双曲正切非线性函数和加均值量化算法对BP神经网络图像压缩处理的影响

Image Compression of BP Neural Network with Hyperbolic Tangent Non-linear Function and Adding Mean Quantization Algorithm

为了有效改善整个网络神经元的信息处理的特性,采用了双曲正切S形非线性传输函数和加均值量化算法,建立了BP神经网络图像压缩处理系统。由实验分析发现采用双曲正切传输函数的神经网络对图像处理的质量有所改善,并证明了这种传输函数的有效性。在此基础上又采用了加均值量化算法来提高神经网络图像压缩处理的训练速度。经过实验证明这种方法不仅不影响神经网络图像处理的质量而且缩短了网络传输图像的时间,由此证明了所提出算法的可行性。

An image compression and transmission system of BP neural network with hyperbolic tangent non-linear function and adding mean quantization algorithm is presented to improve the information processing ability of neural network. Through applying the hyperbolic tangent non-linear function to experiments, we have found that the image processing quality of BP neural network was improved, and effectiveness of the function was testified. Based on this, we also adopt the adding mean quantization algorithm to improve the neural network training speed of image processing. By analyzing the experiment results, we found that this method not only keeps the image processing quality, but also shortens the whole network transmission time.