Fabric defect detection using image processing techniques

The aim of this thesis is to design a defect detection system using image processing techniques. Inspection process is very important for textile industry. Defects decrease the profits of manufacturersand cause undesirable loses. Therefore, to reduce loses manufactures initially started to employ experts to detect the currently available defects on the fabrics. However, human experts have some drawbacks such as tiredness, boredom, and inattentiveness which cause to reduce the detection of faults. Because of that reason, textile industry started to develop new methods. Fortunately, the computer vision technologies with new developments in software and hardware have been applied to textile industry to increase the effectiveness of defect detection system.In our work, Difference of offset Gaussian (DOOG) filter has been used as the main part of fabric defect detection system. So far DOOG filter have not been used in textile defect detection systems. But, it was used for analysing the textures.Fabric patterns have symmetric and regular structures and whereas the defects disrupt the regularity. It is suitable condition to apply DOOG filters to images. DOOG filter is easily understandable and have simple structure. The system first transfers the filter and image in frequency domain then calculate the convolution of them. After this, filtered image convert back to time domain and the results can give information about the defects. In addition to DOOG filter, histogram analysis and thresholding is used to increase the detection of faults.In finding the faults and testing the system performance real fabric images are used. This provided a real test to the detection system. The results have shown that all of the defects have been correctly identified.