Image semantic segmentation method based on GAN network and FCN model

Abstract

In order to improve the accuracy of image semantic segmentation, an image semantic segmentation method based on generative adversarial network (GAN) and fully convolutional network (FCN) model is proposed. First of all, the network structure of the generator is improved. Introducing the residual module in the convolutional layer for difference learning makes the network structure sensitive to changes in the output, so as to better adjust the weight of the generator. Second in order to reduce the number of parameters and calculations, a small convolution kernel is used to halve the number of channels of the input feature map before using the large convolution kernel. Finally, the output of the convolutional layer and the output of the deconvolutional layer are connected by using the idea of a U‐shaped network to avoid low‐level information sharing. The proposed method was experimentally demonstrated on the PASCAL VOC 2012 and CamVid datasets. Experimental results show that the proposed method effectively improves the accuracy of image segmentation, and avoids inaccurate detection caused by insufficient image pixel information and noise interference. Its mean pixel accuracy (MPA) and mean intersection over union (MIOU) are higher than other comparison methods.

Image Segmentation in Complex Backgrounds using an Improved Generative Adversarial Network

“…On the other hand, existing GANs may encounter issues such as low quality of generated images and unstable model training when dealing with image segmentation in complex backgrounds. This limits the breadth and depth of application of GANs in image segmentation under complex backgrounds [33][34][35].…”