Effects of Image Resolution on Automatic Face Detection
Ossama Haneefl, Sana Maqbooll, Farhana Siddiquel, Zahid Mahmood, Shahid Khattakl, Gul Zameen Khan · 2019
This study presents comparison of four well-known face detection algorithms. Face detection algorithms investigated in this study are, (i) Viola Jones (VJ) face detection algorithm, (ii) Normalized pixel difference (NPD), (iii) Histogram of Oriented Gradients (HOG), and (iv) Skin Color Based (SCB) face detection algorithm. Simulation results reveal that on the LFW database, and for image resolution of 40×40 pixels and below, only Skin Color-Based face detection algorithm was able to locate faces with a mean accuracy of 56% at the cost of highest computation complexity. Whereas, for image resolution of 60×60 pixels and above, the Viola Jones algorithm has the highest detection accuracy of 61% with average execution time of 0.605 seconds. On real life images that contained up to 64 faces in the input image, the Viola Jones algorithm surpassed the companion algorithms followed by the NPD and the HOG. The SCB algorithm is found to be least effective on real life images.