Face Detection and Synthesis Using Markov Random Field Models

Markov Random Fields (MRFs) are proposed as viable stochastic models for the spatial distribution of gray level intensities for images of human faces. These models are trained using data bases of face and non-face images. The MRF models are then used for detecting human faces in test images. The number of human face images in the training data base can be increased by simulating face-like as well as non-face like images from the trained MRFs. These simulated images are added to the existing training data bases and the corresponding MRF parameters are re-estimated. We show that the resulting face detection algorithm detects a significantly less number of false positives. We investigate the performance of the face detection algorithm for two classes of MRFs given by the first and second order neighborhood systems. Key words and phrases: Markov Random Fields, face detection, maximum pseudolikelihood estimation, simulated annealing, site permutation.

• Type: Conference
• Year: 2002
• Where: ICPR
• Authors: Sarat C. Dass, Anil K. Jain, Xiaoguang Lu
• Keywords: Computer Vision, Face Detection Algorithm, Human Face Images, ICPR 2002, Markov Random Fields