An all-zero block detection algorithm based on linear classifiers is proposed to reduce the complexity of the H. 264 video encoder and to improve the lower detection rate of common all-zero block detection methods. Five effective features are selected according to the characteristics of the encoder. Then the algorithm uses two different linear classifiers to separate all-zero blocks from non-all-zero blocks based on the encoding situation of the reference block. The training samples are selected from different videos, and the weights of the features and the thresholds of all-zero block detection under different quantization parameters are trained using the Fisher criterion. The thresholds under different quantization parameters are fitted into a quadratic polynomial of the quantization parameter using the least square method. Experimental results show that the proposed algorithm achieves better detection rate than that of other algorithms, and can effectively shorten the encoding time while maintaining the video quality.%针对H.264视频编码中传统全零块检测方法检测率较低的问题,提出了一种全零块检测的线性分类器算法.根据编码特性选择5个变量作为全零块检测的特征,基于参考块的全零块情况,设计了2个不同的线性分类器来区分全零块和非全零块,然后选取代表不同运动程度的视频样本,利用Fisher准则训练得到2个分类器的权系数和不同量化参数下全零块检测的阈值,最后用最小二乘原理将各个量化参数下的阈值拟合成量化参数的二次多项式,从而得到最终的全零块检测的线性分类器.实验结果表明,新算法在基本不降低视频质量的同时,能获得比其他现有算法更好的检测率,并且有效地缩短了编码时间.
展开▼
