An auto threshold adaptive algorithm is suggested for image segmentation of the eggplant images acquired in natural light. The segmentation is conducted with the gray-scale, three orthogonal feature vectors based on R, G, B's linear transformation, and the hue component of HSV color model. The experiment result of the segmentation with Matlab shows that the target and background have a good separation based on Otsu algorithm and the optimization of the gray-scale histogram. However, it is also observed that there is the impact of area light reflection on the region of eggplant surface. A new segmentation experiment based on improved Otsu algorithm is conducted with the color feature vector of (2G-R-B)/4 and the result shows that the new algorithm can partially eliminate the impact of light reflection. Finally, the improved Otsu algorithm with hue are tried in the segmentation and the result shows the impact of light reflection completely is effectively removed. The segmentation effect is further improved by removing the noise of hue-segmented binary image through mathematical morphology.