An Approach in Auto Valuing for Optimal Threshold of Viola Jones

Abstract

The biometric features for people identification on the face include eyes, nose, and mouth. In this study, the three features are detected by using Viola Jones method. The affecting parameter on the accuracy of face feature detection by using Viola Jones is the threshold value because it is the input criterion for Mergethreshold parameter on Matlab. The appropriate threshold value will decrease the detection error so that yielding appropriate ROI. Therefore, this study employs a process of finding out the automatic threshold value by using Viola Jones feature detection. The input data of the system is in the form of frame video from 4 respondents which yield totally 938 face frames as the result of automatic detection that is extracted with matrix 130x110. The result of the testing yields accuracy as much as 79.17% for eyes feature, 94.81% for nose feature, and 99.49 for mouth feature. The whole features yield approximate accuracy as much as 91.15% while the testing result of the time needed by 4 respondents with total 20 video frames is 0.368 seconds for eyes feature detection, 1.065 seconds for nose feature, and 4.665 seconds for mouth feature.

Access to Document

Other files and links

• Link to publication in Scopus
• Open Access Version Available

Fingerprint