【作者】 卢超；

【导师】 唐慧明；

【作者基本信息】 浙江大学 ， 信息与通信工程， 2010， 硕士

【摘要】 随着视频处理与视频编码技术的发展,视频监控系统得到了广泛应用。人们对图像质量和压缩性能的要求是无止境的,高效的视频编码技术成为人们研究的热点。而现有的视频编码技术并没有充分考虑视频监控的特点,针对视频监控的视频编码还有较大的性能提升空间。论文首先研究了监控视频的特点与需求,并分析了现有编码技术在监控应用中的不足,然后提出了针对性的视频编码方法,主要包含以下几个方面:在监控视频中,常常存在大面积的静止区域,这部分区域不是监控的重点,现有编码标准对这些区域编码造成大量的码流浪费,为了更充分地压缩静止区域图像,同时又要保证运动区域图像的编码质量,论文提出了一种结合运动目标检测的编码算法,将图像分解成两个图层图像编码。一个是静止图层,可以拉长静止图层的图像组长度,采用图层跳过模式,有效地提高了静止区域的压缩率;另一个是运动图层,通过采用层间参考预测,提高了编码质量。实验表明这种编码方法是一种高效的视频编码方法。在监控视频中,常常含有噪声,特别是光线不好的场景,监控视频中的噪声将变得很大,对这种视频的压缩率很低。针对这种情况,论文提出了一种视频滤波与自适应合成参考帧相结合的编码方法。视频滤波对噪声有很好的抑制效果,提高了视频编码的压缩率,但同时也会滤除图像纹理和运动细节,这在监控应用中是不利的;另一方面,合成参考帧是一种经过简单滤噪的参考帧图像,能提供更好的预测参考,有效地提高了编码图像的质量,论文提出了自适应合成参考帧的编码方法,实验证明自适应的合成参考帧能进一步地提高含噪视频的编码性能和图像质量,但这种编码方法没有对待编码图像进行滤波。论文充分考虑了视频滤波与合成参考帧的优缺点,将二者相结合,在保证图像细节和编码效率的前提下有效地抑制了图像噪声,同时大幅度提高了压缩效率。更多还原

【Abstract】 Along with the development of the video process and video coding technology, video surveillance systems have been used widely. The requirement for the quality of the picture and the performance of compressing is endless, so that efficient video coding technology becomes the hot research topic. However, the existed video coding technology doesn’t consider the characteristics of video surveillance enough. Therefore, there is more space to improve the performance of video coding for video surveillance.Firstly, this paper studies the characteristics and the demands of surveillance video, and analyzes the disadvantage of existing coding techniques for surveillance, then proposes some video coding schemes, mainly includes the following aspects:In surveillance applications, there are often large static areas, which are not the focus. It is a waste of bit stream to code these static areas. In order to compress these areas largely and improve the quality of the image in the motion areas, this paper proposes an algorithm combined with the motion detection technique, which divides the picture into static layer picture and motion layer picture. For the static layer picture, the algorithm adopts a larger GOP (group of the pictures), and adds a layer skip mode, which improve coding efficiency greatly. For the motion layer picture, the algorithm adopts inter-layer prediction to improve the quality of the picture. The experiment shows the algorithm is very efficient.There is a lot of noise in surveillance video, especially in low light. It’s difficult to improve coding efficiency for this kind of video. The paper proposes an algorithm, which combines video filter and adaptive synthesized reference (SR) picture. Video filter can achieve a better compression performance by reducing the noise. However, it will blur the texture details and the motion details at the same time. This is disadvantage in the surveillance applications. On the other hand, the SR picture is a new reference picture which is filtered simply. The SR picture can provide better prediction to improve the quality of the coding picture. Based on this, an adaptive SR picture is proposed. The experiment shows the adaptive SR picture can improve the performance of the video coding further. However this method just filters the reference picture but not the picture to be coded. The paper takes video filter and adaptive SR picture into consideration, combines both of them to suppress the video noise. At the same time, the proposed scheme improves the coding efficiency and preserves the picture details.更多还原