Face Region Based Conversational Video Coding Bing Xiong, Xiaojiu Fan, Ce Zhu, Senior Member, IEEE, Xuan Jing, and Qiang Peng
Outline Introduction Motion-based sub-image rejection for face detection Face Region Based Bit Allocation in H.264/AVC Experimental Results and Discussions
Introduction Human visual system pays more attention to region of interest(ROI). Bit allocation How to find the proper face region Liu et al. [9] attempted to alleviate the artifact by smoothening the ROI coding priority mask with a mean filter.
Face detection Pyramid Structure in Face Detection Efficient Motion Detection Using Block Mean Difference Sub-Image Rejection Using Motion Features Verification With Facial Color Statistics
Pyramid Structure in Face Detection
Increase S will decrease N
Efficient Motion Detection Using Block Mean Difference Motion detection using the mean difference
Efficient Motion Detection Using Block Mean Difference
Sub-Image Rejection Using Motion Features
Thresholds are empirically set to 10% set to 10%
Sub-Image Rejection Using Motion Features
Remaining sub-images will be classified by Viola’s face detection method[12] Integral image Ada boost Cascade classifier
Verification With Facial Color Statistics skin-tone based face verification scheme can be found in [14]
Face Region Based Bit Allocation in H.264/AVC Face Contour Extraction Facial Feature Priority Based Bit Allocation Face ROI Based Video Coding Architecture
Face Contour Extraction The snake algorithm converges slowly Use “normal line search” on initial contour
Facial Feature Priority Based Bit Allocation T[i] remaining bits before coding ith MB QP[i] function introduced in [22]
Facial Feature Priority Based Bit Allocation
Face ROI Based Video Coding Architecture
Environment 100 frames,first frame intra-frame (I frame) and the rest frames as inter-frames (P frames) H.264/AVC codec JM GHz Intel(R) Core(TM)2 Duo CPU and 2G memory running Microsoft Windows XP
Experimental Results and Discussions