Abstarct: Superresolution is a technique that combines information of some low resolution images of a scene to make an image with higher resolution. In most of the previous related works, the blurriness that is applied to low resolution images is assumed to be due to the integral effect of the image sensor of the digital camera. But in practice, there are other sources of blurriness as well, such as, defocus, atmospheric and motion blur that may be applied to low resolution images. In this dissertation, we assume that the integral effect of the sensor is not the only factor of blurriness of the low resolution images and propose the following two approaches to consider the low resolution image blurriness. In the first approach, we use a neural network to determine the desired blur kernel. To this end, first, the blur kernel of the low resolution images are estimated. These blur kernels are fed to the neural network to determine the desired blur kernel that is applied to the high resolution image in the acquisition process. baxsed on this idea, we propose two distinct methods. In the first method, the neural network is used as a classifier that predicts the desired blur kernel. In the second method, the neural network is used in a regression frxamework that produces the desired blur kernel. The experimental results show using the determined blur kernel increases the quality of the reconstructed high resolution image. In the second approach, we propose a method for superresolution of defocus blurred images with unknown blurriness. In this method, the high resolution blurred image is estimated and then deblurred. Since the blurriness is unknown, the deblurring process is a blind image deconvolution. Deblurring is an inverse and ill-posed problem and solving such problems needs regularization. Hence, a proper regularizer and cost function is proposed. The experimental results show using the proposed cost function increases the quality of the reconstructed high resolution image.