Abstract

Surveillance videos have the feature of static background, which has been exploited in surveillance video coding by the background reference techniques. However, in long surveillance videos that last for several hours to several days, the appearance of background is varying due to the change of illumination, which incurs inefficiency of the background reference techniques. Moreover, intra frames in video coding cannot make use of the background reference and thus incur bit-rate burst. To solve this problem, we propose to separate illuminance out of appearance, and introduce a new kind of references known as reflectance reference (RefRef). RefRef consists of the reflectance component of the background in surveillance videos. The RefRef is invariant to illumination change, and thus can be put always in memory when encoding/decoding the surveillance videos of a specific camera. Especially, RefRef can be used for intra frames, together with the encoding/decoding of the illuminance component. Since the illuminance is usually more smooth than the appearance, RefRef provides higher compression efficiency for intra frames. Experimental results show that the RefRef method leads to more than 50% BD-rate reduction, compared to the High Efficiency Video Coding (HEVC) anchor, on typical surveillance videos under all-intra configuration.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call