This paper proposes a novel method for object category detection in images, which is based on spatial configuration of local shape features. We show how simple, redundant edge based features overcome the problem of edge fragmentation while the efficient use of geometrically related feature pairs allows us to construct a robust object shape matcher, invariant to translation, scale and rotation. These prerequisites are used for weakly supervised learning of object models as well as object class detection. The object models employing pairwise combination of redundant shape features exhibit remarkably accurate localization of similar objects even in the presence of clutter and moderate view point changes which is further exploited for model building, object detection and recognition.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.