An efficient algorithm for oval shape face detection in color family photos

Detecting faces in an unconstrained practical environment remains a challenging task due to wide variation in quality and setup of the capture devices, different lighting conditions, whereby the subjects themselves may pose differently using various facial expressions with or without glasses. In this paper we present a novel architecture for face detection. The system has been trained and tested on a database that consists of 500 family photos with 932 faces. The results demonstrate a high accuracy face detection rate of 94.8%. The detector first locates the faces using the Viola-Jones method [1, 2], whereby each detected face is represented by a sub-image. However, a problem of background inclusion may occur for some sub-images. For such a case, a method that is based on pixel intensity [3] to detect eyes and mouth region is used to update the sub-images. Then, a skin-based method that comprises a color selection scheme based on hybrid approaches of Bayesian estimation and a fuzzy membership function is performed. Finally, an oval shape detection method is used to locate the oval-like shape of the face.

PDF ID 27208.pdf Restricted to Repository staff only Download (406kB)

Actions (login required)

View Item