Motion Deblurring From Single Images Based On Wavelet Transform

The restoration of blurred images is an important branch of digital image processingtechnology, and motion blur is a common phenomenon which is caused by the relativedisplacement of the target and the imaging device. The restoration is conventionally dividedinto two aspects: firstly the researchers extract the degradation function, secondly they use theextracted degradation function to deblur the image.Except the motion blur, the influencing factors of actual motion-blurred images areuncertain and complicated, such as image noises. So, the researchers need to do image pre-processing at the start of image restoration, mainly the image denoising. Considering thecontradictory between eliminating noise and maintaining image edge details, this paperproposes an algorithm of motion deblurring from single images based on wavelet transform.The main work in this paper is concluded as follows:(1) By means of analyzing the degradation model of linear motion-blurred images，thispaper establishes the way that representing the point spread function (PSF) by blur scale andblur angle. After that, the common deblurring methods are analyzed through the experimentsand Wiener filtering algorithm is chosen as the general method. Then, this paper introducesseveral methods to evaluate the quality of restored images.(2) This paper proposes a method to estimate PSF based on fourier spectrum’s geometriccharacteristics, then does some error analysis to this method and verifies its anti-noiseperformance. Firstly, the method transforms the input image to fourier frequency domain.Secondly, combining with the edge detection algorithm, fourier spectrum’s geometriccharacteristics can be extracted through the hough transform. Finally, PSF will be estimated.(3) Based on the Mallat algorithm to2-D discrete wavelet transform, this paper clarifiesthe main idea of wavelet image processing: firstly the author decomposes image via wavelettransform, then processes the wavelet coefficients in transform domain, finally the authorfinishes the image reconstruction work through inverse wavelet transform. This paper alsodoes some research on wavelet threshold denoising, and illustrates the contradictory betweeneliminating noise and maintaining image edge details.(4) Combining with PSF estimation method presented previously, this paper proposes analgorithm of motion deblurring from single images based on wavelet transform. And theauthor verifies its advantages through contrast experiments with other methods. The mainidea of this algorithm: firstly the author does preliminary denoising to the input image anddecomposes it via wavelet transform, then does further denoising to the high frequencycoefficients and does motion deblur to the low frequency coefficient, finally the authorfinishes the image reconstruction work through inverse wavelet transform. (5) This paper does some expanding research to the targets which aren’t involved in thesimulation experiments, including the motion-blurred images who’s length and width aren’tequal, and motion pattern is non-linearity. In the end, the author does some experiments to therealistic images and acquired satisfied effects, including camera shake images and localmotion-blurred images.