Abstract
In digital optical imaging systems, especially digital cameras, autofocus based on image processing (ABOIP) has been widely implemented. Here we present some autofocus techniques for digital cameras that take into account some proverbial human vision characteristics.
None of the current focus measures, which are functions used to judge the sharpness of images, uses the color information in images as the human vision does. These focus measures fail when the target images are of different colors but almost equivalent in gray levels. We propose to improve these focus measures by analyzing the RGB components of the target images, instead of converting them into gray images to calculate the focus measures. This method sharpens the focus measure curves in general, and overcomes the aforementioned shortcoming of the traditional focus measures.
Most commercially available digital cameras with ABOIP choose an evenly sampled small area in the central part of a target image as focus window to reduce the computational consumption. Considering the unevenness of the photoreceptors in the retina, we propose to sample target images in a similar way. The central part is sampled with high resolution, and the sampling is decreased gradually from the center to the periphery. The uneven sampling can reduce the computation dramatically while keep most of the important information.
The human eye tracks the objects of interest with the fovea to keep high sampling of the targets and to sample the non-targets with very low resolution, or even totally ignore them. We propose to utilize two methods to choose focus window in target images dynamically. The first one is using pattern recognition to extract the objects of interest with certain a priori knowledge. The second is tracking the pupil of the photographer so that the region of interest can be determined and chosen as focus window.
SubbaraoM.ChoiT.NikzadA.(1993). Focusing techniques. Optical Engineering, 31(11), 2824–2836