Fabric Defect Detection Based On Non-gaussian Statistical Characteristics

In the fabric industrial, quality control is very important for fabric products. The commonly used detection method is manual detection whose detection effect is heavily dependent on worker's experience, attention and sense. Automated detection is becoming more and more attractive in modern fabric industrial. It is gradually replacing manual detection. Current automated detection is mainly based on the second-order statistical properties of fabric images. These technologies don't take into account the fact that fabric images are subject to non-Gaussian distribution both in image space and transformed space, hence detection rate is low. To solve this problem, a novel efficient method of texture defect detection based on the non-Gaussian statistical properties of texture images is presented, and applied to the fabric products. It improves the automatic detection technology, provides fast and accurate defect detection for fabric products. The main works are as follows:1. Based on the non-Gaussian statistical properties of fabric images both in image space and transformed space, combined it with Independent Component Analysis (ICA) and Wavelet Transform (WT), a novel fabric defect detection technology is presented and applied to the fabric images.2. Research and analyze the description ability of various higher-order statistics for non-Gaussian statistical property, combined it with the robustness of L-moment, a new fabric images description method based on L-moment is presented. It overcomes the sensitivity of external interference.3. To reduce the external interference of the dust and small wool on the fabric surface and uneven light and image noise in the fabric image collection process, a texture enhancement algorithm is used to enhance the difference between defect region and the background. It further improves the defect detection technology based on the non-Gaussian statistical properties.The experimental results have demonstrated that: the novel defect detection technology performed better than traditional methods, the average success rate is more than 95%.