IVML about | r&d | publications | courses | people | links

Robust Feature Detection for Facial Expression Recognition
Video files showing results (feature extraction and expression recognition) from naturalistic videos. Each video contains a 'tune', i.e. an audiovisual segment between two meaningful pauses.

The top half of the video shows the input video sequence, with the extracted feature points shown as yellow dots. The bottom half shows the four outputs of a neurofuzzy network, each corresponding to a quadrant of the 'emotion wheel' (cf. the paper).

As is shown, the confidence measures that correspond to each extracted feature point enable the neurofuzzy network to discard possible false detections and estimate the observed expression based solely on the feature points localized correctly.

High resolution AVI files
1frame_bar7783_7858sub.avi [download 110MB avi]
2frame_bar39926_39969sub.avi [download 64MB avi]
3frame_bar37569_37629sub.avi [download 88MB avi]
4frame_bar50488_50574sub.avi [download 125MB avi]
5frame_bar19435_19515sub.avi [download 117MB avi]
6frame_bar3330_3378sub.avi [download 70MB avi]
7frame_bar28030_28078sub.avi [download 70MB avi]

Streaming Quicktime files
1frame_bar7783_7858sub.mov [mov]
2frame_bar39926_39969sub.mov [mov]
3frame_bar37569_37629sub.mov [mov]
4frame_bar50488_50574sub.mov [mov]
5frame_bar19435_19515sub.mov [mov]
6frame_bar3330_3378sub.mov [mov]
7frame_bar28030_28078sub.mov [mov]

Contact information
Prof. Stefanos Kollias
Dr. Kostas Karpouzis
The Image, Video and Multimedia Systems Laboratory