Feature-based affine-invariant localization of faces

Abstract

We present a novel method for localizing faces in person identification scenarios. Such scenarios involve high resolution images of frontal faces. The proposed algorithm does not require color, copes well in cluttered backgrounds, and accurately localizes faces including eye centers. An extensive analysis and a performance evaluation on the XM2VTS database and on the realistic BioID and BANCA face databases is presented. We show that the algorithm has precision superior to reference methods.

Precise eye localization through a general-to-specific model definition

“…In particular the baseline methods suffer from a drop of the face recognition rate by about the 20% even for small eye localization errors (d eye = 0.05). In such context 0.05 should be considered as the maximum acceptable error, as noted by Hamouz et al [15]. Recently a similar result appeared in the literature [25], where an even more drastic drop of performance is presented (50%, at d eye = 0.05), obtained on the FRGC database and running a different implementation of the PCA technique.…”

“…The ROC curves tell the rate of the automatic eye localization for each level of precision d eye as measured with respect to the ground truth of the eye centers. The graphs compare the performance of our eye detector (SVM-1) against our eye localizer (SVM-2) and, where available, we report the performance of the method [15] denoted as '1 face on the output'. Adopting d eye ≤ 0.25 as the criterion for correct eye detection, we observe that the SVM-1 alone permits to achieve rates of 99.0%, 96.1%, 96.4%, 97.1% and 97.8% over the datasets BANCA, BioID, FERET, FRGC and XM2VTS respectively.…”

“…Indeed, some researchers have faced the problem: the first were Jesorsky et al [16], Smeraldi et al [23] and Martínez [20], and recently there has been a new impulse in this very relevant research field [7,13,15,19,25], while leaving room for further improvements.…”

Webcam-Based Visual Gaze Estimation

“…Furthermore, daylight applications are precluded due to the common use of active infrared (IR) illumination, used to obtain accurate eye location through corneal reflection. Non infrared appearance based eye locators [4,5,6,7,8,9,10,11] can successfully locate eye regions, yet are unable to track eye movements accurately.…”

Real-Time Face Detection Using Illumination Invariant Features

“…When attempting to detect faces (or locate a single face) in a visual representation, image-based and landmark-based methods may be primarily distinguished between [1,2]. This paper focuses on the detection of frontal faces in 2D images and is assigned to the former category.…”