Three-Threshold Hysteresis Method for Human Shoulder Detection in Complex Images

J.C. Bailey and M. Zhang (USA)


Three-Threshold Method, Shoulder Detection, Complex Image, Canny Algorithm, Edge Detection *This research is supported by Christopher Newport University (CNU) 2004 Summer Stipend grant and CNU Applied Research Centre 2004 research funding.


This paper introduces a new technique, called three threshold hysteresis, for improving the detection of human shoulder edges with Canny's edge detection algorithm. Canny’s edge detector, while one of the better techniques in use today, is still dependent on proper adjustment of its input parameters to generate edge maps of images correctly. As a pre-cursor to using human shoulder edges to aid face detection by predicting face locations, this paper focuses on enhancing the localization of these edges. The three-threshold hysteresis technique modifies Canny’s two-threshold approach by adding a mid-level threshold and varies the criteria for edge pixel selection. The new technique is evaluated against Canny’s two-threshold hysteresis on twelve images containing shoulders and shows improved performance, with an 11.3% increase in success rate, using our examples

Important Links:

Go Back