Single-frame super-resolution image reconstruction is one of the issues highly considered and studied in recent research works. In the techniques of super-resolution, besides increasing the size of image,increasing the observable details is considered, too. In this research, a method of single-frame super resolution

is presented. It has low calculation complexity and also good accuracy. In this method, by making a completedatabase and training a Multi Layer Perception (MLP) neural network, we have tried to reconstruct high-resolution text images from low-resolution ones. In this research, it has been shown that the basic information of text images is saved in two most significant bit-planes, so the reconstruction of the high frequency components of this two bit-plane, has been focused in this research. Then by using low-resolution image upscaled by nearest neighbor interpolation, separating six least significant bit-planes of it and finally with the combination of this six bit-planes and those two reconstructed bit-planes, the images with high-resolution are obtained. At the end, the results of our method are compared with the common techniques. To evaluate the proposed method and compare it with some other methods, Structural SIMilarity (SSIM) is used. This quality metric measures structural similarity of two images, that in comparison with PSNR and MSE measures, presents a better concept of similarity between two images.