基于BP神经网络的颜色模糊量化方法

将BP神经网络用于颜色量化过程,提出了符合人眼颜色视觉特性的颜色模糊量化方法。对RGB颜色空间向量进行空间变换,提取得到颜色特征向量。对特征向量标准化处理后,作为BP神经网络的输入向量。将训练样本的期望类别输出做模糊化预处理,用模糊化后的隶属度值作为样本的目标期望输出。利用样本集对改进的BP神经网络进行训练.基于最大隶属原则对神经网络输入特征向量进行分类和量化。使用训练后的BP神经网络进行颜色量化的仿真实验,验证了所提出方法的有效性。

An Approach of Color Fuzzy Quantization Based on BP Neural Networks

An approach of color fuzzy quantization according with human color vision characteristics was proposed. Back Propagation (BP) neural networks were applied in the process of color quantization. Feature vectors were got via a color space transform between RGB and I1I2I3. The feature vectors were then standardized to act as the input vectors of BP neural networks. Training samples were preprocessed by fuzziness at first. The target expectation outputs of the samples were evaluated by the fuzzy membership value. Then the reformed BP neural networks were trained by those samples. The input color vectors were classified and quantized based on the most membership principle. Simulation experiments show the approach is effective.