Dyeing normalization image processing method and device based on improved GAN and product

The invention discloses a dyeing normalization image processing method, device and product based on an improved GAN, and relates to the technical field of image processing, and the method comprises the steps: inputting an image to be dyed and normalized into a trained improved GAN network, and obtaining a dyed and normalized image; according to the improved GAN network, a second decoder is introduced on the basis of a first decoder of a generator of an existing GAN network, so that double decoders are formed; the second decoder is formed by adding a self-attention mechanism layer between the activation layer and the residual layer of each convolution module of the first decoder; and a feature extraction module is also introduced into the GAN network. According to the method, the dyeing normalization effect is greatly improved by using the improved GAN network structure. 本发明公开一种基于改进GAN的染色归一化图像处理方法、装置及产品，涉及图像处理技术领域，方法包括将待染色归一化的图像输入到训练好的改进的GAN网络中，得出染色归一化后的图像；其中，改进的GAN网络是在现有GAN网络的生成器的第一解码器的基础上引入第二解码器，构成双解码器；第二解码器