Abstract :
Summary form only given. In the field of image contrast enhancement, the Retinex algorithm is the most frequently used one. MSRCR(Multi Scale Retinex with Color Restoration) and IMSR (Integrated Multi Scale Retinex) are the two most popular Retinex algorithms in the literature. In this talk, a new algorithm "FH-IMSR" is presented to solve the problems existing in the MSRCR and IMSR algorithms. These problems include distortion of grey out due to the individual treatment of RGB channel, color noise due to log function, and halo effect occurred by the use of the Gaussian filter. The original IMSR algorithm is based on a RGB color model. The separate treatment for each channel in RGB space brings a phenomenon to change the rate of RGB, which causes grey out problem in the restored image. To solve this problem, FH-IMSR proposed to transform RGB color model into HSV color model, and apply the Retinex treatment on the luminance channel V. To remove color noise occurred by the use of log function, the Hue differences in the original image and restored image are minimized. To reduce Halo-artifact, FH-IMSR proposed to use Fuzzy C-means Clustering algorithm to cluster image pixels into dark area, middle area, and bright area, and different strategies are proposed to deal with brightness restoration on the three types of areas. Experiments on a wide variety of images are conducted, and comparisons with MSRCR and IMSR algorithms are made. The results showed that the proposed FM-IMSR algorithm has better performance than MSRCT and IMSR algorithms.
