Evaluating Loss of Texture Definition in Carpets Using Mathematical Morphology: Covariance

Abstract

Carpet textures contain periodic information that varies across constructions and is degraded by mechanical wear. We apply image covariance, a digital implementation of mathematical morphology, to binary carpet images for the purpose of measuring aspects of texture periodicity. Our test materials consist of four kinds of double ply wool carpets of differing textures divided into control, light, and heavy wear samples. Video images were digitized by a True Vision Vista frame grabber. Gray-level images were histogram equalized and converted to binary. Covariance data allow one to measure period frequency, amplitude, and overall mean. Results for our carpet samples show changes in amplitude and mean with wear, and are consistent with findings for a previous paper using grey level co-occurrence analysis. Covariance analysis requires relatively minimal computation for processing and preprocessing, but results may be affected by loss of gray level gradient information. If textural features of interest are preserved, this method is an efficient and easily implemented alternative to co-occurrence analysis. Attention is also given to the covariance analysis of computer generated carpet-like textures. We attempt to duplicate the covariance behavior of our carpet series by altering the placement of the component texture objects and simulate carpet wear by degrading regular textures with noise. We offer some thoughts on modeling carpet texture appearance loss with the aid of simulated texture images.