Abstarct: Abstract Blurring is one of the most common defects in digital images. This happens mainly due to the movement of objects in the scene, camera shake when capture the photo, and the out of focus camera. Reconstructing a clear (latent) image from a blur image is called deblurring, which is one of the current issues in the field of image processing. The main challenge of deblurring is due to little or no information about the latent image or the blurring process, as well as the lack of optimal reconstruction filters to reduce or completely remove the deblurring effect. In addition, the reconstruction can be confounded by artifacts. Artifacts are created in the reconstructed image for various reasons. This is the challenge that is investigated in this thesis. In this thesis, the posterior maximum likelihood algorithm using the heavy tail prior knowledge on the gradient of the latent image is used to remove the blur. To control the blur kernel and prevent the artifact on the latent image, the difference in the energy level of the frequency domain for the latent images with artifact compared to the latent images with non-artifact has been used as an index. A number of masks have been made to calculate the energy in specific ranges to account for the difference in energy levels. The calculated energy is fed to the support vector machine as a feature vector. The support vector machine divides intermediate latent images into two classes of intermediate latent image with artifact and with non-artifact. The heavy tailed algorithm continues to reconstruct the final blur kernel until the support vector machine algorithm finds the last non-artifact intermediate latent image. The results of the evaluation of the proposed method show that the support vector machine has been able to distinguish the non-artifact latent images from the artifact latent images with the test accuracy of 89% and by applying it to the heavy tailed algorithm, it has been able to reduce the artifact latent images by 26%.