Image Preprocessing Research Of Silicon Steel Defect Detection Based On Machine Vision

As an important element in industrial applications, Silicon steel becomes very important that how to reduce its cost of quality testing, to improve efficiency of detection. But the image will inevitably be disturbed by a variety of noise during image acquisition, image transmission and other processing. It is difficult for the image to detect defect. So in order to make image processing, it is crucial to firstly filter noise of the image. There are a variety of image noise, mainly the impulsion noise and Gaussian noise. Edge-preserving filtering has a good effect for smoothing impulsion noise. In recent years, as a powerful tool for mathematical analysis, wavelet transform is widely used in the field of image de-noising. Many algorithms of image de-noising based on wavelet transform are made to filter Gaussian noise better.In this paper we deeply work over filtering of the most common impulse noise and Gaussian noise in the image. Research mainly includes the following three aspects:Firstly, image de-noising methods in the space domain. In the space domain, for the shortcomings of traditional and classical de-noising algorithms: mean filtering, median filtering and edge-preserving filtering, in this paper we propose a new adaptive edge-preserving filtering algorithm. The simulation results tell that the improved algorithm can effectively smooth impulsion noise. And the algorithm also achieves good results when it is applied to the image of silicon steel: compared with the classical spatial domain de-noising algorithm, MSE reduce by at least 0.5, and SNR rise by at least 1.5.Secondly, image de-noising methods in the wavelet transform domain. This paper analyses three wavelet de-noising methods: the modulus maximum de-noising method, the de-noising method of spatial correlation and the threshold shrinkage de-noising method. For shrinkage threshold de-noising method, this paper proposes an improved multi-threshold wavelet de-noising method, and a multi-scale and multi-threshold wavelet de-noising method. Through simulation experiments, it can concludes that the improved multi-threshold wavelet de-noising method is better than the traditional hard and soft threshold de-noising algorithm, multi-scale and multi-threshold wavelet de-noising method is better than MTS algorithm. Through the application of silicon steel image, it also achieves good results: when J = 3, the hard threshold MTS algorithm, MSE reduces by nearly 4.5, SNR increases by nearly 10.5; and the soft-threshold MTS algorithm, MSE reduces by nearly 2.0, SNR increases nearly 6.5.Last, based on filtering method of the space domain and de-noising method of wavelet transform domain, for the noisy image generally containing more than one noise, this paper gives a combination of image de-noising algorithm of the edge-preserving filtering and wavelet threshold de-noising. For the polluted image mixed Gaussian noise and impulse noise, the simulation results tell that de-noising effect of the combining algorithm is clearly much better than a single edge-preserving filtering and a single multi-scales and multi-threshold wavelet de-noising method. The algorithm also achieves good results for the application of silicon steel image: compared with a single edge-preserving filtering algorithm, MSE reduces by nearly 13.2, SNR increases by nearly 5.2; compared with a single multi-scale and multi-threshold wavelet de-noising method, MSE reduces by nearly 11.7, SNR increases nearly 5.6.