In the detection of small defects in a large background, since the existing machine vision systems sample images at a fixed resolution and a preset speed in one inspection, usually there is a large amount of redundancy. To solve this problem, a new machine vision system with parallel and selective imaging function based on visual attention was presented, which divided the image processing into two stages of paralleling pre-attention stage and serial attention stage. Following the mechanism of human visual attention, to remove most of the redundancy image data, an algorithm for identification of suspicious images in the pre-attention based on visual entropy was developed. The experiment results of detecting contaminants in wool indicated that the new machine vision system can efficiently reduce the total image data and the algorithm based on visual entropy is fast and valid. It provided a new approach for improvement of the speed and accuracy of real-time detections of the machine vision systems significantly.
