Low-Complexity Lossless and Fine-Granularity Scalable Near-Lossless Compression of Color Images

We present a method that extends lossless compression with the feature of finegranularity scalable near-lossless compression while preserving the high compression efficiency and low complexity exhibited by dedicated lossless compression methods when compared to the scalable compression methods developed for lossy image compression. The method operates by splitting the image pixel values into their most significant bits (MSBs) and least significant bits (LSBs). The MSBs are losslessly compressed by a dedicated lossless compression method (e.g. JPEG-LS). The LSBs are compressed by a scalable encoder, i.e. in such a way that their description may be truncated at any desired point. A practical way of implementing the scalable encoder is to just rearrange the bit planes, without actually doing compression (since the LSBs are approximately uniform random variables, they cannot be losslessly compressed much anyway). This means that the scalable encoder first puts the most significant bit plane (of the LSB part) in the stream, followed by the respectively less significant bit planes. The compressed scalable data may now be truncated at any bit position in order to obtain the desired compression ratio. We also present a method to automatically and adaptively determine theMSB/LSB split point such that a scalable bit string is obtained without affecting the compression efficiency and without producing compression artefacts for near-lossless compression. To determine the split point, first a low-complexity DPCM-type prediction is carried out on the original pixel values to obtain the prediction error signal. Next, the split point is computed from the average value of the magnitude of this signal. Finally, applying a (lossless) color transform to decorrelate the image color components before compressing them provides a higher (lossless) compression ratio. To be able to provide an efficient completely lossless compression option, the color transform has to be perfectly reversible. In order to still be able to use perceptually relevant error criteria when near-lossless compression is used, we present the LYUV transform, a new lossless color transform that is the lossless equivalent of the YUV transform that is commonly used for lossy image compression.