Biomedical Image Classification Using Joint Histogram and Bhattacharyya Kernel

This paper presents a novel classification model for biomedical images that involves automatic classification of microscopy and endoscopy images by using joint histogram of intensity and distance. Our proposed joint histogram can adequately capture intensity, texture and shape information (distribution) about the object of interest. We derive a kernel based on the Bhattacharya coefficient of joint histogram. Experimental results on blood cell and tumor classifications show the effectiveness of the proposed method. For blood cell classification we obtain improved performance using our feature with Bhattacharyya kernel. For tumor classification our method obtained best classification accuracy among the previously reported accuracies on the same endoscopy dataset.