Stochastic texture image retrieval and similarity matching

Stochastic texture images representing various foil materials (like polymer sheets, nonwoven textiles and paper) provide important information about these materials, and are frequently utilized in industry. It is often difficult to objectively measure the similarity among those images, or to discriminate images of different types of materials. This work proposes a new multi-resolution method for texture image discrimination and similarity matching. The wavelet transform is used to represent the images in multiple resolutions, and to describe them in terms of their orientation and gray-level distributions. It is also proposed a multi-resolution similarity measure based on this representation. Finally, some experiments illustrate the performance of our method, and some conclusions are presented.