Research On Color Image Enhancement And Segmentation Algorithm Based On Human Perception

Image enhancement and image segmentation are the important procedures of image processing and image understanding. In general, the approaches that utilize color information will get more accurate results than those methods only uses gray informatin. By using image enhancement methods, the details of input image will be more strenthened and the color will be more vividly. Image enhancement techniques for color image have important application value in modern communication and multimedia etc. Color image segmentation methods have been widely used in video coding, human face detection and recognition, and content based image retrieval. The human perceptive characteristics on color and image apperance are very important to image enhancement and image segmentation. The main research of this paper is trying to get the more similar results by using the color technology theory of color science, human visual characteristics and human perception in color image enhancement and image segmentation.The main contribution of this paper is as follows:1. This paper analyzed the existing contrast enhancement methods that based on color appearance model. CIECAM02 can not predict the image apperance and iCAM has the immoderate color adatation. This paper presented a new color image enhancement algorithm by integrating color appearance model and image appearance model, to predict the image perceptual attributes that more similar with human eyes. And the proposed method can get the more accurate reference white estimating.2. This paper provides three pixel similartity measure functions which accords with human visual characteristics. Noise, perceptual detail and texture can be effectively distinguished by using those measure functions. The new functions can avoid the bad response to weak edges and detail preserving of traditional bilateral filter. Then a novel Perceptual Model based Multichannel Bilateral Filtering algorithm(PM-MBF) for color image is proposed based on pixel similartity measure function. First, the algorithm transform the input RGB image to another color space that is more fit for human visual characteristic, further, the Gaussian spacial filters for lightness and chroma channel are decided automatically by perceptual model; at last PM-MBF is realized by the defined pixel similartity measure function, The algorithm can preserve more weak edges, textures and details, specially it can keep the real color transition region, while suppress the noise. In additional, it avoided the color shift that occurs in other color spaces, such as RGB.3. This paper proposes a novel edge detection method based on human visual characteristic for color images, which decreased the false edge detection rate. The proposed method first simulates the eye blurring mechanism by Bilateral Filter to remove the weak detail and noises, this step can reduce the false edge detection rate and avoid edge location change; the color image enhancement process based on color appearance gives prominence to the details of the region of interest, this can reduce the true edges'missing rate; a new color gradient enhancement method is implemented on luminance and chroma channel by using various color difference equations, this process can reduce the false color details'detection rate. Finally, to get more accurate edge location, non maximum suppression is implemented and the edge is extracted by dual threshhold strategy.4. This paper analyzed the color image segmentation methods that utilized color and spacial information. For the shortage of region growing in seed selecting and the lower speed, we proposed a new fast color image segmentation approach based on connectiviness region labeling procedure. The method contains two steps. The first step is original segmentaion, which combined the pixels, that adjacent in spacial space and close in color, to one region. Then, the second step merged the small regions to neighbouring big one according to the color information and spacial adjacent relation. The proposed method is simple and effective to segment the input image without color quantization and seed selection.5. This paper analyzed the problem of the existing segmentation methods: ignoring the human visual mechenism, image apperance and the effect of color space's selecting. This paper proposes a framework of a perceptually based color vision model for image segmentation. The framework simulates the eye blurring mechanism by filtering the origin image, which helps to improve the performance to spatially altered images; then uses CIE CAM02 to predict the color appearance under complex viewing conditions, finally color similarity based segmentation approach is implemented after a extensively discussion among various color difference equation. The results indicate that our method helps traditional image segmentation algorithms produce result more similar to human perception.