基于直觉模糊——神经网络的色情图像识别算法

网络中色情图像的传播严重影响了网络信息内容的安全性。为提高色情图像识别的准确度,提出了一种直觉模糊理论和FP（Forward Propagation）神经网络相结合的色情图像识别算法。算法以颜色直方图为底层特征,根据色情图像颜色分布情况,由模糊理论和直觉模糊理论共同构建图像特征矩阵;采用FP网络实现色情图像特征训练过程,其中特征矩阵的权重通过反向传播神经网络训练得到,以加权距离建立球形邻域半径;最后通过球形邻域覆盖情况识别色情图像。实验结果表明,该算法能够在不影响识别速率的前提下,有效的提高识别准确度。

Intuitionistic fuzzy theory - neural network basing pornographic image recognizing algorithm

Spread of pornographic image on the internet impacts security of information content seriously. In order to improve accuracy of pornographic image recognition, an image recognition algorithm combining intuitionistic fuzzy theory and Forward Propagation neural network was proposed. The algorithm used color histograms as basic features. According to distribution of pornographic image colors, it constructed image features matrix by fuzzy theory and intuitionistic fuzzy theory. And it applied FP network to implement pornographic image training. The neural network get weight of features matrix from training of back propagation network. Distance with weight was used to establish radiuses of sphere neighborhoods. The recognition algorithm recognized pomorgraphic images from range of sphere neighborhoods. The experiments showed that this algorithm could improve accuracy of recognition in the case of not decreasing the speed.