Very-low Bit-rate Image-coding On Adaptive Multigrids

Macq, Benoît [UCL] Queluz, MP. Simon, B.

We propose a new approach for very low bit-rate sequence coding (below 64 kbit/s) in which spatial information and temporal changes are encoded using similar tools. In this spatio-temporal integrated approach adaptive multigrids are used for both predicted images (the inter-images) and for the intra-images. The inter-images are mainly encoded by a reliable motion field analyzed and transcribed on an adaptive sampling grid (with variable sampling mesh). The intra-images are encoded using tree segmentation leading to a similar transcription. Universal entropy coding and image reconstruction based on overlapping waveforms are used for both intra-images and inter-images. This coding system is well suited for the cases where regular intra-image refreshes are demanded. Present results suggest that with the proposed method, the range of very low bit-rate can be achieved, maintaining a good subjective quality.