Research On Defect Detection Method Based On Machine Vision For Thin Elastic Pantu-hose

As people are the pursuit of fashion clothes, the panty-hose the market need is more and more, and the material, the style of panty-hose is more and more demanding. At present, the quality of the panty-hose detection is mainly by artificial completed, which has many shortcomings such as great worker's labor intensity and low detection efficiency. And what's more, the method is greatly influenced by inspection personnel's subjective factor, which means the inspection efficiency is low and the miss rate is high, so it is difficult to get accurate test results. So the automatic detection technology used for the process of pantyhose detection is the inevitable trend. In this paper, the theory of wavelet transform based on machine vision technology in the field of detecting defects for thin elastic panty-hose is introduced. The concrete content is as follows:Image preprocessing: in image processing commonly used pre-processing method, analyses the respective characteristics, and through the experiment, studying the three ways to the effect of filter, we design a image filter processing method which is suitable for thin elastic stockings. In this way, the normal texture of the panty-hose is filtered at the extreme and the defect information is highlighted.Wavelet analysis: selecting suitable wavelet base, the wavelet analysis is used to detect pantyhose images, then we can get the high-frequency detail sub-image. In horizontal direction on the details of the image and the details of the vertical sub image will appear the local maximum value of wavelet coefficient, shown as singular points of gray on gray. Because the wavelet transform is sparse, the maximum before than decomposition is more outstanding, more conducive to the defect detection.Defect discriminant: wavelet characteristics analysis of the high-frequency detail sub-image was done. Selecting the four characteristics of energy, variance and entropy as the eigenvalue of the image, and extract the four characteristics from the high-frequency detail sub-image. All the characteristics are normalized and used as the basis for recognizing.Defect segmentation: study the three methods of threshold segmentation, and selecting the maximum entropy threshold segmentation with better effect for this subject. On this basis, we use the methods of morphology operations to binary the image.Through the analysis of the results, the defects identification algorithm proposed in this thesis can exactly detect the defects of hole, end out and double weft on the thin elastic panty-hose.