Discrimination of fabric frictional sounds based on Haar features

Abstract

In allusion to the challenging issue of identifying fabric materials by frictional sounds, this study endeavors to prove the possibility of classifying fabric friction sounds into their material categories using discriminators built upon the Haar features. A total of 32 pieces of fabric falling into four material categories including cotton, wool, silk, and flax are put through a specialized apparatus to collect frictional sound signals. The Haar features on every scale and position of the acquired signal are extracted to establish a feature space. For each point in the feature space, a discriminator is built to approve all positive samples of a certain category and deny as many negative samples as possible. To relieve the heavy burden produced by the huge number of discriminators, progressive selection is performed on the discriminators to form a queue in which a discriminator is liable to fix some errors of the former. The outcome is a much-reduced version of the unordered discriminators with the same discriminability. The improved Haar feature is also investigated and is found to be capable of reducing the size of the discrimination queue, thus further improving the efficiency of the mechanism. It is also revealed that additional samples involved can help achieve a perfect accuracy. The discrimination mechanism advanced by this effort can provide a basis for identifying fabric materials by frictional sounds.