A fractal-based keypoint computation method for solid textures

This paper describes a fractal-based keypoint computation method for solid textures. Solid textures have been used for a number of years for various applications including medical imaging, geographical data analysis and biological data analysis. It is important to accurately compute pattern features from solid textures, because pattern features can be used for classifications, detections, and retrievals. Recently, the use of local pattern features and the bag of features (BOF) approach has become popular for certain applications. In the BOF approach, software programs choose proper keypoints from the solid textures for computing local pattern features. Our fractal-based technique identifies keypoints from solid textures, and the techniques are examined using a solid texture benchmark dataset. Preliminary experiments indicated that the use of the fractal-based keypoints shows relatively better classification results compared to the use of random keypoints.