Recognition of morphological-variation oriented patterns in textile structures

Abstract

A pattern recognition process, allows the identification of the different issues of the threads in the image of a textile piece. For this purpose, a morphological-variation operation has been defined. This operation generates an image in which the bright pixels show the density and distribution of a periodical structure of related to threads in the textile piece, showing the existence of possible defects in the textile piece.

The final classification is done through an analysis of the space distribution characteristics and its decrease, by means of a frequency distribution analysis, by finding the similarity with the original pattern through classification systems using algorithms of artificial neural networks.

The morphological-variance method is an alternative procedure for recognition of patterns oriented in the same direction, which are defined in a vector or matrix of morphological characteristics. In this case, liftingin gray-level images, where the classification process is determined by the morphological similarity with kernel pattern.