A method for signalling block-adaptive quantization in baseline sequential JPEG

The traditional JPEG compression standard, unlike JPEG2000, does not prescribe a means for spatially-variable quantization of transform coefficients. This prevents useful features such as region-of-interest (ROI) coding, where, for example, the subject in a photograph may be lightly compressed while the background is heavily compressed to hide irrelevant details. In this paper, we show how the quantization may be adapted in each block and, more importantly, how the adaptation may be signalled to the decoder in a memory-efficient manner. The method takes advantage of previously unused slots in the Huffman tables employed for entropy coding. The adaptation allows, for example, ROI coding of subject and background. We demonstrate how the encoder may be modified to produce an output that is standards-compliant, and how the decoder may be modified to correctly recover the adaptively quantized image. We show that the dequantization of an adaptively-quantized image with a standard, non-adaptive decoder, amounts to filtering in the DCT domain. Examples are shown to illustrate adaptive quantization, signalling, and decoding.