Abstract:
From security systems to human-computer interaction, face recognition technology is a key component in
many different applications. In this domain, the Local Binary Pattern Histogram (LBPH) algorithm has
shown great promise due to its ability in texture-based feature extraction and robustness.
In this study, the performance of LBPH algorithm is improved by optimising important parameters: radius,
number of neighbours and grid configuration. Tweaking these parameters systematically enable us to get
the most success out of this algorithm. LBPH algorithm parameters has been tuned into a 5x7 dimensional
matrix, which gives us total of 35 grids with equal width and height pixels. In other words, one central
pixel plus 34 neighbouring pixels where the radii of 3 square neighbourhoods can be adjusted.
The study combines the experimental exploration with fine-tuning via machine learning approaches to
optimize LBPH algorithm. Finally, we have provided experimental results on intra-class and inter-class
feature distribution analysis conducted on selected images taken under constraints and unconstraint
environments. The performance of our novel 34N-LBPH algorithm showed very low values by obtaining
intra-class average Means of 0.031, 0.078, and 0.101 for three classes under constraint environment
indicating the proposed 34N-LBPH is robust for facial feature extraction. The findings suggest that our
enhanced algorithm, 34 Neighbour Linear Binary Pattern Histogram (34N-LBPH) can effectively handle
variations in lighting, expressions and occlusions, contributing to the advancement of facial feature
extraction for face recognition.
