In this paper, we present a compression technique which preserves edges in the compressed image and adapts itself to the local nature of the image. Inter block redundancies were seldom considered in the earlier design of compressors. We take this into acount, achieve a good compression ratio and also preserve good visual quality at the same time. In this paper, the range value presented by Nosiopoulos is used to classity the image blocks into four coding techniques, respectively, which are used to compress the associated class image block. The four coding algorithms are: mean coding, 2_level MMSE quantizer for blodk truncation coding (BTC), hybrid BTC (HBTC) and quad-tree coding. Experimental results show that considerable improvements over the work by Nasiopoulus can be achieved. As for complex images which contain sharp edges and even single-pixel-width grey level sudden changes, their visual quality is very well-preserved. Thus,our technique is suitable for practical applications if image quality is requested.