Automatic recognition of woven fabric structures using digital image processing and artificial neural networks.

Abstract

The current method to recognize structures in textile mills in Uganda like Southern Range Nyanza limited involves the use of magnifying glasses and sometimes pulling of single yarns from the woven fabric. This method is not only of low efficiency and time consuming but also very tedious and subjectively affected by the knowledge and experiences as well as mental and physical condition of inspectors. It limits the number of structures that can be produced by a textile mill. It also causes strain to the eyes hence there is need for an automatic programme for recognizing woven fabric structures. This report consists of new methods established for automatic recognition of woven fabric structures based on digital processing and artificial neural networks. This method is a combination of image processing techniques (grayscale, histogram equalization and median filtering only), textual feature extraction using GLCM (correlation, contrast, entropy, energy, homogeneity) and the algorithm of single layer feed forward back-propagation of the stochastic gradient descent rule. In this case, the images were captured through a scanner with a resolution of 12001 x 2400 dpi. The program was provided with a user friendly Graphical User interface.