Research On Removing Impulse And Gaussian Noise For Surface Defects Image Of Strip Steel

No doubt that steel plays a role same as food in the steel industry, and the cold steel strip is one of the important products in the steel industry. It is not only an important original material of the military, shipbuilding, aviation aerospace, machine manufacturing and chemical engineering, but also gets extensive usage in the automotive, home appliances necessary for people’s lives. However, the existing of surface defect affects the quality of the cold steel strip seriously. Its real-time monitoring is a significant way to improve its quality. Thus, it is very necessary to develop a system rapidly, which can detect strip surface defect accurately.The detection system is inevitably affected by noise in the process of imaging. CCD image sensor used to obtain images, the degree of illumination, temperature sensors, signal processing circuit are the main factors that generate a large amount of noise in obtaining images. The noise create blur in the picture and is probably recognized as defect, resulting in decline in the reliability of the detection system. So, it is vital to eliminate the noise that the picture contains in the pretreatment of the defect detection.Firstly, for the pulse noise, this paper researches the basic principles and implementing process of traditional median filter, also researches the identities of the traditional median filter and its deficiencies. A new method(TTNA, Two Threshold Norm Algorithm) for eliminating noise is carried out in this paper, based on the front research. This new method holds a especial character, which can differentiate the noise points and non-noise points wonderfully and then filters the noise points using the median filter. Through a large number of simulation experiments, this method has been proved simple and applicable, and is evidently better than other traditional filter methods in the field of noise de-noising and details reservation. In addition, this method has a wild applicability under the circumstance of little noise and also has some application value in engineering. Secondly, for Gaussian noise, this paper researches wavelet threshold de-noising methods and the methods based on bayesian estimation theory. The former is widely used because of its conciseness, but construction of threshold function is sightless to some degrees and the thresholds have their own defects, which reduce the noise reduction effects. The latter gets better performance but always has complicated prior probability statistical models that cause large calculated quantity. In view of the above questions, this article proposes a improved method based on a simplified model. The proposed method sees wavelet coefficients of original images as zero mean Gaussian distribution, and evaluates it from wavelet coefficients of noise images using MAP, finally filtering residual noise of reconstructed image with smoothing. Through a large number of experiments, this method is simple,and has been proved more efficient than methods based on wavelet threshold, can filtering Gaussian noise efficiently in the strip surface defect detecting system.Finally, the establishment of an objective evaluation of the quality of image denoising system proved the superiority of the two denoising methods.