Embedded image-domain adaptive compression of simple images

Popular transform coding techniques provide excellent lossy compression for natural images but fall short with "simple" images, i.e., images represented by only a small portion of the available intensity values. In this paper we propose a novel image-domain compression technique aiming at simple images, or, more generally, simple regions within a compound image. The proposed method is a context-adaptive bit-plane coder based on a binary arithmetic coder. The context information allows us to exploit the strong correlation across the bit-planes of simple images. Given the bit-plane coding structure our coder produces an SNR-scalable embedded bit stream which can be lossless when completely decoded.