Transcript Document

A Survey of Affect Recognition Methods:
Audio, Visual, and Spontaneous Expressions
Zhihong Zeng, Maja Pantic, Glenn I. Roisman,
Thomas S. Huang
Reported by
Chengsheng Mao
2011年1月11日
1
The Description of Emotion
• Discrete categories description: the most popular
example of this description this description is the
basic emotion categories, which include happiness,
sadness, fear, anger, disgust, and surprise. This
description of basic emotions was specially supported
by the cross-cultural studies conducted by Ekman [40],
[42].
• Dimensional description: the evaluation and activation
dimensions are expected to reflect the main aspects
of emotion. the evaluation and activation dimensions
are expected to reflect the main aspects of emotion.
2
Audio and/or Visual Databases of
Human Affective Behavior
3
Related work
4
Challenges
• Data collection for emotion recognition:
–
–
–
–
–
Spontaneous versus posed
Lab setting versus real-world
Expression versus feeling
Open recording versus hidden recording
Emotion-purpose versus other-purpose
• Labeling date for emotion recognition:
– If constantly asks a user for his/her emotion, we can be quite sure that
eventually the response would be that of anger or annoyance.
– Further research is required to achieve maximum utilization of
unlabeled data for the problem of emotion recognition
5
The research method on emotion
Experiment
design
Original
signals
Signal
preprocessing
1. Emotional model
2. SAM (self assessment
Manikin)
3. Others
1. Referencing
2. De-noise
Pure signals
Feature
extraction
Features
Data mining
1. linear (AR model, FFT, ...)
2. non-linear (complexity,
entropy, ...)
1. data preprocessing
2. classification and prediction
3. cluster analysis
4. discriminant analysis
6
Data mining
• Data preprocessing
– Data cleaning is applied to remove noise and correct
inconsistencies in the data.
– Data transformations, normalization may improve the
accuracy and efficiency of mining algorithms.
– Data reduction techniques can be applied to obtain a
reduced representation of the data set that is much smaller
in volume, yet closely maintains the integrity of the original
data.
7
Classification and prediction
• Classification and prediction
– A classifier or predictor based on a certain algorithm is built
by analyzing or learning from a training set made up of
database tuples and their associated class labels or values.
(supervised learning)
– For classification, the classifier is used to classify the test
data. Then the classification accuracy is calculated to
estimate the classifier.
– For prediction, the values are predicted through the
predictor and then an error based on the difference
between the predicted value and the actual known value is
computed to estimate the predictor.
8
Cluster analysis and discriminant analysis
• Clustering is the process of grouping the data into
classes or clusters, so that objects within a cluster
have high similarity in comparison to one another but
are very dissimilar to objects in other clusters.
(unsupervised learning)
• Discriminant analysis find the discriminant functions
based on the training datum and their class labels.
Then classify the data unknown category according
to the discriminant functions.
9