Defect Detection on Printed Fabrics Via Gabor Filter and Regular Band

Abstract

Two methods are proposed in this paper to inspect printed fabrics. One method is to apply a genetic algorithm to select parameters of optimal Gabor filter. Optimal Gabor filter can reduce the noise information of printed fabrics, which can achieve defect detection of printed fabrics. The other is in utilizing distance matching function to determine the unit of printed fabrics. Extracting features on a moving unit of printed fabrics can realize defect segmentation of printed fabrics. Two approaches of defect detection have their own advantages. Detecting method with Gabor filter using genetic algorithm has perfect detection results of random printed fabrics, the other method based on statistical rule can receive better defect detection results of regular printed fabrics. Both methods can be realized in practice and detection time of proposed methods can occupy little in total detection time.

Visual-Based Defect Detection and Classification Approaches for Industrial Applications—A SURVEY

“…They achieved 96.2% success rate with a Gabor wavelet network to extract optimal texture features from a defect-free image and then 97.1% with an only real Gabor filter for defect detection. Recently, Kang et al [39] proposed two approaches: an optimized Gabor filter and a distance-matching-based method called regular band. They achieved a 71.4% detection rate with 0% false alarm with Gabor filtering and a 93.1% detection rate with 4.9% false positives on 85 sample images from TILDA database with regular band.…”

Fast and Parallel Summed Area Table for Fabric Defect Detection

“…The main advantage of using statistical features descriptors is the rotation and translation invariance that means a defect has the same feature description independent from orientation and position. 23 The spectral approach is used to extract points and edges using several spatial or frequency domain¯ltering techniques such as Fourier Transform (FT), 10 Wavelet Transform (WT), 29,49,56 Gabor Transform (GT) 3,8,18,20 and¯ltering. 34 In the model-based approach a random¯eld of an image can be represented as a stochastic modeling by a simple function of an array of random variables.…”