Invariant supervised texture recognition using multi-channel Gabor filters
A method of rotation and scale invariant texture recognition is proposed, which can also be employed for multi-object analysis and object recognition based on pattern analysis in noisy images. The recognition method uses data base textures, which are compared with a texture to be recognized and relies on extraction and classification of features. The features are extracted using multichannel polar logarithmic Gabor filtering of the data base textures and the texture to be recognized with the same definite filter bank. The polar logarithmic orientation of the Gabor filters guarantees rotation and scale invariance. The classification of the features is carried out by symmetric phase-only matched filtering. The performance of the method has been tested on Brodatz textures, hone textures, and textile textures with perfect recognition results. Rotation angle and scale factor can be determined with arbitrary precision by the classification scheme.