This article presents a tractable and empirically accurate algorithm realizing a midlevel visual process for pattern recognition. The algorithm takes advantage of hypotheses provided by a highlevel visual process, thereby, attempting to extract a region in an image based on these hypotheses. The main focus is to recognize quadrilateral as well as arbitrarily shaped objects from synthetic and real-world images. The novel approach is based on a study of the Hough Transform and its generalized version. To show overall usefulness of the algorithm, an extensive series of experiments was performed. In particular, occlusion and multiple object-instances were tested, indicating the effectiveness of this work's approach.