In this work, we present a face detector that establishes the new state-of-theart
in terms of both accuracy and speed. It follows the “boosted cascade structure
+ simple features” principles. We use simple pixel differences as feature which
bring advantages on the efficiency. Our detector takes only 28.6 milliseconds
for a VGA image, more than 1000 times faster than [32]. It also achieves the
best accuracy on the challenging datasets [32, 9, 22], significantly outperforms all
existing academia solutions including [32, 24], and is on bar with the commercial
system in Google Picasa. Figure 1 shows our example detection results under
large viewpoints, occlusion and poor lighting
In this work, we present a face detector that establishes the new state-of-theartin terms of both accuracy and speed. It follows the “boosted cascade structure+ simple features” principles. We use simple pixel differences as feature whichbring advantages on the efficiency. Our detector takes only 28.6 millisecondsfor a VGA image, more than 1000 times faster than [32]. It also achieves thebest accuracy on the challenging datasets [32, 9, 22], significantly outperforms allexisting academia solutions including [32, 24], and is on bar with the commercialsystem in Google Picasa. Figure 1 shows our example detection results underlarge viewpoints, occlusion and poor lighting
การแปล กรุณารอสักครู่..
