Research On Motion Deblurring Using Coded Exposure

Motion deblurring is a traditional and important issue in computer vision applications and consumer photography. A motion blur is usually caused by a relative motion between the camera and the object during the exposure, such as hand shaking, object motion, etc. Removing motion blur to obtain a clear image using motion deblurring methods is of great significance for surveillance, criminal investigation, and security.It is well-known that motion deblurring using a traditional camera is ill-posed, due to the fact that high-frequency information is lost in the captured blurred image and cannot be recovered. Coded exposure can preserve the high-frequency information and turn the image deconvolution into a well-posed problem by fluttering the camera's shutter open and closed during exposure time. This paper presents improved motion deblurring algorithms using coded exposure with the help of a single image and multiple images respectively.On one hand, as for motion deblurring using a single image, although the coded exposure makes the blur kernel (or point spread function, PSF) invertible, PSF estimation remains very important, In this paper, a symmetrical code is designed firstly to be used for fluttering the camera's shutter. Thus manual selection of motion direction is avoided when estimating PSF, which will ease the estimation of PSF. Then a coarse but relatively accurate PSF is estimated automatically from the captured motion blurred image based on the state-of-the-art PSF estimation algorithm efficiently. However, there remains improvement space for the accuracy of the preliminarily estimated PSF because of the existence of burrs and break points at some portions of the coarse PSF. Thanks to the knowledge of the code structure, these portions can be found and then remedied through the removal or interpolation of pixels. Finally, a much more accurate PSF is obtained and then used for non-blind deconvolution to get a clearer deblurred image. Experiments demonstrate that the proposed method outperforms state-of-the-art motion deblurring algorithms.On the other hand, this paper combines the coded exposure and motion deblurring using multiple images for the first time. Traditional multi-image motion deblurring methods aim to take advantage of the complementary information in each blurred image to restore a clear view of the same blurred object. And the deblurring quality is usually better than that of single-image motion deblurring methods. But these traditional multi-image motion deblurring methods usually presume that the direction of relative motion between the camera and the object differs greatly in each blurred image, due to the fact that similar motion directions can only provide redundant information rather than complementary information. Because of the inertia of object motion, however, the motion direction in each blurred image rarely differs greatly. For example, for consecutively captured multiple blurred images of a vehicle in motion, the motion direction of the vehicle is almost the same, thus multi-image motion deblurring degenerates into single-image motion deblurring. This paper proposes to design different codes for the capture of multiple images using coded exposure to make each image can also provide complementary information even when the motion directions are similar. Thus deblurred image of higher quality can be obtained. Experimental results on both synthetic and real images clearly demonstrate that the proposed method is able to overcome the problem of traditional multi-image motion deblurring methods when the motion directions are similar and outperforms the single-image motion deblurring methods.