摘要
本文提出了一种基于H.264降低图像尺寸转码中的宏块类型快速选择方案.文中以原始图像中对应四个宏块的非零系数总个数为准则选择宏块类型,首次提出以指数模型来描述阈值和重新量化参数的关系(Th-Qr模型),并将其线性化为一元线性回归模型,进一步使用最小二乘法来估计模型中的参数.帧内模式选择中,Th-Qr模型是用来区分I4×4和I16×16宏块;帧间模式选择中,Th-Qr模型区分出当前宏块所处区域的运动性质,进而简化候选宏块类型个数.实验结果表明,与全搜索法相比,在保证压缩性能损失不大的情况下,本文提出的帧内模式选择方法的速度能提高4到5倍,帧间模式选择方法的速度能提高约35倍.
A fast macroblock mode selection method in downscaling transcoder based on H.264 is proposed in this paper, which the total non-zero coefficients of responding four macroblocks in pre-coded flame is utilized as the criterion to select macroblock mode. We propose an exponent function model between threshold and re-quantizafion parameter （ Th-Qr model）. This model is converted into a linear regression model, and least square method is adopted to estimate parameters in the model. In the intra mode selection part, Th-Qr model is used to classify I4 × 4 or I16× 16;In the inter mode selection part, Th-Qr model is used to classify the motion activity in downscaled flame, and different candidate macroblock modes are used according to classified motion activity areas. We compared the PSNR performance and computational complexity of proposed method with the flail search algorithrn. For the intra mode selection part, we can speed up about 4 - 5 times, and for the inter mode selection part, we can speed up about 35 times.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2008年第5期848-851,共4页
Acta Electronica Sinica
基金
山东省科技攻关项目2005年(第三批)(No.2005GG3201117)
