Abstract: Affected by many factors, outdoor scenes vary greatly, so using a single feature(color or texture) to complete the outdoor scenes recognition can not achieve satisfactory results. A natural object recognition algorithm based on SVM and color/texture combination features was presented. Firstly, the color histogram based on the RGI3 color space was extracted as color feature. Then, the texture feature was extracted based on multi-channel Gabor filters. At last, the color/texture combination features were presented. An image database of training samples including sky, road, house, tree and grass was created,which is obtained from Pasadena Houses2000 database of California Institute of Technology. And the natural object recognition based on SVM using a single feature and color/texture combination features was completed respectively. Experimental results show that this algorithm has good recognition effect on the images in which each natural object varies greatly from each other.

Key words: Natural object recognition, Gabor filter, Color histogram, Color/texture combination features, SVM