Research On Image Restoration Of Motion Blurred Images

In the areas of aerospace, military reconnaissance, road transportation monitoring,health care, the useful information in the image is very important. However, because ofthe relative motion between the camera and the scene during the exposure time, motionblurred image would be achieved in the acquisition process, which brings aboutdifficulty in image feature extraction, target recognition and image analysis. In order tosatisfy the needs of reality, a clear image can be obtained from the blurred imagethrough a restoration processing. Although there are many mature image restorationmethods, due to the great challenging problem of the image restoration technology,identification of the point spread function (PSF) is still at the primary research stage.In this paper we fully research on the blur parameters of motion blurred imagecaused by uniform rectilinear motion. The principles of motion blurred imagedegradation and restoration are analyzed, and the common image quality evaluationmethods are presented. Through analyzing the degraded model and restoration model,the PSF model is built. What’s more, the key to determine PSF is to obtain blur lengthand blur direction. The wavelet basis and decomposition level suitable for motion blurbased on detailed study of wavelet transform theory and experiments are proposed.In order to accurately identify PSF of motion blurred image, a robust method basedon two-dimensional discrete wavelet transform (2D-DWT) and cepstrum is proposed inthis paper. The key idea of the proposed method is to estimate the blur length and theblur direction by employing2D-DWT on motion blur, then detecting the decomposedimage with cepstrum analysis. Mathematical morphological method and canny edgedetection are employed to extract the parameters of PSF. Experimental results covering180noiseless images and96noise images artificially blurred images validate theusability and accuracy of the proposed method.In order to choose the robust classic image restoration method, through discussingthe existing restoration methods (ie inverse filter, wiener filter, constrained least squaresmethod and Richardson-Lucy method), the advantages and disadvantages are presented.Based on the PSF estimated by our method, the simulation experiments for four kinds ofclassic image restoration algorithms are performed under the case of noiseless and noisy.By using four kinds of image quality assessment functions (ie peak signal to noise ratio,mean squared error, gray mean srads and laplacian sum), we get the figures of recovery images and the results of image quality evaluations. Through estimating PSF forreal-word blurred images and restoring images, we come to the conclusion that theproposed method is robust and effective.