Fusion of global and local information for object detection

Abstract

This paper presents a framework for fusing together global and local information in images to form a powerful object detection system. We begin by describing two detection algorithms. The first algorithm uses independent component analysis to derive an image representation that captures global information in the input data. The second algorithm uses a part-based representation that relies on local properties of the data. The strengths of the two detection algorithms are then combined to form a more powerful detector The approach is evaluated on a database of real-world images containing side views of cars. The combined detector gives distinctly superior performance than each of the individual detectors, achieving a high detection accuracy of 94% on this difficult test set.