TITLE
PERFORMANCE ANALYSIS OF HAAR CASCADE-BASED FACE DETECTION IN MULTI-FACE IMAGES UNDER DIGITALLY SIMULATED LIGHTING CONDITIONS
AUTHOR(S)
Ivan Sarkocevic1, Vladimir Maksimovic2, Branimir Jaksic2*, Petar Spalevic2, Bojan Prlincevic1
ABSTRACT
In this study, the performance of the Haar Cascade Classifier was evaluated in images containing varying numbers of faces captured from both frontal and non-frontal perspectives. The images extracted from the Face Detection Dataset (FDD) were digitally manipulated to achieve different lighting conditions by adjusting brightness, contrast, and gamma parameter values using a Python program. The Haar Cascade Classifier, implemented in a separate Python program, was used to carry out the face detection process. Two objective metrics were employed to estimate detection accuracy: the F-measure, derived from GroundTruth data of the reference images, and Det.F, representing the total number of detected faces. The findings indicate that variations in brightness, contrast, and gamma parameters slightly affect detection outcomes, while changes in the angle of perspective have a far more significant impact. The Haar Cascade Classifier achieved the best results in images showing frontal faces, especially when there is a small number of faces represented in an image, regardless of the lighting conditions; images containing a large number of non-frontal faces confirmed the classifier’s limited robustness under such complex conditions.
DOI
How to cite this article:
Ivan Sarkocevic1, Vladimir Maksimovic2, Branimir Jaksic2*, Petar Spalevic2, Bojan Prlincevic1, PERFORMANCE ANALYSIS OF HAAR CASCADE-BASED FACE DETECTION IN MULTI-FACE IMAGES UNDER DIGITALLY SIMULATED LIGHTING CONDITIONS, UNITECH – SELECTED PAPERS - 2025