基于JND模型的H.264算法及视频传输系统实现

发布时间：2018-06-19 01:02

本文选题：H.264 + JND视觉模型　；参考：《电子科技大学》2014年硕士论文

【摘要】：随着互联网技术和流媒体技术的迅猛发展,使实时视频通信技术也取得了很大的进步。今天,实时视频传输系统已经广泛应用于军事指挥,安全监控,网络社交等多个领域[1]。与此同时,人们对视频的传输速度和视频的质量的要求也越来越高。要在有限的带宽条件下流畅的传输高质量的视频图像,就必须采用高性能的硬件和高效的编码算法。H.264是适用于低码率视频传输的新一代视频压缩标准,它比以往的编码标准的编码效率都要高,编码质量也更好,网络适应性更强。本文首先对H.264编码标准进行了阐述,介绍了其中几个比较关键的编码技术,例如帧间预测、帧内预测、DCT变换和量化等,为本文后面将JND模型应用于H.264编码算法奠定了理论基础。随后,本文介绍了人眼的亮度特性、空间频率敏感特性以及对比度掩蔽特性,然后根据Wei提出的应用于JPEG标准的JND模型,提出一种改进的应用于H.264标准的JND模型。本文JND模型的亮度自适应因子不是线性函数,而是抛物线,更加符合人眼的视觉特性,而且根据DCT系数的频率分量对宏块进行分类,计算出的对比度掩蔽因子更加准确,同时通过反复试验,对参考模型中的一些参数进行了修正,使模型适用于H.264的离散余弦变换。我们对JND模型的优劣进行了仿真测试,结果表明本文的JND模型性能良好。接着,本文将JND模型应用于H.264编码算法,利用模型计算得到JND阈值来去除DCT系数中的视觉冗余,实验结果表明,添加JND模型之后编码大约节省了6%的编码码率;同时我们还利用JND阈值对H.264编码图像宏块时使用的编码模式进行筛选,缩小了模式选择范围,降低了编码计算量,提高了编码效率。最后,本文利用虚拟机、FFMPEG、Jrtplib以及JND视觉模型搭建了一个仿真的实时视频传输系统,测试了系统的实时性;同时,利用大恒相机、FFMPEG、MFC、PC机实现了一个真实的实时视频传输系统,详细描述了系统软件的编写,并对软件的视频播放、视频保存、视频截图以及视频翻转等功能进行了测试,测试结果表明软件所有功能得到实现,系统运作正常。
[Abstract]:With the rapid development of Internet and streaming media technology, real-time video communication technology has made great progress. Today, real-time video transmission systems have been widely used in military command, security monitoring, social networking and other fields [1]. At the same time, the speed of video transmission and the quality of video are increasingly demanding. In order to transmit high quality video image smoothly under limited bandwidth, it is necessary to adopt high performance hardware and efficient coding algorithm. H.264 is a new generation video compression standard suitable for low bit rate video transmission. It has higher coding efficiency, better coding quality and better network adaptability than previous coding standards. In this paper, the coding standard of H.264 is introduced, and some key coding techniques, such as inter-frame prediction, intra-frame prediction, DCT transform and quantization, are introduced, which lays a theoretical foundation for the application of JND model to H.264 coding algorithm. Then, this paper introduces the luminance characteristics, spatial frequency sensitivity and contrast masking characteristics of human eyes. Then, an improved JND model applied to H.264 is proposed according to the JND model proposed by Wei for the JPEG standard. In this paper, the brightness adaptive factor of JND model is not a linear function, but a parabola, which is more in line with the visual characteristics of human eyes, and classifies macroblocks according to the frequency components of DCT coefficients, and the calculated contrast masking factor is more accurate. At the same time, some parameters in the reference model are modified through repeated experiments to make the model suitable for the discrete cosine transform of H. 264. The simulation results show that the JND model has good performance. Then, the JND model is applied to H.264 coding algorithm, and the JND threshold is calculated to remove the visual redundancy in DCT coefficients. The experimental results show that the coding rate is about 6% less after adding JND model. At the same time, the JND threshold is used to filter the coding mode used in H. 264 image macroblock, which reduces the range of mode selection, reduces the amount of coding computation and improves the coding efficiency. Finally, this paper uses virtual machine FFMPEGG Jrtplib and JND vision model to set up a real time video transmission system, and tests the real time of the system. At the same time, a real time video transmission system is realized by using FFMPEGG MFCPC. The system software is described in detail, and the functions of video playing, video saving, video capture and video flipping are tested. The test results show that all the functions of the software have been realized and the system is running normally.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN919.8

【参考文献】