Analysis of Defect Classification Approaches for Fabric Images based on Four DFT Sector Features

Textile is an integral part of human society. The Textile industry drastically needs a solution to detect and classify defects present in any fabric. This will help the quality control mechanism of fabric manufacturing and processing units. Various types of flaws such as oil stains, punches, thread condensation, etc. may get introduced in the fabric; intentionally or unintentionally. These or many such examples of errors present in cloth made up of different kinds of materials such as Silk, Jute, etc. So it is need of the hour to understand the relationship between the fabric material and defect types. An effective mechanism can be developed to point out the specific features required for accurate classification of defects. In this paper, we have experimented with various classification algorithms to classify the defects based on proposed DFT features. These ranked results can be further analyzed for better categorization. We have used TILDA Fabric databases containing 3200 images of the following fabric materials: Silk, Jute, Diamond pattern, and Flower pattern with seven different types of defects such as Oil stains, punches, thread condensation, foreign body, wrinkles, camera distortion and lighting conditions for this experimentation. Results are analyzed based on subjective and objective analysis methods. It has been observed that Flower and Diamond pattern fabric materials have closer classification rates after applying SVM, Grid Search, and Random Forest algorithms. At the same time, oil stains and punches, these defects are found to be have an average classification rates similar to each other across all fabric types and classification methods.