Using cross-examples in viola-jones algorithm for thermal face detection

Detection of the face region is a key step in a face recognition system. Thermal images are widely used in many applications where normal visibility is reduced, impaired or ineffective, such as night surveillance and fugitive searches. However, low spatial resolution brings challenges to face detection in thermal images. Viola-Jones is an object detection method widely used for face detection. The algorithm suffers from missed faces and wrongly detected non-face objects due to low resolution of thermal images. To improve the face detection performance for thermal images, we propose to incorporate cross-examples into our framework. In addition to using negative samples of non-face thermal images, we utilize non-face visible images as part of the negative samples (cross-examples). Cross-examples effectively increase the discriminability between the positive samples and negative samples. Experimental results show that the proposed scheme can effectively reduce the non-face objects and thus improve accuracy of face detection.