Research On Image Denoising Algorithm For Metal Surface Defects

The evolution of different materials marks the progress of human society.Today,with the development of science and technology,the demand and application scenarios of metal parts are increasing,especially in key fields such as national defense,aerospace and nuclear industry.Naturally,these important fields are very strict for the quality control of all kinds of metal parts.Therefore,the research team of defect detection of metal parts at home and abroad is growing.Especially in the application of machine vision,various automatic equipment are used to replace human eyes for metal parts defect detection,in which the preprocessing of collected images is very important to the defect detection results.On the basis of studying the research status of existing image denoising algorithms at home and abroad,aiming at the metal surface defects of metal parts in this project,this paper studies the image denoising method for metal surface defects by using improved wavelet analysis algorithm and improved nonlocal mean filtering algorithm.(1)For the elimination of salt and pepper noise in complex metal images,based on the original non local means(NLM),an improved non local means denoising algorithm is proposed.The novelty of the algorithm lies in: the eight direction sobel edge detection algorithm is added to the original NLM algorithm to extract more accurate edge details.At the same time,the edge part is included in the measurement range in the original weight calculation method,and the weight is determined in cooperation with euclidean distance.Finally,compared with NLM algorithm,the results show that the improved algorithm increases the peak signal-to-noise ratio by about 5 dB and the structural similarity by about0.01 compared with the original NLM algorithm.(2)For the elimination of gaussian noise in complex metal images,based on the existing partial wavelet threshold denoising algorithm,an optimized wavelet threshold function denoising method is proposed,and two key problems are analyzed in detail,namely,the selection of threshold function and threshold.Several other improved algorithms in the literature are introduced,and a series of comparative simulation and analysis are carried out based on this.The simulation results show that the improved wavelet threshold denoising algorithm not only has good denoising effect subjectively,on the data,the peak signal-to-noise ratio increased by about 7 dB,the structural similarity increased by about 0.1,and the mean square error decreased by 76%,reflecting the improvement of its denoising effect.(3)According to the characteristics of the two denoising methods studied in this paper,the combined denoising of improved wavelet threshold denoising and nonlocal mean filtering is carried out for the complex metal image with high-density pepper and salt noise.Experimental research shows that the effect of using this combined denoising method is better than that of using only a single method,and it has certain practicability and effectiveness.