A comparative analysis between three local descriptors
Shekhar Karanwal · 2023
Literature reports comparative analysis between numerous descriptors (local) in form of their research article. These studies although achieve good results but the best combination of local, global and classification methods are missing in the earlier research. As a result, the performance of these are not as encouraging as it be in the harsh light variations. This work eliminate this demerit and executes the best combination of local, global and classification methods in Face Recognition (FR). Precisely, this work presents the comparative analysis of three local descriptors. These three local descriptors are 6x6 MB-LBP, MB-ZZLBP and MRELBP-NI. All three captures the macrostructure and microstructure information therefore their comparison is performed. The idea is to check which descriptor produces better results in harsh illumination variations. For all three descriptors global histograms are extracted, which develops histogram size of 256. For compression and matching FLDA and SVMs are utilized. It is MRELBP-NI which attains better results than other two, as result suggests. The best accuracy achieved by MRELBP-NI is 93.49% and 99.33% on EYB and YB datasets. All testing is conducted in MATLAB R2021a settings.