| Summary: | 碩士 === 國立臺灣大學 === 電信工程學研究所 === 100 === Based on the increasing of accessible data and the fast development of the computational technology, machine learning attracted lots of attention in the last ten year because of the great demand of automation in human life. Now in the disciplines of pattern recognition, robotics, artificial intelligences, computer vision, and even economics, machine learning has been an indispensible part to extract and discover the valuable information from data.
On the other hand, human face related topics such as face detection and recognition became important research fields in pattern recognition and computer vision during the last few decades. This is due to the needs of automatic recognition and surveillance system, the interest in the human visual system on human face perception, and the design of human-computer interface, etc.
In this thesis, we focus on using machine learning techniques for facial expression recognition. A facial expression recognition framework is proposed, which includes four steps: feature extraction, denoising mechanism, dimensionality reduction, and facial expression determination. The widely-used local binary pattern feature (LBP) is modified and combined with a new feature extraction method, local phase quantization (LPQ) to represent the facial expression. Since the extracted features are noisy and contain unrelated information for expression recognition task, a denoising mechanism is proposed. Due to the denoising mechanism, the denoised features are more representative for facial expression. Different from the existing dimensionality reduction algorithms, an expression-specific dimensionality reduction algorithm is proposed based on the special properties of facial expression. Finally, the reduced features with more meaning for facial expression are fed into the widely-used Support Vector Machine (SVM) and K-nearest neighbor classifier. From the experimental results, the proposed framework and algorithms achieve the highest recognition rate against the existing methods based on the JAFFE database.
|
|---|
