Semantic segmentation network for complex background characters based on lightweight UNet
Towards the problems of traditional complex background character segmentation algorithm,a semantic segmentation network for complex background characters based on lightweight UNet is proposed.The network structure is based on UNet.In the feature extraction module,the traditional convolution is changed into deepthwise separable convolution,which greatly reduces the number of parameters and computation of the network feature extraction module.The residual learning module is introduced to solve the network degradation problem.Experiments were performed on the self-made dataset and H-DIBCO2018 open dataset,and compared with FCN8s,AttationUNet and UNet.Experimental results show that the proposed network has both computational efficiency and segmentation accuracy,and is practical.
万方数据
维普
