Aiming at the traditional shape searching which needed to search in the whole shape space each time, this paper proposed a new approach based on classified face shape searching. This approach began with a shape space that contained diverse shapes. First it optimized the random forest classifiers by the feature selection method based on correlation and trained the random forest classifiers by training samples. Then it divided the shape space into several sub-spaces by random forest classifiers, and searched the sub-space that was the most similar to the current shape, and estimated the center of the sub-space and probability distribution. Finally, this paper employed cascaded regression to realize face alignment. The experiment result demonstrates the proposed approach obvious decreases the searching time and has good robustness in unconstrained environment on the 300-W database.

[2019-04-05] Printed Article /doc/2017.11.0774/v2