The Research Of Color Cast Correction Of Color Image

With the development of multimedia technology, the spread of broadband network and the applications of image acquisition devices, such as DV, digital cameras, especially network and mobile phone cameras, the way in which people get images and video become more and more. As the window through which people can understand the world, the influence of image on people's daily life is increasing. About 80% of the information from the objective world is obtained by the human visual system. When people observe the objective things, human visual system is most sensitive to color, then less sensitive to the shape and surface texture, and least sensitive to details. So it is important to get the information of the objective world from the color features of the visual image.Image acquisition devices are different from the human visual system which has the characteristic of color constancy. Human visual system can adjust to the change of the environment light source automatically, remove the influence of the light source and perceive the color of the objects without mistakes. But when collect images, image acquisition devices are easy to be affected by environment light sources, the reflection properties of objects and the photosensitive coefficient of acquisition devices. So there are some differences between the color of acquisition images and the color of real images, which is called the phenomenon of color cast. Especially, now the most cameras can only implement some simple functions, such as image collection and image transmission, but cannot ensure the quality of the image. When the color image is collected, the color values of image collected from the same scene in different illuminants are very different, which brings errors to the analysis of color image. So the color cast correction is important to image processing and computer vision.There are many reasons leading to color cast of color image, so this paper only considers the color cast which is caused by the changes of the environmental light source, and the main research is to correct the color cast caused by changes of the color temperature of the environment light source and underexposure. The former correction is divided into two steps: color cast detection and color cast correction. Firstly, we detect that whether the image has color cast and determine the type of the color cast, and then only correct real color cast image. In this step, the color cast correction is based on von Kries's model of color constancy. The latter correction is on the basis of the Retinex's model of color constancy for the underexposed color cast correction directly.Based on the classical algorithm, the following tasks have been finished in this paper:1. Analyzing the classic color cast detection algorithm, such as histogram statistics, gray balance, white balance and two-dimensional color histogram method, Because these algorithm have limitations, for example the detection results of gray balance is not very accurate and the two-dimensional color histogram algorithm needs to be conversed the color space. So this paper, based on these two color cast detection algorithm, comes up with a improved color cast detection algorithm, which runs in RGB color space, determines whether there is a color cast in the image and the type of the color cast by using the ratio of the chroma mean(r/g) and the ratio of the chroma mean square (R_RMS/G_RMS) after normalizing the three-channel of RGB and uses the information of color value of RGB channel to determine the NNO region of the image. The simulation results show that the improved algorithm can effectively detect whether the image has a color cast, and if the color cast image the type of the color cast is real color cast or nature color cast. At the same time, the improved algorithm can determine the NNO region without the color space conversion and determine NNO region more exactly than using Lab color information.2. Analyzing the classic color cast correction algorithms which are based on von Kries's model of color constancy, such as max-RGB, Grey_World, SoG and GSI algorithm. The improved color cast correction algorithm is based on SoG and GSI algorithm. Using the NNO region determined by the improved color cast detection algorithm as the initial conditions of the iteration algorithm, it corrects the image which has the real color cast. The simulation results show that the improved color cast correction algorithm can effectively correct the color cast caused by the color temperature of the environment light source. The improved algorithm is influenced by the number of the iteration. However, we can learn from the subjective and objective simulation results that the improved algorithm is effective. The angle error and chroma error (Euclidean distance of chroma) of its corrected image are lower than the classical color cast correction algorithm.3. Analyzing classical color cast correction algorithms which are based on Retinex's model of color constancy, such as Retinex algorithm which is based on global features and Retinex algorithm which is based on local features. In basis on the classic local Retinex algorithm, the improved algorithm divides the output image which corrected by SSR,MSR,MSRCR algorithm into two parts: Luminance part and chroma part. When keeping the chroma parts unchanged, we calculate the histogram of the Luminance part and do linear stretch to the histogram which truncated at both ends. The simulation results show that the improved color cast correction algorithm can effectively correct the color cast caused by underexposed. From the subjective and objective simulation results we can learn that the improved Retinex algorithm can improve the image's information entropy with the contrast of the image unchanged.