Parallel Optimized Pearson Correlation
Condition (PO-PCC) for Robust Cosmetic Makeup Facial Recognition
Kanokmon
Rujirakul and Chakchai So-In
Department of Computer Science, Faculty of Science, Khon Kaen University, Thailand
Abstract: Makeup changes or the
application of cosmetics constitute one of the challenges for the improvement
of the recognition precision of human faces because makeup
has a direct impact on facial features, such as shape, tone, and texture. Thus, this research investigates the possibility of integrating
a statistical model using Pearson Correlation (PC) to
enhance the facial recognition accuracy. PC is generally used to determine the relationship between the
training and testing images while leveraging the key
advantage of fast computing. Considering the relationship of factors other than
the features, i.e., changes in shape, size, color, or appearance, leads to a
robustness of the cosmetic images. To further improve the accuracy and reduce
the complexity of the approach, a technique using channel selection and the Optimum
Index Factor (OIF), including Histogram Equalization (HE), is also considered.
In addition, to enable real-time (online) applications, this research applies
parallelism to reduce the computational time in the pre-processing and feature
extraction stages, especially for parallel matrix manipulation, without
affecting the recognition rate. The performance improvement is confirmed by extensive
evaluations using three cosmetic datasets compared to classic facial
recognitions, namely, principal component analysis and local binary pattern (by
factors of 6.98 and 1.4, respectively), including their parallel enhancements
(i.e., by factors of 31,194.02 and 1577.88, respectively) while maintaining
high recognition precision.
Keywords: Cosmetic, facial recognition, makeup, parallel, pearson
correlation.