A technique for shape recognition that is invariant to scale and rotation is presented. This technique employs the number of bit quads, the basic 2 x 2 element of binary (0,1) imagery, of each object. The feature vector is a scaled version of the number of bit quads, which allows a distance to be defined between unknown objects and a collection of known prototypes. Recognition is accomplished by utilizing this distance metric as a classifier. An example is provided that recognizes an automobile shape from a set of six prototypes. Several experiments are performed that change the scale and relative rotation of the unknown. In all cases the correct automobile is identified from the set of six prototypes. A second example considers the effects of boundary noise on classification and points out the advantage of employing noise smoothing prior to feature extraction. The technique presented has the advantage of simplicity, pipeline implementation, and low storage requirements.