Nowadays, applications in image processing require high-resolution images for better visual quality and more accurate analysis. The goal of image enlargement is using information from a single image, from multiple images, or from large image databases to generate a high-resolution image. Image enlargement is an ill-posed problem due to many factors: insufficient information from low-resolution image, unknown downsampling kernel, and etc. Most of the previous works have focused on low magnification factors. In this project, we present an algorithm that can hallucinate high-frequency details into the enlarged image. Given a small input image, the enlarged result can be generated by utilizing the information extracted from a large image database. Moreover, unlike some of the other state-of-the-art learning based methods, the proposed method does not need any training process in advance. The proposed method also shows considerable improvement in visual quality over standard image enlargement approaches at large magnification factor.