【作者】 何斌；

【导师】 陈纯； 张海翔；

【作者基本信息】 浙江大学 ， 计算机应用技术， 2007， 硕士

【摘要】 随着3G时代的到来，无线网络的带宽将呈现大幅度的增长。这给移动应用带来无限的契机。对于移动可视电话这样的多媒体应用来说，由于视频数据量往往比较大，2.5G带宽还难以承受较高质量视频数据传输。而3G有着2Mbps的室内速率和384Kbps的室外速率，这给移动可视电话的应用提供了很好的网络条件，使得移动可视电话的普及成为可能。由于移动设备硬件性能的局限性和可视电话的“移动”性(人们可能在行走的过程中或在车上等比较颠簸的环境中使用移动可视电话)，本文针对这些特点对移动设备上的编解码器做了相应的优化，主要包括运动估计搜索最佳匹配块算法的改进、以整型DCT替代传统的浮点DCT以及解码端的色彩空间转换优化等。MPEG-4是一种高性能的视频压缩标准。它的压缩过程包括运动估计和补偿、DCT变换、量化和熵编码。其中运动估计占了编码70％左右时间，所以有必要对这部分算法做一些有针对性的优化。本文主要结合三步搜索法和菱形搜索法的特点，对MEPG-4的运动估计算法提出改进，使之更适应移动可视电话视频图像运动变化比较大的特点，提升了编码速度。传统的浮点DCT变换由于含有大量的浮点和乘法运算，计算复杂度高，耗时多。为了改进这一缺点，本文采用整型DCT代替浮点DCT。整型DCT每次变换只使用了45次加法和18次移位操作，所以在编码速度上有较大的提升，而在视频质量上没有明显的下降。另外，由于MPEG-4解码后的图像是YUV格式，需要转换成RGB格式来播放。由于每帧图像在播放前都要经过YUV到RGB的转换，降低了整体解码速度。为了解决这个问题，本文采用查询表来转换色彩空间。这样，每次转换只需要一次查表操作，加快了整体解码速度。更多还原

【Abstract】 With the arrival of 3G, the bandwidth of wireless network will be significantly increased. 3G may bring boundless opportunities for mobile application. At present, the multimedia related applications, such as mobile TV, mobile videophone, MMS and so on, occupy the majority of the mobile applications. As the greater amount of video data, the 2.5G network can’t bear such real-time video data transmission. However, with the development of 3G and its high bandwidth, the mobile videophone application will become more and more popular.This paper focuses on the optimization for mobile video codec, because of the limitation of mobile device performance and the its "mobile" characteristics. The optimization work mainly contains the improvement of motion estimation algorithm, the replacement of Float DCT to Integer DCT and a new method of color space conversion.MPEG-4 is a high-performance video compression standard, which is based on DPCM/DCT model. It includes motion estimation and compensation, FDCT/IDCT, quantization and entropy coding. It is necessary to do some optimizations for the algorithm of motion estimation, because more than 70% coding time is spent by full search motion estimation model. At present, many algorithms for fast motion estimation are proposed, such as TSS, DS, etc. TSS is suitable for strong motion for its long step of first searching. DS is suitable for small motion for its accurate searching. Combination of these two algorithms will produce better results for mobile videophone.Float DCT costs large amount of computing power because of a lot of float operations and multiplications. We can use Integer DCT to replace Float DCT to reduce a lot of coding time. Integer DCT only contains 45 add and 18 shift operations without multiplications. Integer DCT not only improves coding speed, but also ensures the video quality not been decreased significantly.In addition, we suggest a new method to convert image from YUV4:2:0 to RGB565. We use a look-up table to convert image format instead of implementation of converting directly. It really saves a lot of decoding time.更多还原