改进BP神经网络的自适应预测算法

针对视频编码中的耗时和图像不精确问题,提出改进BP神经网络的自适应预测算法。以区域生长、边缘检测和BP神经网络相结合,首先对图像进行对象分割,采用优化的传输结构,自适应调整学习率;然后在分割的基础上,进行自适应预测搜索编码。判断所选择块内是否有边界,若没有,不做运动估计,直接将当前块运动矢量置为零;若有,则进行自适应预测估计。根据不同的宏块特点,自适应地采取相应的搜索模式,减少搜索时间。实验结果表明,该算法与经典搜索算法相比,可取得良好的编码效果。

Adaptive predictive algorithm based on improved BP neural network

Owing to the problem of wasting time and illegibility of images in video coding,an adaptive predictive search algorithm based on improved BP neural network is proposed.This method is combined with region growing,edge detection and BP neural network,using optimized transmission structure to segment video object,adjusting the learning rate adaptively.Then whether the chosen block has boundary is judged.If it hasn＇t,the chosen block is not carried out searching and is given the zero vector.If it has boundary,adaptive prediction estimates are carried out.Depending on the characteristics of the block,the appropriate searching mode is taken adaptively,and the search time is reduced.The new algorithm reduces the computational complexity of the video compression algorithm and improves the accuracy of rebuild images.