Abstract
PERFORMANCE EVALUATION OF APPEARANCE AND GEOMETRY BASED ALGORITHMS FOR FACIAL FEATURE EXTRACTIONS
Ganiyu Rafiu Adesina, Adeyemo Isiaka Akinkunmi, Olabiyisi Stephen Olatunde, Akande Noah Oluwatobi
ABSTRACT
Facial feature extraction is an important step towards the development of any facial recognition system. Many algorithms have been used to improve feature extraction but little research work has been done to evaluate their performances in order to determine the most efficient. This paper focuses on the performance evaluation of appearance and geometry based facial features extraction techniques using Gabor Wavelet (GW) and Local Binary Patterns (LBP) algorithms, respectively. The GW and LBP algorithms were implemented using MATLAB 8.1 (R2013a) with Euclidean distance as similarity metric between each testing and training. Each of the implemented algorithms was used to extract facial features. The GW algorithm was used to extract facial features such as projection axes, edge, valley and ridge contours while features such as histograms, pixel values, lines, edges and corners were extracted using the LBP algorithm. The results of evaluation showed that GW feature extraction technique assumed an average Computational Time of 3.92seconds while LBP yielded an average Computational Time of 6.0seconds. Also, GW yielded 56.2% False Acceptance Rate while 82.5% False Acceptance Rate was recorded for LBP. Similarly, the False Rejection Rate of GW was 88.7% while that of LBP was 48.5%. The results obtained in this study implied that GW facial features extraction technique is more efficient than LBP in terms of Computational Time, False Acceptance Rate and False Rejection Rate. Also, it could serve as a guide for researchers to choose appropriate algorithm for facial features extraction.
[Full Text Article] [Download Certificate]