Image-based material reconstruction technique is one of the methods for obtaining texture assets.To create real-istic material appearances,it is essential to remove the original shadows from texture images.Existing shadow removal methods often face challenges such as limited diversity in datasets and limited range of shadow types that are capable of handling.To address the complexities and current limitations in material texture shadow processing,this research curated a high-resolution image dataset with diverse surface and scene characteristics.Utilising deep convolutional neural net-works,the study conducted training and analysis,aiming to provide insights into material texture shadow processing.Ex-perimental results demonstrated the effectiveness of the proposed method in shadow removal for complex scenes,expand-ing the scope of shadow removal while preserving relevant texture information.
维普
万方数据