Method for repairing overexposed areas in HDR images based on block linear regression and guided filtering

To address phase discontinuity and point cloud data loss induced by overexposed regions in fringe projection profilometry (FPP), this paper proposes an image inpainting method combining block-wise linear regression with guided filtering. The method first calculates the absolute phase to identify regions with phase unwrapping errors as processing areas. Based on a local block-wise linear regression model between Gaussian-filtered low-exposure images and overexposed images, it approximately predicts missing pixel information. To enhance the continuity and reliability of fringe details, guided filtering technology is further employed to optimize the inpainting results. When processing fringe images containing large-area overexposed regions, this method requires only two sets of fringe images at different exposure times (a total of 24 images) to complete the restoration, eliminating the need for multiple exposure images required by traditional methods. Experimental results demonstrate that this method can effectively repair overexposed regions, restore fringe phase information, and maintain the integrity of 3D point cloud data, providing a feasible approach for three-dimensional reconstruction of high-dynamic-range (HDR) objects.