A hybrid video mining approach for Cancerous polyp detection in endoscopy videos

Video mining on endoscopic videos is a challenging task due to interesting features of unstructured data. This paper presents a new feature descriptor to automatically recognize the images with cancerous polyps in endoscopic videos. This approach uses the features of cancerous polyps, determining polyp as cancerous by one feature descriptor is not possible. To overcome this the work combines two different feature descriptors for much more accurate identification. The two feature descriptors are Complete Local Binary Pattern (CLBP) descriptor and the Global-Local Oriented Edge Magnitude Pattern (Global LOEMP) descriptor. CLBP is used to detect the texture information in the image and the Global LOEMP is used to extract the color features. By combining both feature descriptors a stronger technique can be employed to identify cancerous polyps.