An Efficient Projection for Fast Low Bitrate Omnidirectional Video Compression

The high computational complexity of VVC presents a challenge for its widespread adoption, specially in omnidirectional video encoding. To address this, a novel Packed Modified Craster Parabolic projection is proposed. This format is derived by geometrically transforming a modified Collignon projection into a rectangular shape, and employing a Deep Learning based model for post-decoding enhancement. Experimental results demonstrate average encoding time reductions of about 8% to 11%, with minimal impact on perceived coding quality (−0.57%). In fact, the method is particularly suited to low bitrates, where it is able to improve the perceived coding quality by about 0.2%.