In this paper, a novel algorithm for image enlargement is presented. In order to improve the quality of enlarged image, a scheme, which detects the edge directions in local images using compressed domain information and classifies the local images by edge directions into nine categories, is introduced. To each category, a low-pass filter is designed to preserve local edge features when it is enlarged. For a specific local image, the low-pass image enlargement can be implemented by convolving the sparse local image with the corresponding filter. Experimental results show that the quality of images enhanced by the proposed algorithm are greatly improved over that by traditional low-pass enlargement schemes and the computation complexity of the presented algorithm is very simple.