Abstract:This work presents a comparison among different focus measures used in the literature for autofocusing in a non previously explored application of face detection. This application has different characteristics to those where traditionally autofocus methods have been applied like microscopy or depth from focus. The aim of the work is to nd if the best focus measures in traditional applications of autofocus have the same performance in face detection applications. To do that six focus measures has been studied in four different settings from the oldest to more recent ones.
IntroductionIn face detection and face recognition methods (Pentland et al., 1994;Rowley et al., 1998;Gross et al., 2001;Hjelmas and Low, 2001;Yang et al., 2002;Zhao et al., 2003), borders play an important role because they de ne the facial features that appear in the face such as eyes, mouth and nose, which are needed to carry out the task. In a blurred image these facial features are not well de ned, and so the detection or identi cation can not be done. Blurred images can be obtained in a defocused camera because defocusing can be modelled as a low-pass ltering process, opposite to focused images which have a higher frequency content (Nayar, 1994). Thus, it is desirable that the image acquisition system has an autofocus mechanism. (Krotkov, 1987;Nayar, 1994;Lee et al., 1995;Subbarao and Tyan, 1998;Choi and Ki, 1999;Lee et al., 2001;Nathaniel et al., 2001;Kehtarnavaz and Oh, 2003;Kristan and Pernus, 2004;Shirvaikar, 2004;Park and Kim, 2005;Kristan et al., 2006).
Most of the published autofocusing algorithmssolve the problem of planar objects like in microscopy applications (Sun et al., 2004) or single object of interest like depth from focus applications (Nayar, 1994). In these applications, focus measures exhibit an ideal curve with a peak with step slopes at the lens position where the object is focused because there is only one object in the image or because it is a planar image. However, in human computer interaction, people do not always hold the same position in the image and exists more objects in the scene so the focus measure does not exhibit a clear maximum. In digital photograhy this drawback is eluded because the photographer selects the object of interest and centers it.