Improved Shape-from-focus Reconstruction for High Dynamic Range Freedom Surface

The 3D reconstruction of freedom surface has been a major challenge for computer vision as well as metrology. Complex surface structures such as large undulations and deep grooves, along with surface reflection properties like high dynamic range (HDR) pose many difficulties for 3D reconstruction. Shape from focus (SFF) is widely studied for its low cost, ease of deployment, and effectively avoiding occlusion. However, its inherent disadvantages, such as inaccuracy and sensitivity to noise, cannot be ignored. This article demonstrates an improved depth map optimization method to solve these problems. Firstly, a more comprehensive and reliable depth recovery evaluation method is proposed. Then an efficient optimization algorithm is applied to the filtered depth maps. Comparable experiments confirm our method's effective suppression of noise and the reliable compensation of missing signals when measuring freedom surfaces.