Generality of matched filtering and minimum Euclidean distance projection for optical pattern recognition

Area of Science:

• Optics and Photonics
• Pattern Recognition
• Statistical Signal Processing

Background:

• Traditional optical correlation pattern recognition optimizes signal-to-noise ratio (SNR) for single training images.
• Existing methods face challenges with multiple training images, complex noise, and filter design constraints.

Purpose of the Study:

• To extend matched filtering with minimum Euclidean distance projection for advanced optical pattern recognition.
• To address statistical metrics, noise, filter construction, and observable outputs in complex scenarios.

Main Methods:

• The algorithm integrates standard statistical pattern recognition metrics with multiple training images.
• It accounts for additive input noise with known power spectral density and irreducible detection noise.
• Filter construction utilizes arbitrary subsets of the complex unit disk, relying solely on observable correlator outputs.

Main Results:

• The enhanced algorithm optimizes various statistical criteria including Fisher ratio, Bayes error, Chernoff and Bhattacharyya bounds.
• It also improves population entropy, expected information, and generalized SNR metrics.
• Performance is evaluated using the area under the receiver operating characteristic curve.

Conclusions:

• The generalized algorithm effectively handles multiple training images and complex noise environments in optical pattern recognition.
• Different statistical criteria can be optimized using specific complex scalar weights, offering flexibility in filter design.
• This work advances the robustness and applicability of optical correlation pattern recognition systems.