphasize whole or local characteristics with the destination,expand the difference of the image between the different object characteristics, curb thefeatures of no interested to improve image quality and abundant information, strengthen theimage interpretation and recognition result, and improve the intelligibility of the image，according to the different needs to choose appropriate enhancement method. Imageenhancement can be distortion and the result of the enhancement is very important for thesubsequent image segmentation and identify. Along with the increasing popularity of theconsumer and professional digital camera, every day produce mass of image data. Butbecause the influence of the scene condition, many are in the high dynamic range scene, in thedark or foggy weather conditions, the visual effects of taken pictures are not good. This paperstudies Retinex algorithm in solving the above problems to made relatively good effect, intheory and practical application have the important meaning.The basic theory of Retinex is that people perceive color and brightness, is not justdepending on the absolute light that is the point into the eye, and the relevant of thesurrounding color and brightness. In the essence, the original image is divided into theincident image, the reflect by removing the incident image methods to avoid the unevenillumination, reappear the original characteristic of the image. The incident componentbelongs to the part of interference and lows. With this algorithm we can keep the tonereproduction, color constant and enhanced dynamic range compression etc.Thorough study the theory of Retinex，the paper was given the multi-scale Retinex imageenhancement algorithm based on brightness division. The algorithm can be divided into threesteps, first by introducing the HSI color space that is more suitable for the human eyeobservation object, this model can effectively separate brightness information from colorinformation. Only extract the brightness information of the image, to avoid the influence ofsaturation of the color image, greatly restore the color image information. And then import thelogarithmic image processing model (LIP), according to the different images luminancevalues will be given the brightness of the different areas. Finally, according to the different brightness areas to enhance with different scales of the parameters Retinex, and the area factorfusion different scales to enhance the results. The experimental results show that, comparedwith the existing multi-scale Retinex algorithm, the method can completely restore thebrightness information, remove the influence of uneven illumination. The fusion image cannot only keep the meaningful consequential details but also can keep the information of thestructure form the input channel, effectively improve the visual effect of enhanced image, canget better in detail and the total effect of the images.