Pattern recognition of multiple objects using adaptive correlation filters

A new method for reliable pattern recognition of multiple distorted objects in a cluttered background and consequent classification of the detected objects is proposed. The method is based on a bank of composite correlation filters. The filters are designed with the help of an iterative algorithm exploiting a modified version of synthetic discriminant functions. The bank consists of a minimal quantity of the filters required for a given input scene to guarantee a prespecified value of discrimination capability for pattern recognition and classification of all objects. Statistical analysis of the number of required correlations versus the recognition performance is provided and discussed. Computer simulation results obtained with the proposed method are compared with those of known techniques in terms of performance criteria for recognition and classification of objects.