Texture segmentation using directional operators

A method for the classification and segmentation of texture in images is presented. The method derives texture features from the image which are independent of mean gray-scale level but dependent on texture orientation. Each pixel in the original image can then be classified using supervised pattern-recognition techniques. A segmentation of the image can be produced by delineating the boundaries within the classified image in horizontal and vertical directions. The algorithm is efficient and can be implemented using general or specialist processors. Results from a sidescan sonar image of a sea bed and a video image containing samples of real texture are given