基于纹理复用的视频压缩研究

许多视频监控应用的目标是从视频图像中获取前景对象的行为关系信息,因此重构视频只要能够保留这些信息并在整体上符合人类视觉心理即可。提出一种基于纹理复用的视频压缩方法:(1)事先根据应用先验知识构造监控场景背景Sprite并分别预存在监控端和控制中心;(2)用背景差法依空-时次序提取前景对象;(3)对象分类;(4)按照纹理复用算法给出同类对象的复用参数;(5)对基本纹理和复用参数压缩编码。其特点是充分利用既有纹理信息表示同一对象,各处理模块在MPEG-4标准框架下简化了实际压缩信源,因此,其压缩比高于MPEG-4的既有结果,重构视频的主观评分在4分以上。此外,对象整体在确定位置嵌入背景Sprite的计算非常简易,解码实时性强。实验结果验证了方法的有效性。

Study on Video Compression with Texture Reuse

Since the objective of many video-based surveillance systems is to fetch the behavioral relations of the foreground objects from the captured video images,it meets the requirement of video reconstruction if only the above-stated information is preserved and the reconstructed video as a whole is in consistence with human visual psychology.A novel approach with texture reuse for video compression is presented:(1) the scene background sprite is built using a priori constraints offline and pre-stored both on the controlled side and the monitoring center respectively;(2) get the detected objects frame by frame using background subtraction in spatial-temporal order;(3) classify the objects;(4) evaluate the texture reuse parameters using the proposed algorithm;(5) do compressive coding on the representative texture and the reuse parameters.This approach is featured by exploiting the existing information to express the same object,under the framework defined by MPEG-4,the actual compressed source greatly simplified for each processing modular,thus its compression ratio is higher than the existing counterpart of MPEG-4,with the subjective visual score 4 or higher.In addition,it is computationally easy to fix a VO into the specified region of the Sprite,featured with strong decoding real timeliness.Experimental results have substantiated its effectiveness.