Image Shadow Removal Algorithm Based On End-to-End Depth Network

With thecontinuous development of science and technology,images and other multimedia information continue to increase.However,in the process of image capture,it is susceptible to different factors,causing the image quality to decrease.Shadowing is a phenomenon of quality degradation caused by imaging conditions,which seriously affects the processing effects of subsequent image editing,target recognition and other tasks.Therefore,in order to remove its influence,increase the visual authenticity and physical authenticity of image editing and processing.This article focuses on the following research work on the shadow area of the image:(1)Aiming at the problem of easily mistake dark areas as shadow areas in the current shadow detection process,an image shadow detection method based on near-infrared information is proposed.Based on the knowledge of image shadow detection and near-infrared(NIR)information,the model is constructed,and some important attributes of NIR information are used to distinguish the shadow region from the shadow-free region.Temporary shadow candidate images obtained by fusing visible light images and near-infrared images,and the effect is enhanced by calculating color and NIR ratio images,and then the two are fused to form shadow area candidate images.Finally,the threshold selection method is used to calculate the optimal threshold to obtain a shadow mask to achieve the purpose of shadow detection.Through quantitative and qualitative analysis of the experimental results,the algorithm performance is fully verified.(2)For the existing shadow removal methods,there are problems such as lack of fully automatic models,poor texture restoration,and easy to ignore semantic information.An end-to-end deep network model is proposed for image shadow removal tasks to further improve the effect of image shadow removal.The model is constructed from two networks,an encoder-decoder network and a small and refined network.The former is used to predict the shadow scale factor and the latter is used to refine.At the same time,a new dataset(RSDB)is constructed for the problem of the small scale of the current shadow removal dataset.In order to verify the performance of the model,both quantitative and qualitative evaluation methods are used.In the quantitative analysis,the peak signal-to-noise ratio and structural similarity index are used as evaluation indicators,showing good performance.In the qualitative analysis,the model obtained a higher quality shadow-free image.Experimental results show that the model is more robust in both subjective vision and objective quantification.