A detection system based on hybrid of magnetic method and radiography imaging technique is adopted to detect the hidden causes of accidents and ensure the safe operation of steel-cord conveyor belt. It is necessary to implement auto defect extraction of X-ray image of steel-cord conveyor belt for the real-time performance of the whole detection system. This paper lays special stress on studying the auto defect extraction method taking connector stretch for instance. The malfunction images of steel-cord belts are denoised and trimed firstly to decrease useless data. Then the interference from steel-cord is reduced using horizontal gray differential method. Image binarization based on histogram as well as denoise algorithm based on area of continuous bright pixels are implemented to obtain the pure defect signal. Finally the position and size informations of defect are extracted. The experimental results evaluate the effectiveness of the algorithm.