Study On Image Denoising Based On Space And Frequency Domain And Its Application

The main purpose of image denoising is to effectively improve the quality of given images and the quality of the real images which are polluted by the noise.Denoising technology can do good to the quality of images,meanwhile,the image peak signal to noise ratio(PSNR)is increased to enhance the information of original images.As an important means of image processing,the studies of image denoising algorithm is unfailing.In the existing image denoising algorithms,some may obtain good results in the low dimensional signal image denoising processing,but they are not suitable for high dimensional signal image denoising processing;some may obtain good results,but they lost a part of edge information;some devote to detect the edge infoemation,but they lost the details of images.Finding a right way to achieve a balance between resisting the noise and keeping details is the key problem we solved in the thesis.The main contributions of this thesis are as follows:(1)The research status of image denoising is reviewed firstly,and the related basic theories of noise types and denoising method are introduced,and advantages and disadvantages of existing denoising algorithms are analyzed.(2)Considering shortages in spatial domain filtering method and frequency domain filtering method,we combine them effectively in the thesis to improve the results of image denoising.Firstly,the bilateral filter in spatial domain is selected to keep effectively the edge information of the image.Secondly,the wavelet transform with short-time Fourier transform is selected to keep the texture information of the images.Finally,the two methods above are combined to denoise.Experimental results show that the image denoising method combinated spatial domain with frequency domain can keep the edge and texture information in visual effects(ie,denoised images,edge detection images and the image histogram)and quantitative performance(for example,the Boat image whose noise density sigma is 0.2,the PSNR value is 37.88 dB in this thesis,which is higher than that of the other algorithms).(3)Based on the image denoising,the segmentation and recognition method of vehicle flow information are proposed.Firstly,the key frames are obtained by the image denoising method based on spatial domain and frequency domain;secondly,the vertical projection regional characteristics of key frames is combined with Hough transform to roughly locate the results of segmentation;thirdly,based on the the results of segmentation,SVM is taken to segment the license plate and recognise the plate numbers automatically.Experimental results show that the proposed segmentation and recognition method of the license plate image based on of mage denoising increase 2.01% than traditional methods.In this thesis,the image denoising method based on spatial domain and frequency domain improve the effect of image denoising,and we apply the method to segment and recoginise the vehicle flow information.Experimental results show that there are theoretical meaning and application value in the proposed methods.