Face Detection
Face detection is the process of automatically locating human faces in visual media (digital images or video). The appearance of faces is governed by a large number of factors, and can therefore vary widely from one face to the next, and one image to the next. Some of these complicating factors include variations in age, gender, ethnicity, head pose, facial expressions, lighting, image quality and compression artifacts. PittPatt’s pioneering face detection technology overcomes these barriers, and has been extensively trained and tested on real-world face imagery. It is widely acknowledged to be the leading technology in automatic face detection.
Head Pose
Our face detectors have been trained to robustly find faces regardless of head pose, ranging from frontal to full profile. As a consequence, we are also able to automatically estimate three-dimensional head pose.
Facial Variation
Facial appearance can vary widely as a function of age, gender, ethnicity and facial expressions. Our face detectors have been trained over diverse data sets to account for these variations, as well as appearance variations due to sunglasses, hats, and hair styles.
Low-Resolution Faces
Our face detectors can detect faces down to very small resolutions. How small? In our extensive training and testing, we have found that we can reliably detect faces with as few as 8 pixels between the eyes (corresponding to a face width of approximately 16 pixels), and are able to detect the majority of faces with resolutions as small as 6 pixels between the eyes.
Image Quality
We train our face detectors on non-ideal, real-world data. Consequently, our face detectors handle complex image backgrounds, grayscale as well as color images, and a large range of image quality and lighting conditions. This includes poor image quality factors such as blurring, compression artifacts and image noise. We have also implemented a variety of lighting correction methods that allow the face detector to accurately detect faces for many diverse lighting conditions, from saturated (over-exposed) to heavily shadowed faces.
