Abstract: The goal of texture probing is to find the locations of the observed textures in an image by probing points. For an observed texture image, the topographic independent component analysis (TICA) is employed first to learn the un-mixing bases. The bases, acting as filters, are selected further by the criterion of the maximum response. All the selected filters from the observed texture images are collected together to construct a filter bank. Then, the mosaic texture image to be tested is decomposed into filter channels to calculate the texture features of the probing points. To group these probing points, a spectral graph is constructed by treating them as its vertices. In light of the cut on the spectral graph by a spectral cluster (SC), probing point sets are finally separated recursively. Experiments on mosaic texture images show that the proposed method gives satisfactory results.