The color feature can effectively characterize the color distribution of images. However, most of the existing algorithms are based on a single color space, and the color moment is the most commonly used as global color feature vector, thus the lack of color spatial information lead to retrieval errors. Aiming at the problems, a block color feature extraction algorithm based on the mixed color space is proposed. The extracted color feature and texture feature are combined for classification and recognition of image. Experimental results show that the precision and recall of the color feature extraction algorithm in the mixed color space are improved in the process of classification and identification for Chinese painting and general image, when it compared with the single color space feature extraction algorithm. And they are further enhanced after image segmentation.