Texture feature extraction based on primitive analysis

The authors examine feature extraction from textures with weakly coupled primitives. Three simple techniques are presented for primitive extraction. Texture features are defined as means and standard deviations of the first-order distributions of primitive attributes such as area, perimeter, compactness, orientation, eccentricity, color, and contrast. The effectiveness of the extracted texture features is illustrated in the context of texture classification. Experimental results show that the extracted texture features are particularly effective for macro- and regular texture discrimination but not so effective for micro- and more random textures