Adaptive median operators in image segmentation

The authors introduce an adaptive image segmentation algorithm for automated surface inspection vision systems. Also, a computationally efficient implementation architecture is proposed. The method is based on a multiresolution histogram algorithm which detects the defects within a variable size data window by monitoring the stationarity of the image. A good estimate of the stationarity is obtained by comparing the moving mean and median computed from the image. It is shown that using a large data window for the statistically stationary part of the image and decreasing the size of the window to the minimum during sharp edges results in high-quality defect segmentation. The proposed segmentation algorithm has been tested with images obtained from a cutting line of a cold rolling copper mill. Examples depict the performance of the proposed approach