Dynamic focus window selection using a statistical color model

We previously showed the necessity of utilizing dynamic methods to select focus window for passive autofocus in digital imaging systems. One possibility is to track the photographer's pupil through a modified viewfinder so that the region of interest in a target image can be determined, as previously described. Yet this assumes that a user is on site and he/she looks through the viewfinder, which is less and less practiced as a result of the availability of liquid crystal displays (LCD) on most consumer digital imaging systems. An alternative is to use pattern recognition to select focus windows when the imaging targets are known in advance and can be extracted from their background. In this paper, one of such cases, where the imaging targets are humans, is discussed in detail. The theoretical basis for dynamic focus window selection is briefly reviewed. And an example is given to demonstrate the effects of different focus windows on the imaging results. Then the focus window selecting technique using a statistical model of human skin colors is described in detail. The incoming target image in RGB color space is transformed into 2-dimension (r, g) space. Each pixel is binarized according to the relationship between its (r, g) value and the skin color distribution. Thus skin regions in the image are extracted. Morphological operations are then applied to the resultant binary image to reduce the number and irregularity of the skin regions. A rectangle can be fitted to the extracted skin region and used as the focus window. Experimental results are given to demonstrate the advantages of the proposed method.