3D Shape Reconstruction Using Focus Clue From Defocus Blur Detection

3D shape reconstruction has numerous applications in computer vision. Shape from focus (SFF) is one of the 3D shape reconstruction techniques that employs the focus information of the images. Accurate estimation of the focus values of image pixels is a critical step in SFF such that it plays a significant role in determining the overall quality of the reconstructed 3D shape. Traditionally, in SFF, hand-crafted focus measure (FM) operators are used, which do not perform well under diverse and complicated situations. In this work, we propose to exploit the pre-trained deep convolutional neural network (CNN)-based model to estimate the focus values of image pixels. Experimental results showed that the proposed method yields better focus measurements which leads to high-quality shape reconstruction. This opens a window for a vast variety of deep learning-based single image focus/defocus estimation techniques to be employed for the 3D shape reconstruction in SFF.