Information Technology Reference
In-Depth Information
Fig. 5 Example of selected features for Subject D
Ta b l e 2 Average and standard deviation of the computation time (in seconds)
Process
Average S.D.
Feature computation
0.787
0.045
Facial expression recognition
1.656
0.305
Emotional scene detection
0.001
0.000
Total
2.443
0.301
to obtain the facial feature points is not included because this process is performed
by an existing software application.
Facial expression recognition requires slightly long computation time due to the
ensemble clustering. On the other hand, the emotional scene detection is consider-
ably fast owing to the concise detection algorithm. As a whole, the proposed method
is fully efficient because it takes less than three seconds to detect emotional scenes
from a 5-minute-long video.
7Con lu ion
This paper proposed an emotional scene detection method for the better utilization
of lifelog videos. The proposed method is suitable for various lifelog videos because
it is based on unsupervised learning and does not require any training data sets. In
addition, the proposed method is fully efficient by introducing an ensemble learning
framework using a few useful feature values.
The experimental results show the effectiveness of the proposed method to de-
tect the emotional scenes with smiles. However, the performance of the proposed
method for the other emotions is unclear. Thus, evaluating the proposed method for
a wider variety of emotions is included in the future work.
Search WWH ::




Custom Search