Matting-based restoration and enhancement for high performance panoramic imaging

It is known that images seen by human eyes and those captured by camera lens are evidently different. In general, a human eye can see the majority of the view field. However, a lens has limited visual range due to various specifications and uses. We can obtain a panoramic image that has a wider view field using stitching techniques from input images captured by putting the lens in different view angles. However, the panorama image may not be clear enough due to various reasons. In this study, we address a challenging problem where each input image is focused to a certain view point with a different distance. Since the lens has a finite depth-of-field, part of the image must contain some out-of-focus blur. To synthesize a clear panorama image from such kind of blurred input images, it is necessary to de-blur the input images first. In this paper, we assume that the spatially variant out-of-focus blur in an input image is approximately uniform, and study a matting-based method for restoring the image via de-convolving the input image.