FDICS: A VISION-BASED SYSTEM FOR IDENTIFICATION AND CLASSIFICATION OF FABRIC DEFECTS
In todays global market, the key to a manufacturing enterprises success lies in being competitive. To achieve this, the enterprise must make use of state-of-the-art techniques such as Computer-Integrated Manufacturing (CIM), Just-in-time (JIT) and Total Quality Management (TQM). Computer vision is a relatively new technology that combines computers and video cameras to acquire, analyze and interpret images in a way that parallels human vision. The objective of this research was to develop an automated defect inspection and classification system using the principles of machine vision, image processing and pattern recognition. The importance of quality for the textile/apparel industry and the need for vision-based inspection systems have been discussed. Research has shown that texture plays a vital role in the inspection of the surfaces of many materials such as fabrics, wood and metal. Moreover, texture-based classification systems have been successful in domains such as wood defects, semi-conductor inspections, etc. Various methodologies involved in image analysis and classification have also been discussed in detail.
The design, development and use of Fabric Defect Identification and Classification System (FDICS), a vision-based system for identification and classification of fabric defects have been discussed. FDICS is made up of an image acquisition module, a feature extraction module and a classification module. The image acquisition module obtains the digitized image of the fabric sample using a video camera and stores it as an image file. The feature extraction module extracts the tonal and texture features from the image. The classification module classifies an unknown fabric sample into on of five fabric classes based on the Mahalanobis classifier. FDICS has been shown to provide a higher percentage of correct classification than a similar system reported in literature for defects considered by both systems. The relative accuracy of using only either the tonal or texture features was studied. The latter set of features gave a higher percentage of correct classification compared to the former; however, the percentage was the highest when both sets of features were used together.