Adaptive window selection for 3D shape recovery from image focus

In shape from focus (SFF) techniques, focus measure plays an important role. Conventionally, two fixed windows at two stages have been used to compute and enhance the focus measures, respectively. The window sizes largely affect the accuracy of the recovered depth map. A smaller window is unable to compute focus quality in textureless and smooth areas in the images. Whereas, a large window may over smooth or distort the object shape. Moreover, the use of any fixed window may not provide optimal results. In this paper, instead of using a fixed window at each stage, an adaptive window is proposed to compute and enhance the focus measure. At first stage, to compute the focus measure, a criterion based on dispersion of image intensities determines the size of the mask. Similarly, at second stage, to enhance the focus measure, a criterion based on dispersion of initial focus values estimates the window size. The proposed scheme is tested using image sequences of simulated and real objects. The comparative analysis has demonstrated the effectiveness of the proposed adaptive windows.

► 3D shape is reconstructed by using adaptive window selection.
► Adaptive window processing provides more accuracy than fixed window processing.
► Optimal window size can be determined by using statistical approach.