A Reflection Removal Algorithm Based on Multi-angle Perception and Edge Guidance
When images are taken,the clarity and integrity of target image is often damaged by reflections from glass,windows,etc.These parts that cause interference to human vision are called reflection layers.The reflected image in a real scene is a complex coupling of the background layer and the reflection layer,and it is difficult to separate them.Rarely image reflection removal methods utilize the information of the reflection image itself to guide reflection removal,and how to effectively utilize this information is also a key issue.We propose a network for image reflection removal based on a multi-angle perception and edge information.Firstly,we propose a multi-angle perception recovery network to construct a coarse reflection removal result.Then,the following edge-guided removal network uses the image's edge structure information to optimize the coarse result and generate a final reflection removal result.In particular,the proposed multi-angle perception module is used to enhance the perceptual ability of the model and suppress the generation of reflection sequences.In addition,the edge structure of the image provides structural guidance for the reconstruction of the image from different scales,which further recovers the lost detail features.Experiments demonstrate the proposed network model has superior performance relative to existing reflection elimination methods.
国家科技期刊平台
NSTL
万方数据
