Multiscale Blob Features for Gray Scale, Rotation and Spatial Scale Invariant Texture Classification

This paper proposes to apply a series of flexible threshold planes to the textured image and then use the topological and geometrical attributes of the blobs in the obtained binary images to describe image texture. The proposed multiscale blob features (MBF) is invariant to linear gray-level scaling and rotation, and is insensitive to uniform spatial scaling. The experiment results show that MBF offers very low error rate on the entire Brodatz texture database, and confirm its invariance properties