Multiscale texture classification and retrieval based on magnitude and phase features of complex wavelet subbands

This paper proposes a multiscale texture classifier which uses features extracted from both magnitude and phase responses of subbands at different resolutions of the dual-tree complex wavelet transform decomposition of a texture image. The mean and entropy in the transform domain are used to form a feature vector. The proposed method can achieve a high texture classification rate even for small number of samples used in training stage. This makes it suitable for applications where the number of texture samples used in training is very limited. The superior performance and robustness of the proposed classifier is shown for classifying and retrieving texture images from image databases.

Graphical abstractFigure optionsDownload full-size imageDownload as PowerPoint slideHighlights► A computationally inexpensive yet efficient texture classifier is proposed. ► A multiscale feature vector is generated from complex wavelet subbands. ► Magnitude and phase information are utilised to achieve superior performance. ► The phase information contributes in identification of texture structures. ► Proposed multiscale classifier is also efficient in texture retrieval.