Rapid-transform based rotation invariant descriptor for texture classification under non-ideal conditions

• A rotation invariant descriptor is proposed for adapting non-ideal conditions.
• The proposed descriptor is based on Rapid-transform which is shift invariant.
• Feature selection approach is designed to improve the performance of the descriptor.
• More robust descriptor is proposed under noise condition.

Rapid-transform based descriptor is proposed for texture classification against rotation variations, illumination variations, as well as noise effect. The proposed descriptor is based on the local circular neighborhood and the local feature vector is obtained by means of Rapid-transform. The local feature vector is rotation invariant because of the cyclic shift invariance property of Rapid-transform. Combining several descriptors with different (N, R) parameters the spatial multiscale is obtained. Feature selection approach is designed to improve the classification accuracy and reduce the computing cost. The issue of noise is discussed and more robust descriptor based on Rapid-transform is introduced under noise condition. Texture classification experiments were carried out on the Brodatz and Outex databases, and promising results are obtained from those experiments.