【摘要】 为了提高神经网络在数码冲印颜色空间转化应用中的收敛效率,设计了一种基于反向传播(BP)神经网络的自适应优化模型.通过等间距分割由曝光实验测得的青、品、黄染料累积色差曲线,构建了在CIE Lab色域空间内分布相对规则的大容量学习样本.基于对网络训练过程中隐含层神经元节点间相关性和离散性的动态分析,合并或删除了冗余的节点结构.根据学习速率对网络收敛效率的影响,引入全局平均误差(GME)作为权值,对学习速率进行即时调整.仿真结果表明,与传统的BP神经网络相比,优化后的神经网络模型收敛成功率显著提高,收敛速度加快.输出精度能够满足色差要求.更多还原

【Abstract】 To improve the convergence efficiency of neural network in color space conversion of digital photofinishing,an adaptive optimization model based on BP neural network was designed.By dividing the exposal-tested color-difference curves of cyan,magenta and yellow cumulative dye with constant steps,a large number of samples distributing relatively regularly in CIE Lab color space were generated.Based on the dynamic analysis of hidden-neuron-node’s relativity and radiativity in the process of network training,the unwanted nodes were combined or deleted.According to the influence of learning rate on convergence efficiency of neural network,global mean error(GME) was implemented as weight to adjust learning rate in real-time.The simulation results indicate that the adaptive optimization model of neural network has better convergence probability and speed than the traditional BP neural network,and that its output precision can meet the demand of color-difference.更多还原