Summary: | 碩士 === 國立臺灣科技大學 === 電機工程系 === 105 === Facial deception detection is becoming a challenging problem for automatic inspection of surveillance videos. In this thesis, we propose a novel algorithm for differentiating the deception and truth based on visual clues. The Random Forest classifier is applied to track the facial landmark points, which is utilized to analyze the facial action unit based on the movement of the facial feature points. In addition, the biological and geometrical features are also considered, and the sequential forward floating selection (SFFS) is integrated to select the best feature combinations. The proposed method employs least-mean-square filter to significantly improve the robustness of the extracted features. To verify the extracted features for deception and truth identification, the pre-trained least-mean-square filters and the Support Vector Machine (SVM) are utilized. Experimental results demonstrated that despite the uncontrolled factors, illumination, head pose and facial of sheltering, in the videos, the proposed method is consistent in achieving promising performance compared to that of the former schemes.
|