bject recognition is a key problem in machine vision.The issue to be addressed related to the representation of the objects and the method used to establish correspondence. In this thesis, a new technique for three-dimensional (3-D) polyhedral objects recognition on the basis of a single two-dimensional (2D) view of 3-D scene, are proposed. The binary gradient image of the scene is converted into Hough domain using a modified form of Hough Transform method-for straight line detection, which is proposed by this thesis to cope with Hough problems, through performing pixel contiguity check and lines' best fitting algorithms. The cluster patterns (object 's descriptions) originating from straight line features of the image are explored by reasoning in Hough space. This yield symbolic object descriptions, which is compared with objects model's descriptions. For the purpose of matching, models wire-frame are constructed through system's object editor. These models are represented by a set of similar descriptions of the object to be recognized, from a multiple viewing angles. Matching between the ofcgect and models is performed through an efficient search strategy, using two matching stages, where the first stage selects the most likely candidates models for unknown object, and the next stage identified unknown object, through perform a detailed comparison. The elegance in our introduce techniques lies in to cope with noisy images.