Color Image Segmentation Based On Saliency

Make full use of the color information contained in a color image can help to improvethe segmentation accuracy. A saliency-based color image segmentation method was proposedin this paper. The concrete research contents are as follows.1. Color image segmentation based on RGB comprehensive information saliency. Firstly,the captured RGB color image was separated to R, G and B color channels, and the colorinformation for each channel was calculated. Secondly, the R saliency for each pixel in the Rchannel, the G saliency for each pixel in the G channel, and the B saliency for each pixel inthe B channel was calculated respectively. Then comprehensive saliency map was obtained bythe weighted R, G and B saliency. The weights for the R, G and B saliency were determinedby the corresponding color information of each color channel. At last, the targets wereseparated out from the comprehensive saliency map using a threshold method.2. Color image segmentation based on HSV comprehensive information saliency. Firstly,the color image was converted from RGB color space to HSV color space. The convertedcolor image in HSV color space was divided into H, S and V channels. The hue channel Hand saturation channel S were treated as the color information of the image, and the valuechannel V as the bright information of the image. Secondly, color information saliency wascalculated from H and S, and bright information saliency was obtained from V. Thirdly,comprehensive saliency map is obtained by weighted color information saliency and brightinformation saliency, where the weights were determined by the quantity of color informationand bright information. At last, the ultimate targets were separated out from thecomprehensive saliency map using the Otsu’s method.3. Results and discussion. In order to verify the validity and robustness of theabove-mentioned methods, the foreign fiber color images captured in the AVI system wereused as materials. The proposed methods were programmed in Matlab and were run in theWindows XP operating system. The results indicate that proposed methods can obtain highersegmentation accuracy and are more robust than the classical segmentation methods.