Perceptual Distance Normalization for Appearance Detection

In this paper we develop a novel contrast-invariant appearance detection model. The goal is to classify object-specific images (e.g. face images) from generic background patches. The novel contribution of this paper is the design of a perceptual distortion measure for comparing the appearance of an object to its reconstruction from the principal subspace. We demonstrate our approach on two different datasets: separating eyes from non-eyes and classifying faces from nonfaces. On the eye database, for a true detection rate of 95% we demonstrate a nine-fold improvement in the false positive rates over a previously reported detection model [5]. We also compare our detector model with a SVM classifier.

• Type: Conference
• Year: 2004
• Where: ICPR
• Authors: Allan D. Jepson, Chakra Chennubhotla
• Keywords: Appearance Detection Model, Computer Vision, ICPR 2004, Perceptual Distortion Measure, True Detection Rate