The purpose of this thesis is to develop a computer vision system which recognizes objects from real images. The system accepts the real image of a simple world consisting of blocks, pyramids, and cylinders. The recognition of objects is done by using the concept of region descriptions. First, significant features such as lines and vertices are extracted from the input image. Then primitive regions are determined from the feature lists and predictions of overlapped regions are made. The region description is obtained by using region adjacency graphs(RAG) and T-junctions. Using the structural relations, objects are finally recognized. The identity and location of the object are determined by using the geometric measurements.