Motion blur analysis based on image segmentation and blind deconvolution

The problem of blurring caused by object motion in a gray level image is analyzed, and an algorithm combining image segmentation and blind deconvolution based on statistical features of objects and background is introduced to estimate visual motion and restore the image. Regions consisting certain geometrical information of pixels are regarded as suspected moving objects and segmented on the base of directional derivative of the image. Simple connected regions are selected by the use of mathematical morphological algorithm and level set method. Convolution kernels of regions larger than a given threshold are inferred through ensemble learning, and blurred regions can be restored individually. Radon transform is adopted to determine motion patterns of objects. Experimental results show the effectiveness of the algorithm for visual motion estimation and deblurring in a gray level image.