Most of the current object recognition systems involve matching the scene object against every predefined object in the model database one at a time. The major problem with this approach is that as the number of models is increased the computational complexity and time requirements of the systems are greatly increased. In this paper, we present a linear supervised classifer that uses a weight vector as a filter to modify the flow of the input information. The optimal weight vector is derived from a set of training samples in the sense of minimum least-sqares error between desired and actual output responses. The computational complexity of the weight vector is independent of the number of models. Assignment of a scene object to one of several classes of objects is directly determined by the weight vector and the input image of the object.Experimental results on both man-made workparts with simple shapes and natural objects with complex shapes are studied in this paper. The occluded and distorted versions of these test objects are also included to evaluate the performance of the linear weight classifier.