Sub-population prediction using enhanced correlation filters
Minimum average correlation energy (MACE) filters are initially developed and widely used for image pattern recognition tasks. A novel method leveraging the enhanced MACE filters is proposed to tackle classification problems from a new perspective. By employing 1D Fourier transform, establishing new identification metric, and improving numerical stability, the proposed method constructs an enhanced correlation filter to select a sub-population of the un-labelled data, and subsequently outputs labels. Experiments show our method achieves 100% precision on multiple datasets considered, including two public benchmarks and one obtained from semiconductor industry addressing an emerging task.