Biologically inspired template matching using scene context

Template matching is widely used in pattern recognition and computer vision. However, the performance of traditional template matching approaches is often sensitive to large intraclass variance, occlusion, minor variety of poses, low-resolution conditions, background clutter etc. To resolve the problem, this paper presents a scene context based template matching method, which is inspired from two biological plausible ways, scene-centered viewpoint and change blindness phenomenon. The proposed method is quite different form conventional template matching methods, since this technique measures the similarity of a template and a sub-image by putting both into scene context, in a semantic representation. This work focuses on object detection task using only one object instance without training. Extensive experiments on three datasets and a video clearly show the superior robustness of the proposed method.