The fusion feature wavelet pyramid based on FCIS and GLCM for texture classification

Local binary pattern (LBP) is an effective texture feature extraction algorithm, but sensitive to noise. This paper proposes an improved LBP algorithm named fusion center interval strategy (FCIS) for this problem. FCIS introduces the fuzzy method and presents an adaptive fusion center interval selection based on the local statistical information. To improve the completeness of feature extraction, a systematic texture feature extraction algorithm named the fusion feature wavelet pyramid (FFWP) has been proposed. FFWP is built on a set of flexible selectable filters of non-separable bidimensional Shannon type wavelets and fuses FCIS and gray-level co-occurrence matrix (GLCM). The FCIS and GLCM fusion features provide information on contrast, homogeneity, and local anisotropy. The wavelet pyramid with multi-level decomposition makes for multi-scale FCIS and GLCM fusion features with no need for the fixed radii. Validation experiments are performed on publicly available datasets, Kylberg, UMD, and UIUC datasets. Compared with 18 relative algorithms, the results show the effectiveness of FCIS with less time consumption and the supremacy of FFWP on robust and comprehensive texture representation.