Automatic gastroscopy video quality assessment

摘要: Automatic endoscope video analysis is an essential function for medical robot and computer-aided diagnosis system. However, the performance of these algorithms are often degraded by low quality images under uncontrolled environment, where some them difficult even human ourselves analysis, such as over-saturated reflection, too dark or obscure. In this paper, we formulate problem gastroscopy evaluation a supervised framework detect non-informative frames from sequence. order to achieve goal, HSV histograms, pyramid histograms orientation gradients uniform Local Binary Pattern extracted represent frames. And then Random Forests classifier used classify Experimental results in our new dataset with about 110000 demonstrate that accuracy method 95% false positive rate lower than 1.3%.