标题: | 基于位元-失真理论之H.264/AVC位元控制研究 The Study of Rate-Distortion Theorem forH.264/AVC Rate Control |
作者: | 许益宾 Hsu, Yi-Pin 陈福川 Chen, Fu-Chuang 电控工程研究所 |
关键字: | 位元控制;位元失真理论;H.264/AVC;Rate control;Rate-distortion theory;H.264/AVC |
公开日期: | 2011 |
摘要: | 位元控制在讯号处理上扮演相当重要地位,对于视讯压缩的应用更是重要。其中,位元控制有两个取舍点作为相互影响的因子,一为资料品质,也就是失真度,另一为资料量。在视讯压缩系统上,共定义出两种不同的影像压缩型别,分别是Intra-Frame与Inter-Frame形式。 Intra-Frame扮演整个影像压缩的起始点,除了要保持高度清晰的影像品质外,在位元的使用上更需精确应用,主因为接下来的影像需参考此一影像并使用剩下的位元进行压缩。为了增加压缩效能,或是减低位元需求,常会加入大量的Inter-Frame形式的影像。在即时的视讯压缩系统中,藉由前一张影像的资料品质与资料量来快速预测量化系数是位元控制所扮演的脚色。然而,真实的影像资料是会随物体以及时间产生变化,往往两张画面就有可能产生极大的落差。因此,如何藉由前一张影像的资料品质与资料量来预测出精确的量化系数,因而产生稳定的影像品质或是使用精确的位元数成为重要的研究课题。 本研究以视讯压缩H.264/AVC为主,架构于原始的位元-失真理论,而此基准点也是目前H.264/AVC在位元控制所采用的方法,探讨其现阶段位元控制的法则,得到一个可行性的失真度预测方法,进而将之应用于Intra-Frame与Inter-Frame的两个层面。更近一步的讨论,基于这两个层面,在不同的应用考量上,共推导出四个不同的位元控制演算法。 基于Intra-Frame的位元控制演算法。 利用所推导得到的失真度预测方法,藉由常见的标准测试影像进行参数训练,进而得到快速且精确的Intra-Frame位元控制演算法。 基于Inter-Frame的位元控制演算法。 利用所推导得到的失真度预测方法与位元估测方程式,透过Lagrange最佳化的推导,提出一个解析性的精确位元控制演算法,改善原本H.264/AVC以及现有着作在低位元率限制下表现不佳的缺点。 在实务应用上,优先考量嵌入式系统的适用性。藉由H.264/AVC内建的量化系数预测技术,得到初始的量化系数,利用所推导得到的失真度预测方法以及影像直方统计曲线,进行量化参数修正。 藉由所推导出的方程式来有效的预测影像复杂度,以达到较先前研究更精确量化参数数值。目的在于解决低位元率的视讯传输之应用。 由原始的位元与失真理论,推导出失真度预测方法。为验证此一方法其效能的适用性,本文基于四种重要且常见的应用,将此一方法加入,藉由实验结果指出,此一方法确实能有效运用位元来提升影像压缩品质。此外亦有将嵌入式系统的硬体限制考虑,让此方法更能贴近产业界应用。 Rate control plays a vital role in signal processing and video compression. Two degrees of distortion and consumed counts for measurement are selected in a tradeoff consideration. The video compression system defines two distinct frame types, as follows: intra-frame and inter-frame. The start frame in encoded video sequences is assigned as an intra-frame, and the goal is to maintain a higher image quality and lower encoded bits because the intra-frame is referenced after the encoded image, and the remaining bits are used for compression. In a real-time video compression system, referred to as H.264/AVC, the determination of the quantization parameter by rate control must refer to the encoded results in the previous frame. However, sudden occurrences such as fast-moving objects or complex backgrounds in two neighboring frames are usual. Thus, because of encoded results in previous images, producing stable image quality or accurate bits consumption is a crucial topic. This study focused on the H.264/AVC video compression system and rate-distortion theory to obtain a workable distortion prediction method for intra-frame and inter-frame rate control. Subsequently, we used this distortion prediction method to develop four rate control algorithms for various applications. Based on the intra-frame rate control algorithm According to the distortion prediction method and a large number of test images, we obtained a set of system parameters to develop a fast and accurate intra-frame rate control algorithm. Based on the inter-frame rate control algorithm We propose an analytic inter-frame rate control algorithm according to the distortion prediction method, H.264/AVC rate prediction method, and Lagrange optimization. This algorithm can solve the problem of low bit rate limitation in H.264/AVC. In real applications, we first considered the limitation of embedded systems. The initial QP value can be obtained through the built-in technology of QP calculation. We developed the intra-frame rate control algorithm according to the initial QP value, the distortion prediction method, and the information of image histogram. In the proposed distortion prediction method, we can predict the measurement of image content complexity efficiently to obtain a more accurate QP value than by using the current algorithm. This algorithm was developed for low bit rate video conferencing. We derived the distortion prediction method according to the original rate-distortion theorem. We applied the method on four differing applications to verify the results. Based on the simulation results, the method improved compressed image quality efficiently. We also considered hardware limitations for industrial application. |
URI: | http://140.113.39.130/cdrfb3/record/nctu/#GT079412803 http://hdl.handle.net/11536/40729 |
显示于类别: | Thesis |
文件中的档案:
If it is a zip file, please download the file and unzip it, then open index.html in a browser to view the full text content.