Abstarct: Objective in this thesis is to develop a method for establishing an object recognition system baxsed on the matching of image primitives. A primitive is extracted from image by the Scale Invariant Feature Transform(SIFT) descxriptors. This method can be applied to a number of computer vision applications such as object recognition (in general) and image retrieval.The motivation for using descxriptors as image primitives is that they can be invariantly to a group of affine transforms and stable under scaling and rotation. Each object in databaxse modeledby a single frontal image. The recognition task is to detemine the peresence of object(s) of interest in scene images. Attributed Relational Graph(ARG) proposed for represenation image primitives. Each node in ARGdescribed bytwo properties: Unary and Binary measuments. The unary measuments use to describe any node in graph individually. In order to describe relation with two nodes binary measuments used. The probalistic realxtation labelingapplied for graph matching. Exprimental result devided into two stage: Virtual Environment(VE) and real senario.Since in the real environment is difficult to control control system parameters, Virtual Environment(VE) constructed. Then structural nosie added in this VE for evaluating two algorithm baxsed on probabilisic relaxation labeling: Ahmadyfard and Kostin matching. Later two algorithm in real senario tested. The first alghorithm was reliable than second algothim. Against the second find the more match than first algorithm. Finally A new hybrid algorithm baxsed on two algorithm proposed and applied to a real senarion. The prposed algorithm performs from both reliability and recognition rates point to view.