An Efficient Shadow Detection Algorithm Based on Hybrid Attention Mechanism
In the computer vision task,the presence of shadow pixels will have a negative impact on algorithms.Therefore,it is of certain research significance to improve the performance and effect of shadow detection network.The existing shadow detection algorithms using attention mechanism are not sufficient to extract cross-channel features and global pixel information.To solve this problem,this paper studies shadow feature information and combines the design idea of hybrid attention mechanism to build a new network Res-CCNet that integrates channel attention and spatial attention,and uses dense connection and feature fusion to reuse ne-glected features.Experiments are conducted on datasets SBU and UCF with three different evaluation criteria SER,NER and BER.The results show that the network algorithm has efficient shadow detection capabilities and application prospects.
shadow detectionattention mechanismCNNfeature extractionsemantic information
国家科技期刊平台
NSTL
万方数据
