Image Shadow Removal Method Based On Residual Neural Network

The removal of image shadows is a challenging task that requires researchers to fully understand the semantics of surrounding scenes while detecting shadows in the image.Most of the methods of de-shading first locate the shadow area and then use the reconstruction algorithm to remove the umbra and penumbra.However,it is a very difficult task to accurately detect shadow areas.The shadow removal method based on physical factors ignores the semantic information of the surrounding scene while adapting to high quality images.The statistical-based shadow removal method requires manual annotation of the shadow features,and the effect of shadow removal largely depends on the accuracy of manually marking the shadow regions.The deep learning based method has achieved competitive results in the field of image processing such as deshaping.Many scholars have used deep learning methods to remove shadows from images and achieve high quality results.This article removes shadows from a single image based on a depth residual network.This paper is based on deep neural network,combined with physical factor based shadow removal method to achieve single image deshaping.Based on the residual network training model,the residual network can effectively avoid the gradient dispersion and gradient explosion phenomenon due to the excessive number of network layers.The network model trained in this paper has better generalization ability and less interference from the external environment.High-quality shading can still be achieved with complex background textures,rich textures and non-fixed light sources.The network trained in this paper can accurately detect the shadow area and achieve the removal.The method presented in this paper can achieve high quality results on both public data sets and experimental data sets.