Face region detection using DCT and homomorphic filter

Authors:
Kyunghwan Baek;Yeongsoo Chang;Daeuong Kim;Youngtak Kim;Boohyoung Lee;Hwanik Chung;Youngjoon Han;Hernsoo Hahn
Affiliations:
School of Electrical Engineering, Soongsil University, Seoul, Korea;School of Electrical Engineering, Soongsil University, Seoul, Korea;School of Electrical Engineering, Soongsil University, Seoul, Korea;School of Electrical Engineering, Soongsil University, Seoul, Korea;School of Computer Science & Engineering, Kongju National University, Cheonan City, Chungnam, Korea;Dept. of Internet Information, Kyungbok College, Gyeonggi-do, Korea;School of Electrical Engineering, Soongsil University, Seoul, Korea;School of Electrical Engineering, Soongsil University, Seoul, Korea
Venue:
ISPRA'07 Proceedings of the 6th WSEAS International Conference on Signal Processing, Robotics and Automation
Year:
2007

Citing 5
Cited 0

Omni-face detection for video/image content description

MULTIMEDIA '00 Proceedings of the 2000 ACM workshops on Multimedia
Detecting Faces in Images: A Survey

IEEE Transactions on Pattern Analysis and Machine Intelligence
Lambertian Reflectance and Linear Subspaces

IEEE Transactions on Pattern Analysis and Machine Intelligence
Robust Real-Time Face Detection

International Journal of Computer Vision
Global contrast factor - a new approach to image contrast

Computational Aesthetics'05 Proceedings of the First Eurographics conference on Computational Aesthetics in Graphics, Visualization and Imaging

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper proposes a new approach for the detection of face regions where the facial features such as eyes, nose, and mouth may exist with very high probability. It uses the characteristic of DCT (discree cosine transformation) that concentrates the energy of an image into lower frequency coefficients. Since the facial features are pertained to relatively high frequency in a face image, the inverse DCT after removing the DCT's coefficients corresponding to the lower frequencies generates the image where the facial feature regions are emphasized. For an efficient application of the proposed method, homomorphic filter is used with the global contrast factor. The DCT coefficients are eliminated so that the global contrast factor can be maximized. Once the inverse DCTed image is obtained, the face region is found by analyzing the topology of facial features. The proposed algorithm has been tested with various images collected randomly. The experimental results have shown a superior performance even when an image has a complex background.