【作者】 闫小涛；

【导师】 张翠芳；

【作者基本信息】 西南交通大学 ， 信号与信息处理， 2009， 硕士

【摘要】 要在有限的码率下得到效果好的图像质量,有效的码率控制算法必不可少。但是在H.264标准码率控制算法G012中,仅仅在编码前对一帧进行目标比特分配,没有对实际编码比特和目标比特的差距进行处理;MAD(MeanAbsolute Difference)预测算法只采用单个宏块预测,精确度不高。本文通过增加一个帧层补偿算法,对实际编码比特和目标比特的差距进行处理;给出一种改进的宏块层MAD预测算法:大小三角形预测算法,利用多个宏块预测当前宏块MAD。通过仿真实验,证明改进后的算法有更好的性能。绪论主要介绍了码率控制技术的国内外研究现状以及本文的主要研究内容。基本知识是码率控制的相关技术介绍,主要介绍了帧层的码率控制技术和宏块层的MAD预测技术。帧层码率控制的改进算法:帧层比特补偿算法。此算法主要针对连续几帧甚至更多帧出现实际编码比特总是大(小)于目标编码比特的情况,在帧层增加了一个补偿算法:当连续出现的帧数较少时,采用比特补偿;当连续出现的帧数较多时,直接改变量化参数。通过仿真实验证明,改进后的算法PSNR值更高,实际缓冲更加接近目标缓冲,实际编码比特也更加接近于目标编码比特。宏块层改进的MAD预测算法:大小三角形预测算法。此算法主要针对利用单个宏块预测当前宏块MAD精确度低的情况,利用多个宏块进行预测当前宏块:利用宏块之间的相关性,将相关性大的宏块赋予相对大的权值,相关性小的宏块赋予相对小的权值。通过仿真实验可以得知:改进后的预测算法有更高的PSNR值,实际缓冲更加接近于目标缓冲,实际编码比特更加接近于目标编码比特。总之,通过改进,各项性能指标都有所提高,这说明改进后的算法优于对比算法。更多还原

【Abstract】 In order to get a better picture using limited bits ,an effective rate control algorithm is indispensable . In G012, only a target bit is allocated to a frame before coding it, but the difference between target and real bit is not processed; MAD predicition is only operated using linear prediction, so precision degree is not high. This paper addes a frame compensation algorithm and study a new macroblock MAD prediction algorithm:the Big-Small triangle prediction algorithm.Through simulation experiments, it proves that the new algorithm is better.The introduction mainly introduces rate control’s present research situation in home and abroad ,and the major research content of this text.Some basic knowledge is mainly the correlation technique introduction of rate control: it mainly introduces the frame rate control technique and macroblock layer MAD’s prediction technology .Frame layer compesation algorithm mainly aimes at the situation when real bits is bigger(smaller) than target bits for several frames even more frames. In order to settle the problem , bits compensation algorithm is given. When the number of continuous frames is not so more, bit compensation is adopted; otherwise, the quantification parameter will be adjusted. Experiments prove that the improved algorithm’s PSNR is higher, the real buffer bufferfullness is closer to the target bufferfullness, and the real bit is closer to the target bit.The improved macroblock MAD predicition algorithm aims at former algorithm’s lower prediction precision . In order to settle the problem , current malblock MAD’s prediction uses more malblocks combinating with malblock correlation , and the malblocks are devided into two kinds : neithboring current malblock and not. I use defferent ways towards defferent malblocks . Experiments prove that the improved algorithm’s PSNR is higher, the real bufferfullness is closer to the target bufferfullness, and the real bit is closer to the target bit.Anyway, the improved algorithm has a better performance indicators, and it proves that the improved algorithm is better.更多还原