The TEXSCALE Methodology: Assessment of Multi-Scale and Multi-Orientation Classification of Textures

A methodology for texture analysis termed TEXSCALE is described that involves a hierarchical approach to the problem of texture recognition and image segmentation by texture. TEXSCALE involves the determination of texture class 'tuned' masks that determine whether a texture belongs in a particular texture class, coupled with the use of 'tuned' masks that differentiate between members of the same texture class. This approach is of special interest in that it combines in a single stage the usual two step classification scheme involving feature extraction followed by classification rule construction. In the work reported here, the TEXSCALE methodology has been applied to 15 texture classes, each class comprising samples of 24 variants of particular Brodatz textures. of 8 different orientations and 3 scales of rotated and scaled We have assessed the classification effectiveness of the TEXSCALE methodology, primarily through the use of a confusion matrix, applied to 1080 distinct samples.