论文部分内容阅读
在视频编码中，图象序列宏块活动性的正确分类有着非常重要的意义。本文提出了一种神经网络的图象序列宏块活动性分类新算法。算法首先计算宏块中16个亮度子块的亮度子块纹理掩蔽能量函数值，用来作为神经网络的输入信号，然后用BP神经网络来综合评价宏块的视觉活动特性，将宏块分成平坦区，边缘区，纹理区或皱纹理区。实验结果表明，本文提出的算法，充分考虑了人眼的视党特性，并用BP神经网络来划分宏块的不同活动性，可以得到比较正确的分类结果。正确的分类可以有效地应用于基于块变换的视频织码算法的量化器中，改善解码图象主观视觉的效果，降低图像编码主观和客观失真，保证国家质量。另外，由于每个宏块分类的独立性和神经网络的高度容错性与独立性，可以采用一个并行处理器实时地实现图象序列宏块活动性的分类。
In video coding, the correct classification of the motion of the macroblocks of the image sequence is of great importance. In this paper, a new algorithm for the classification of macroblocks in image sequences based on neural networks is proposed. Firstly, the algorithm calculates the brightness sub-block texture masking energy function values of 16 luminance sub-blocks in the macroblock to be used as the input signal of the neural network. Then the BP neural network is used to evaluate the visual activity of the macroblock comprehensively, Area, marginal area, textured area or wrinkled textured area. The experimental results show that the proposed algorithm fully considers the visual characteristics of the human eye and uses BP neural network to separate the different activities of the macroblocks to get more correct classification results. The correct classification can be effectively applied to the quantizer based on block transform video coding algorithm to improve the subjective visual effect of the decoded image, reduce the subjective and objective distortion of image coding and ensure the quality of the country. In addition, due to the independence of each macroblock classification and the high degree of fault tolerance and independence of neural networks, a parallel processor can be used to classify macroblock activity in real-time.