Skip to Main Content
This paper proposes a new method to detect objects in images. Boundaries contain shape of the objects. To detect objects in cluttered images, we use boundary fragments. Boundary fragments are obtained by our new training procedure. Poisson equation is used to divide edges and extract generic model of the object. Gaussian Mixture Model (GMM) is used to model shape of the object. This creates relation between local boundary fragments. To localize our desired object in image, a voting process is performed in Hough space. A scale adaptive search algorithm is used in this voting space to find the most probable regions of object presence. By learned GMM, shape of the object is obtained from test image. We remove false positives, usual problems in object detection tasks, by matching generic model to the extracted shapes. Local orientations histograms of the objects model are used for matching. We evaluate our method and compare with state-of-the-art methods. Experimental results show the power of our proposed method in detection and its robustness in face to scale and translation variations.