Extraction of regions of interest from face images using cellular analysis
COMPUTE '08 Proceedings of the 1st Bangalore Annual Compute Conference
Hi-index | 0.00 |
In this paper, a novel face detection algorithm is proposed which can detect both face and eye pupils efficiently at the same time. Interestingly, the results of face detection can be used to identify the regions of interest for pupil detection, and the results of pupil detection can be further contributed to detect more precise faces. A cascaded face filter is first constructed by using an adaboosting algorithm, which can rapidly filter out the non-face regions and keeps the possible face regions. Based on a radial-symmetry transform, the signal of eye pupil in a face candidate is considerably enhanced and becomes easy to detect. With an eye-pair checking process, the two pupil candidates are chosen as the outputted pupils if their corresponding face image obtains the highest verification score which is also larger than a predefined threshold. Experiments on the famous BioID face database have shown that 90.0% of faces can be successfully detected, and among the detected faces about 98% of eye pupils are detected with highly acceptable precision.