Most of the existing lossy encoding schemes rely on the stochastic image model with a spatially uniform correlation and show decreasing performance as sharp edge features increase in images. To cope with images where sharp edge features convey more important information, a relatively new encoding scheme utilizing a Laplacian edge detector is proposed and experimented with prototype sample images. In this scheme, the Laplacian operator is applied to an image to get a corresponding edge field and, in turn, the edge field is encoded with the Huffman code. A quantization, a fundamental information reduction step, is also carried out in the edge field, and the corresponding error field, the overall intensity variations, is approximated with some discrete sine transform coefficients. A compression ratio up to 5 or 6 can be achieved with the current implementation without a severe degradation of image quality. An implementation with an image divided into tiles and an extension to an hierarchical model will be studied in the future.