Object-based video compression
To answer the needs of the deaf and hearing impaired community, the Telesign project aims at proposing pre-competitive technological solutions for transmitting, at an affordable cost for the end-user, conversations in sign language over ISDN networks. Here, the technological lock lies in jointly accounting for such constraints as low bitrate (64 kbps), optimal trade-off between video rate and quality, and preservation of message integrity and semantics.
ARTEMIS contributions have concerned:
- The specification of an interoperable approach based on video compression (rather than on gesture recognition) leading to define the required compression rate by coding gestures as well as face and chest which are also involved in sign language communication;
- The design of a non-supervised segmentation method into meaningful video objects (hands, face, chest and background);
- The implementation of an MPEG-4 object-oriented video codec, allowing for bitrates compatible with ISDN network capacity;
- The development of a validation testbed for assessing the intelligibility of transmitted messages.
Extensive experiments carried out over 3 finger-spelling and French sign language datasets, have demonstrated the relevance of the proposed framework.
Partners: Orange Labs - France Telecom R&D