Lossless Compression and Interpolation for High Quality Still Images and Video Sequences

In the beginning lossless coding of still images and video sequences have been studied only for scientific purpose, such as biomedical and astronomical imagery, where even little losses could be relevant for the correct interpretation of the compressed data. In these last few years, the new technological innovations lead many professional users to the “digital world”. Professional photographers and the movie makers are day-by-day more interested to these new digital possibilities and they ask for higher image quality and the possibility to directly access the data captured by the sensor, without any automatic post-processing such as the white balance or the colour interpolation. In this thesis we present new lossless compression algorithms based upon the optimal prediction theory. We focused our research on the prediction step rather than on the prediction error modeling and context-based coding steps; in this way, we exploited all the possible correlation existing in the input images or video sequences. We started developing an optimal prediction for grayscale images, then we modified it for the colour image first, and, at the end, for the colour video sequence. We propose several way to exploit spatial, spectral and temporal correlations showing that they could be used to either improve the final compression ratio or reduce the complexity of the algorithm. We used the optimal prediction theory also to code hyperspectral images. In this case, the images present a strong spectral correlation which must be suitably exploited to achieve compression ratios that could be close to 4:1. We exploits this correlation via a data clustering used by both prediction and arithmetic coding. These studies on the lossless compression are also used in the European project IST-MetaCamera on the digital cinema. During this project we looked for the most suitable lossy-tolossless algorithm ables to work at the output data-rate of a professional 35mm format digital camera. Then we reduced the complexity of the selected algorithm and we designed a new “data-flow” which allows for the real-time implementation of the entire coding system. For the same project, we developed two colour interpolation algorithms for Bayer pattern images. The first one ensures a very high quality with a complexity lower than that of other less performant algorithms. The second one is a post-processing step useful to improve the perceptual quality of bayer colour images reconstructed using low-complexity demosaicking algorithms.