3540222189 by Unknown

Author:Unknown
Language: eng
Format: epub

340 Jae Sik Chang, Eun Yi Kim, and Hang Joon Kim The contour split up into several parts, and finally two of them are converged to boundaries of faces and others are disappeared. This result represents that the proposed method runs independently on the number of faces in the given image. This experiment was performedwith initial contour and the threthold value ofstopping criteria is 5. 3.2 Performance Comparisons and Discussions To assess the validity of the proposed method, it was compared with the geometric active contour model presented in [4]. The model represents and evolves the contour using level sets. In the method using the model, the contour shrinks iteratively and the contour stops on the boundary that has high gradient values. For each video frame, ground truth was created by manually constructing accurate boundaries of each facial region. This is used to calculate the accuracy and the miss detection rate so as to evaluate performance of the proposed method. The accuracy and the miss detection rate are definedby where = pixels offacial regions detectedby the proposed method} = pixels offacial regions detectedmanually} : groundtruth ; # ofpixels in the region. Fig.3 shows the result of face detection using the geometric active contour model. The experiment performed on the same initial condition that is used on experiment 1(a). Fig. 3. Results of face detection using the geometric active contour model proposed by Caselles elal. [4] (experiment 3).

Face Detection with Active Contours Using Color Information 341 Fig. 4. The comparisons of the two methods in terms of accuracy, miss detection. Fig. 4 shows the comparisons ofthe two methods in terms ofaccuracy, miss detection rate and energy. As you can see, accuracy (miss detection rate and energy) ofthe proposed method increases (decreases) dramatically and maintain stable phase.

342 Jae Sik Chang, Eun Yi Kim, and Hang Joon Kim Conclusions4 In this paper, an active contour was used for detecting facial regions, regardless of pose, viewpoints, and noise. An input image is modeled using aMarkov random field (MRF), which is effective in describing the spatial dependency of neighboring pixels and robust to degradation and noise. And MAP was used for optimality criterion, so that the face detection is formulated as an energy minimization. For minimizing the energy, we used a active contour model based on the color information of human faces. In the active contour model, a contour is presented by zero level set of level function and evolved via level set partial differential equations. Experimental results show the effectiveness ofthe proposed method. However, the proposed method is a semi-automatic method, where it need a initial contour inputted by manually. Future works include determining the initial contour efficiently and adaptation the proposed method to real applications such as video monitoring system, face recognition system, etc. Acknowledgement This work was partially supported by both Brain Korea 21 and grant NO. R04-2003000-10187-0 from the Basic Research Program of the Korea Science & Engineering Foundation, respectively. References 1. 2. 3. 4. 5. 6. 7. 8. 9. M. H.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.