Design of Correlation Filters for Pattern Recognition Using a Noisy Training Image

Abstract

Correlation filters for object detection and location estimation are commonly designed assuming the shape and graylevel structure of the object of interest are explicitly available. In this work we propose the design of correlation filters when the appearance of the target is given in a single training image. The target is assumed to be embedded in a cluttered background and the image is assumed to be corrupted by additive sensor noise. The designed filters are used to detect the target in an input scene modeled by the nonoverlapping signal model. An optimal correlation filter, with respect to the peak-to-output energy ratio criterion, is proposed for object detection and location estimation. We also present estimation techniques for the required parameters. Computer simulation results obtained with the proposed filters are presented and compared with those of common correlation filters.