New texture features based on the complexity curve

Abstract A new set of texture features based on the complexity curve is proposed for texture analysis. The complexity curve is obtained by computing the number of black-to-white transitions observed in a binary image for all possible threshold values in a gray-level texture image. The new features characterize the first-order statistics of the complexity curve, and therefore give a second-order description of texture. The set of features is applied to texture classification, and to capture the directionality and the periodicity of texture. For texture classification, two experiments are performed. First, some Brodatz textures are used to evaluate the classification performance of the features. Some of the traditional texture features, extracted from the co-occurrence matrix, have also been used for comparison. Secondly, the features are applied to the classification of Transmission Electron Micrographs of premalignant and malignant cell nuclei in a controlled mouse liver carcinogenesis experiment. Directionality and periodicity are two high-level texture features that guide the process of perceptual grouping of texture images. Therefore, two additional experiments have been performed to evaluate the performance of the complexity curve features in estimating the texture directionality and periodicity. The experimental results demonstrate that the new texture features based on the complexity curve are very useful tools for texture analysis.