Abstarct: One of the long-standing challenges in photography is motion blur, which is generated from relative motion between a camera and a scene during exposure. The final purpose of image deblurring is reconstruction of a sharp image from a blurred version. In this thesis, we deal with image deblurring, which is due to the motion of a rigid object during exposure. The first problem in image deblurring is to extract the place of moving object. The second problem is to estimate the blur kernel, and, then, to reconstruct a natural looking image. To extract the blur region, a blur metric must be introduced in order to estimate the amount of blurriness in the pixels of the image. The blur kernel estimation is an ill-posed problem, because, this purpose must be come off using only a single blur image. The image reconstruction is challenging because the high frequency image contents attenuated by blur. To extract the blur region, in this thesis, a blur metric was proposed baxsed on a discriminative feature of sharp images. Using the proposed blur metric, the blur map, i.e. a matrix that encodes the blurriness value of individual pixels, is estimated. Then, by employing a pixon-baxsed segmentation method on the blur map, the given image is segmented into blur/non-blur regions. The segmented image is used as a mask in the remaining stages of deblurring. In this thesis, for blur kernel estimation, a blind deconvolution process baxsed on MAP was employed. In this process, the blur kernel and the sharp image are iteratively and alternatively estimated. To do this, two MAP sub-problems, one for kernel estimation and another for latent image reconstruction, have been introduced. In kernel estimation sub-problem, the data fitting term was introduced baxsed on salient edges in the image. Using the salient edges rather than the pixels is due to the blur kernel is better estimated from salient edges rather than from a smooth region. The Laplacian distribution, representing sparsity feature, has been used as kernel prior. In image reconstruction sub-problem, a new prior on sharp image baxsed on both the first and second image gradients has been proposed. Using this prior causes to precisely sharp edge estimation, consequently, better kernel estimation. For final latent image reconstruction, a non-blind deconvolution baxsed on MAP has been used. To introduce MAP, likelihood (data fitting) and image prior (regularization) terms, which cause to better reconstruct image details, have been proposed. The proposed prior knowledge, which is baxsed on the proposed blur metric, favors sharp images over blurry ones. It means that the regularization term corresponding to the prior knowledge, which is embedded in a MAP problem, has the minimum value for the true sharp image. Several experiments have been performed to validate the proposed methods in various sections of the thesis. The results show the superiority of the proposed methods on the methods introduced in literature.