Shape-From-Focus 3D measurement with image alignment

When using the Shape-From-Focus (SFF) approach to reconstruct 3-D profile for minute object such as BGA, it was found that the alignment of the image series become quite difficult. In addition, the image shift in terms of image position is nonlinear. In order to overcome this problem, the position shift for each frame was calculated, and images were reconstructed using the interpolation technology. It was found that the quality of 3-D profile reconstructed is greatly improved. In this paper, we also test a couple focus measures. It was found that using a Gaussian smoothing operator between the SLIM operator and the enhanced modified Laplacian had a better performance than others. After the calculation of the focus measure, there are several approaches to determine the surface profile for each individual image pixel. Medium filter without Gaussian fitting was suggested for shape determination after this test. Although the cloud data are still a little bit noisy, but with shape fitting, the ball height and the ball radius can be properly calculated.