Provided are an apparatus and method for effectively compressing and restoring edge areas in an image. Therefore, by predicting values of pixels positioned at each directional edge, among pixels constructing a 2×2 block in a current image, using values of neighbor pixels of the pixels positioned at each directional edge, selecting an edge mode corresponding to a directional edge in which the difference between the predicted values and actual values of the pixels is a minimum, and outputting edge mode data indicating the edge mode instead of the predicted values corresponding to the edge mode, it is possible to effectively compress edge areas in which little similarity exists between pixel values, and accordingly improve a compression rate of edge areas.