Khang Nguyen

Depth from Defocus applied to Auto Focus

Depth from Defocus (DFD) is known as the technology which is able to estimate depth in the scene by a monocular camera without any additive devices. Using this advantage of DFD, we improved the speed of Auto Focus (AF). To apply DFD to AF, it is necessary to capture 2 images that have a small amount of “difference in focal positions” (we call DiFP). In this paper, we show the performance of the depth estimation when the DiFP is changed, and verify that Low Pass Filter is a good solution to get robust depth.

DFD2.0: Motion robustness by amplitude domain approach

Depth from Defocus (DFD) is known as a technology which is able to estimate depth by a monocular camera, without any additive devices. However, it has to get two blurred images with different focused distance sequentially so that subjects or the camera itself can move in the short interval of consecutive shots. Conventional DFD is intolerant to such image shift especially in the weak texture part. In this paper, a new evaluation function for DFD is shown in order to achieve the motion robustness in frequency domain. The Conventional DFD algorithm is modified considering the image shift, and in the derived formula both shift and phase components vanished. As for DFD technology, the formula is insensitive to the image shift because the defocus blur appears only in the amplitude part. As a result, we also confirmed that our algorithm overcomes conventional methods especially for the real life with rapidly moving subjects.