A novel multi-exposure fusion-induced stripe inpainting method for blade reflection-encoded images

The poor reflection of aircraft engine blades often leads to the loss of structured light encoded stripes on the surface, which affects the accuracy of the reconstructed 3D point cloud. Advanced multi-exposure fusion methods are not suitable for situations where stripes are severely missing. A new scheme for multi-exposure image fusion and image inpainting based on deep learning is proposed to address the above issues. Firstly, we propose an engine blade stripe image inpainting dataset (EBS-II). Secondly, for damaged blade stripe images generated by multi-exposure fusion networks, we designed a two-stage generation network based on edge discrimination guidance (TGEDG-Net) to recover the damaged areas in the image. The generator uses a progressive repair network to gradually recover missing structural and texture information, and the discriminator combines edge prior information to improve the ability to distinguish stripe edges and blade structural features. In addition, the network model has been extended to an interactive system that allows users to draw free masks to guide the network in targeting areas where reflections and stripes are missing in the actual scanned image sequence. The experimental results show that the proposed fringe image inpainting method is progressiveness. Meanwhile, the generated high-precision point cloud model verifies the effectiveness and practicality of the proposed scheme. Our dataset and code can be obtained at the following website: https://github.com/VDT-2048/TGEDG-Net.