Local feature extraction and recognition under expression variations based on multimodal face and ear spherical map
Yihang Li, Zhichun Mu, Tingting Zhang · 2016
In unconstrained scenes, the change of expression and pose may lead to mismatching of the human face and ear images, and the recognition rate may also decrease. A method fusing depth and texture information is proposed to deal with the problem. We employ different recognition strategies based on the different characteristics of the spherical depth map and the spherical texture map. The learning to rank approach is applied to select the key points of high repeatability and stability. An improved SIFT method and a LBP-like algorithm are applied in the following process. Sparse representation is used for recognition and then Bayesian decision-level fusion for the improvement of final results. And the experiments prove the effectiveness of our approach evaluated on MARS map.