Lesion detection of electronic gastroscope images based on multiscale texture feature

Electronic gastroscope has been playing an important role in the examination of gastrointestinal tract. However, due to its great dependence on the doctor´s experience and skills, the rate of misdiagnosis is still high. Therefore, an automatic lesion detection system is a huge help for doctors. In this paper, we design a new scheme for gastroscopic image lesion detection. Two new multiscale texture features are utilized and compared which combine contourlet transform with gray level co-occurrence matrix (GLCM) and local binary pattern (LBP) respectively. Combined with color feature and with AdaBoost as a classifier, experiments show that it is promising to utilize the proposed scheme to detect lesions in gastroscopic images. The best performance comes from the combination of color feature and contourlet based local binary pattern feature with false negative rate of 11.94%, false positive rate of 16.10%, and error rate of 13.99%.