A new method for undersized part surface defect detection is proposed. In order to meet workpieces with different sizes and different surface geometry characteristics
we use the rectangular LED array with variable incidence angle as the light source. The original facing image is captured with CCD and treated with a series of pre-processing procedures. Then optimized Hough transform is adopted to fit circle rapidly for end face detection. Radial offsets from pixel point to fitted standard circle are calculated to acquire defected pixel points and then realize the quantitative evaluation for the end face defect. Experimental results show that this method can realize on-line detection of undersized part surface quality promptly and accurately as to improve the efficiency of existing production lines.