Edge-based synthetic discriminant function for distortion invariant object recognition

A new edge correlation based matching method has been presented as an alternative to the typical implementation of intensity correlation based matching. The synthetic discriminant function (SDF) is well known correlation filter for distortion invariant object recognition, and the edge-based SDF (ESDF) presented here utilizes the ´structured information´ rather than the fragile color or intensity constancy assumption. To satisfy the constraint of the same correlation peak value, an iterative method is proposed to adjust the coefficients of training images. We apply this ESDF to stamp sheet automatic segmentation and stamp perforations quality monitoring. The ESDF approach clearly outperformed in the extensive comparative experiments, showing its capabilities to perform distortion invariant object recognition.