用H.263标准对甚低码率图像编码时，经过帧间预测后得到的运动补偿数据通常很小，对这些数据再进行DCT和量化后往往成为全零块，Alice Yu算法和周算法是预先判别全零系数块的较为有效的方法，但在对较为复杂的序列图像进行预测时分别出现了较大程度的误判和漏判。针对这些缺点，该文提出了一种新的全零系数块的判别方法，它具有能随量化级的变化自适应地调整全零块的判断阈值、无需任何附加运算和对图像序列内容复杂程度不敏感的优点，将该方法应用于H.263编码器中，对Miss America和News图像序列进行仿真实验。实验表明，大约有40％-80％的块可以在做DCT和量化前被判别为全零系数块，大大减少了编码的时间，同时图像质量的下降控制在0.0005 dB以内。
For most sequences at very low bit rates, the value of motion compensation data usually is very small, and will all be reduced into zero after DCT and quantization. Alice Yu method(1997) and Zhou method(1998) both are effective to predicting these All-Zero Coefficients (AZC) blocks, but these two methods will, to a large extent, mis-justice and leak out the AZC blocks respectively, especially for complex sequences. In order to overcome these disadvantages, a new method of early predicting AZC blocks is proposed in this paper. In this method, no additional computation is needed and the threshold can be changed with the quantization level. Applying this method to H.2G3 encoder, about 40%-80% blocks of Miss America and News picture sequences can be determined to be AZC blocks before DCT and quantization. It can greatly reduce the encoding computation time and is very useful to the software implementation of H.263 in real-time. In addition, PSNR of decoding sequences is controlled within O.OOOSdB.