Local Patterns and Big Time Series Data for Facial Poses Classification
Hayet Mekami · Journal of Computers · 2018
The problem of identifying and analyzing faces in images is a fundamental task in computer vision.Though great progress has been achieved in face detection, it is still difficult to obtain the pose estimation.In this paper we propose a pose estimation approach that is based on time series representation.We have converted input images of faces into big time series datasets, and we then used a dimensionality reduction method to convert the original series to a symbolic representation.Classification algorithms are then applied using the distances between the symbolic sequences of time series.Since external conditions when capturing images are not always optimal, pose estimation can become a challenge.In order to overcome such problems, we propose to use the gradient image and the Local Binary Pattern (LBP) combined with dynamic morphological quotient image (DMQI-LBP), where these descriptors are robust to changes in illumination.Classification algorithms such as K-means, SVM and KNN were evaluated to classify frontal vs profile faces poses, and the obtained experimental results show that the proposed method is very efficient.