Scale and orientation matching for texture analysis and recognition.

ABSTRACTTexture is a necessary visual feature for image analysis. Many approaches have been proposed to model and analyze texture feature. Generally, researches try to use statistics results on visual features at different scales along various orientations in spatial or frequency domain as texture feature. These features are later compared without considering the scale of texture at the time of extracting information. This neglect probably leads to mismatching right texture matches. This paper contributes three matching schemes (SMS, OMS, and SOMS) to solve the problem. In these schemes, texture feature is not only extracted across scales along various orientations but compared at different scales and orientations as well. A database including texture images at different sizes, scales and orientations is generated to evaluate the performance of schemes. Besides, three modified versions of K-means algorithm are also applied on the database to automatically cluster the patterns. The purpose is to exhaustively examine the performance of proposed schemes under unpromising conditions. The experimental results show that the proposed schemes works well on generated database with various types of texture presented at difference sizes, scales and orientations.