Abnormal region detection in gastroscopic images by combining classifiers on neighboring patches

Gastroscopy is widely used for the clinical examination of gastric diseases. The computerized methods capable to detect abnormal regions can help the physicians to identify the suspicious regions in gastroscopic images. The patch-based technique with the boosted stumps is adopted to detect all kinds of abnormalities in this paper. Considering that the responses of patch classifiers on the neighboring image patches are coherent, a flexible detection model is proposed which combines the patch classifiers´ outputs in the products of experts form to enhance the coherence of patch classifiers. The detection methods are evaluated on a large gastroscopic image dataset containing 2949 images of 413 patients. Experimental results show that the proposed method can improve the detection performance.