Local Feature Evaluation for a Constrained Local Model Framework

We present local feature evaluation for a constrained local model (CLM) framework. We target facial images captured by a mobile camera such as a smartphone. When recognizing facial images captured by a mobile camera, changes in lighting conditions and image degradation from motion blur are considerable problems. CLM is effective for recognizing a facial expression because partial occlusions can be handled easily. In the CLM framework, the optimization strategy is local expert-based deformable model fitting. The likelihood of alignment at a particular landmark location is acquired beforehand using the local features of a large number of images and is used for estimating model parameters. In this learning phase, the features and classifiers used have a great influence on the accuracy of estimation in landmark locations. In our study, tracking accuracy can be improved by changing the features and classifiers for parts of the face. In the experiments, the likelihood map was generated using various features and classifiers, and the accuracy of landmark locations was compared with the conventional method.