Image segmentation in weld defect detection based on modified background subtraction

In computer vision, the background subtraction is an important method to detect moving objects. The background reconstruction algorithm is based on the hypotheses that the background pixels intensity appears in image sequence with maximum probability. This paper proposes a real-time weld defect detection algorithm using a modified background subtraction method based on the assumption that the background pixel intensity appears in image sequence with maximum probability and the distribution of the pixels of background conforms to the Gaussian distribution. The algorithm has been successfully applied to the on-line weld defect detection. Our approach can perfectly extract and roughly classify the weld defects. Experimental results show that the proposed algorithm can meet the requirement of the efficiency of on-line continuous detection of weld defects and detect weld defects automatically and successfully.