Improved detection of woven fabric defect by optimized and adoptive cylindrical band-reject filtering

Abstract

For assuring the quality of woven fabric, use of machine vision system in the weaving industry has become an essential demand. In this paper, we present a novel approach for detecting defects on woven homogeneous/inhomogeneous fabrics by generating 3D fabric image containing information of the entire range of inhomogeneity of the fabric class, followed by the designing of a fabric class specific optimized 3D cylindrical band-reject filter and defect specific energy threshold value. The method was validated on 10 types of fabric defects taken place on 582 defective and 79 non defective fabric images from TILDA database. Detection success rate of 97.5%, false positive rate of 1.8%, recall value of 96.77% and precision of 98.17% was achieved during detection of defects by this method. The method showed promising results during detection of defects from Salt & Pepper and Gaussian noise corrupted fabric images.

Keywords:

• Fabric defect detection
• 3D Fourier transform (3DFT)
• KL divergence
• particle swarm optimization (PSO)
• deep auto encoder (DAE)

Disclosure statement

No potential conflict of interest was reported by the author(s).