Browsing by Author "Chang, Tien-Lung"

Now showing 1 - 2 of 2

BETTER FACE DETECTION WITH VANISHING POINT-BASED IMAGE RECTIFICATION
(2013)
Chang, Tien-Lung
;
Wang, Ching-Ho
;
Chuang, Jen-Hui
;
資訊工程學系
;
Department of Computer Science
In this paper we propose a novel face detection method based on the vanishing point of vertical lines in the scene to improve system performance in a common surveillance application. While most existing face datasets and detection techniques are based on the assumption that the camera has a similar height as the target faces, in practical situations the camera may be installed at different heights. Such discrepancy often degrades the detection performance of algorithms based on learning with certain (e. g., frontal) face orientation. In this paper we propose a transformation to rectify face images (video frames) such that it is not necessary to collect training data of different face orientations. Furthermore, with the proposed method there is no need to perform complex camera calibration. The only required information is the vanishing point of vertical lines, which can often be estimated easily. Experiments show prominent improvements in face detection performance can be obtained with the proposed image transformation.
Improving local learning for object categorization by exploring the effects of ranking
(2008)
Chang, Tien-Lung
;
Liu, Tyng-Luh
;
Chuang, Jen-Hui
;
資訊工程學系
;
Department of Computer Science
Local learning for classification is useful in dealing with various vision problems. One key factor for such approaches to be effective is to find good neighbors for the learning procedure. In this work we describe a novel method to rank neighbors by learning a local distance function, and meanwhile to derive the local distance function by focusing on the high-ranked neighbors. The two aspects of considerations can be elegantly coupled through a well-defined objective function, motivated by a supervised ranking method called P-Norm Push. While the local distance functions are learned independently, they can be reshaped altogether so that their values can be directly compared. We apply the proposed method to the Caltech-101 dataset, and demonstrate the use of proper neighbors can improve the performance of classification techniques based on nearest-neighbor selection.