Abstarct: Images are usually corrupted by noise and blur when recording. Blurriness is typically originated from common sources, including out-of-focus, atmospheric turbulence, camera shake, or object motion. One of the most widely used issues in the field of image processing is removing blur from a blurry image. Image deblurring aims to recover a clear image from an image that may have been blurred for any reason. If the blur kernel is known, this process is called non-blind deblurring; otherwise, it is called blind deblurring. In this dissertation, the problem of blind deblurring has been studied. Blind deblurring is much more challenging than non-blind deblurring, as only a single input blurry image is available, and we usually directly or indirectly seek to estimate the blur kernel from the blurred image. Finally, non-blind deconvolution methods are used to estimate the blur kernel to remove blurriness. In this dissertation, we use Multi-Scale Latent Structure (MSLS) prior to estimate the blur kernel and recover a sharp latent image. In the objective function we have proposed to restore the sharpness of the blurry image, we have added a term that encourages the blur kernel continuity and improves the sharpness of the reconstructed image and reduces the artifact. The evaluation results of the proposed method show that the quality of blur removal has improved significantly in many cases and for the databaxse used, the mean of PSNR and SSIM are 28.4163 and 0.8672, respectively. This is while the proposed method is much faster than the state-of-the-art methods.