Efficient auto-refocusing of iris images for light-field cameras

Light field photography provides a revolutionary possibility to reconstruct well-focused iris region from a 4D light-field image. However, such a “shoot and refocus” scheme is time-consuming in practice because it commonly needs to render an image sequence for finding the optimally refocused frame. This paper presents an efficient auto-refocusing iris imaging solution for lenselet-based light-field cameras. Firstly, a refocusing point spread function (R-PSF) is derived by detailed analysis of the relationship between refocusing depth and defocus blurriness. Secondly, an initial image is rendered at arbitrary depth. Thirdly, a content independent blurriness assessment method based on SVR (support vector regression) modeling is performed on the rendered image to locate depth shift from optimal focusing plane based on R-PSF. Finally, the optimally focused iris image is selected from a frontal candidate and a back candidate. Because our method only involves three times of image rendering based on precise localization of the optimal focusing plane, it is much more efficient than conventional “rendering and selection” solutions which need to render a large number of refocused images.