Image scaling is widely used in many regions from consumer electronics to medical imaging. Because there is a difference in display resolution between a variety of display devices and multimedia contents, most multimedia devices might need image scaling processing unit. When scaling up source images, conventional methods and previous works make image blurred, especially in an edge area. Therefore, to preserve an edge offset in the source image, our basic approach is that source image is divided into 2x2 blocks, and each block is classified into an edge area or a flat area and interpolated by using different method. To classify two regions, we propose modified edge detection scheme. In the flat area case, bi-cubic interpolation function is applied normally. In an edge area, however, each source sample pixel is divided equally into 2x2 pixels. For keeping edge lines smooth in 2 dimensional spaces, specific 4x4 patterns are defined in advance. Each pattern is picked out according to edge direction and replaces 2x2 original source pixels in the edge areas. After exchanging the pixels, bi-cubic interpolation kernel which has twice frequency is applied into exchanged pixels. Objective results and subjective results indicate proposed method better than any other conventional methods and previous works in terms of sharpness