One of the main quality problems of buttons is scratches. The use of machine vision to detect round button scratches can improve work efficiency and reduce costs. This paper has proposed a set of algorithms to improve the scratch detection. As the button has a plurality of ring regions, and the color of each ring region is not the same, Hough transform can be used to extract the center and radius of circular button, then polar transformation is used to convert the circle into a rectangle. It can be faster to divide the different annular regions and easier to partition a button image and calculate binary thresholds of partitions. Button surface is unsmooth and uneven dyeing, so it will interfere the detection of scratches. This paper presents an algorithm for suppression of ring texture, it can effectively reduce the button background irregular texture interference. As the same ring area may exist a variety of colors and the textures of different ring areas are not similar each other, in this paper an algorithm for finding binary thresholds for annular partitions is proposed to reduce interference of scratch judgment. In fact, many scratches are not continuous but intermittent. An intermittent scratch detection algorithm is also proposed to detect intermittent scratches. Combining all the algorithms the scratches in complex round buttons can be well detected.