Abstract: Generic object detection is the key to object localization, detection and tracking in video analyzing. In this paper, we propose a algorithm of generic object detection based on image color names, by which we can obtain sub-regions containing objects in the form of rectangle. In the beginning, an input image is segmented by using color names and the similarity of different regions is computed. Next, we merge the regions based on the similarity between the adjacent ones. Last but not least, selective search strategy is adopted to realize generic object detection in the images. The Experiments show that the candidate boxes detected by this method can meet the accuracy requirements of generic object detection.