1D Moment Signatures for Random Colored Texture Characterization

In this article, we develop a new method of characterization of colored random textures. This method is based on the use of the chromaticity diagram combined with the 1D-geometric moments. In CIE XYZ color space, each pixel of an image is associated with a point within chromatic space, in which a color is characterized by its wavelength and its purity factor. Thus, we elaborate an attribute vector which includes color and gray-level characteristics. Color characteristics are computed by means of the moments of the purity factor histogram and of the wavelength histogram. The energy assigned to each pixel is taken into account by computing the moments of the gray-level histogram. In addition, the random nature of texture is taken into account by the variance of estimation error of a 2D-AR model. The relevance of this characterization has been evaluated by means of a classification process applied to 720 images of granite stones taken from the ''marbleandgranite.com'' database. We show that a attribute vector of dimension 7 makes it possible to reach a percentage of correct classification of 91 %.