Quadratic correlation filter design methodology for target detection and surveillance applications

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Conventional linear correlation filters (LCFs) are widely used for target detection but have limitations.
• Feature extraction or segmentation is often required for traditional detection methods.
• Existing methods may struggle with complex imagery or require extensive postprocessing.

Purpose of the Study:

• To introduce a novel optimization method for Quadratic Correlation Filters (QCFs).
• To enhance shift-invariant target detection capabilities, particularly in infrared imagery.
• To improve performance and simplify processing compared to LCFs.

Main Methods:

• Quadratic classifiers (QCFs) operating directly on image data without feature extraction.
• Joint optimization of QCF performance using multiple parallel correlators.
• Formulation of a class-separation metric as a maximized Rayleigh quotient.

Main Results:

• QCFs demonstrate superior performance over linear counterparts for target detection.
• The proposed QCF optimization method significantly outperforms previous approaches.
• Simplified postprocessing achieved through combined outputs from multiple correlators.
• Independent evaluations confirm the effectiveness of the QCF method.

Conclusions:

• The novel QCF optimization method offers significant advantages for shift-invariant target detection.
• QCFs provide a powerful alternative to LCFs, especially for infrared surveillance.
• The approach yields improved accuracy and efficiency in image analysis tasks.