Nowadays with advances in technology and the pervasiveness of using 2D and 3D images, we need measures that assess the visual quality of images. Based on amount of information from reference image, existing method are divided into three categories: no-reference, reduced-reference and full-reference. There are many full-reference and reduced-reference methods for 2D images, but in many cases there is no access to the original image. Most of the existing no-reference visual quality assessment methods are application specific universal no-reference schemes for 2D images have high computational complexity which are not suitable for real-time applications. In this thesis, a fast general-purpose method for quality evaluation of images is presented. Spatial and wavelet domain features are extracted from two-dimensional images that change with the type and severity of distortions. The most effective of these features are selected by a feature selection algorithm. In addition to maintaining the performance, reducing the number of features, also minimizes the complexity of the procedure. Afterward, the selected features are used as inputs to the neural network. The obtained values were tested on a standard database. Experimental results show that the proposed method is not only computationally simple, but is highly in accordance with the subjective scores. Stereo image quality assessment is known as one of the challenges in image processing. Compared with 2D image quality assessment methods, current stereo techniques have less performance. Most of them are full-reference and almost all of the no-reference ones, are application specific. The only no-reference general method documented in this context, has shown poor performance compared to the two-dimensional methods. In this thesis a no-reference general-purpose method is proposed that creates the binocular combined and anaglyph images from left and right stereo images. After spatial domain feature extraction, features of each synthesized image are given into the neural network separately. The third neural network, takes the outputs of these two networks in order to improve the results. This criterion was tested on a standard database and experimental results show that the proposed approach is consistent with the results of human evaluation. Keywords: 2D Image quality assessment, Stereo image Quality Assessment, Natural Scene Statistics, Feature subset selection, Nearal network, Synthesized image.