A novel image descriptor based on dot-diffused block truncation coding with bit probability

With the advent of mobile smart devices and ubiquitous network connections, digital images can now be conveniently captured, edited, and shared online worldwide. The ever-increasing number of pictures poses a technically challenging, yet unavoidable problem of efficiently and accurately finding a desired image. In this work, we develop a new image retrieval system based on an image compression technique, namely dot-diffused block truncation coding (DDBTC) with bit probability. Specifically, the color feature derived from color distribution and the bitmap feature (BF) derived from both edges and textures jointly describe the image. The degree of similarity between two images is then measured by their respective color and BFs by using the modified Canberra distance metric. Experimental results show that the proposed feature descriptor achieves superior image retrieval performance as compared to the former DDBTC image retrieval feature and conventional non-DDBTC based features. Additional experiments on image classification verify that the proposed feature descriptor outperforms conventional image classification methods.