Inclusion of filter modulation in synthetic-discriminant-function construction.

Area of Science:

• Optical Engineering
• Image Processing
• Pattern Recognition

Background:

• Matched spatial filters (MSF) are crucial for pattern recognition and object detection in optical systems.
• Conventional synthetic-discriminant-function (SDF) filters, including phase-only filters (POFs) and binary-phase-only filters (BPOFs), face limitations in achieving precise correlation responses.
• The synthesis of SDF MSFs typically involves projection methods that can struggle with accuracy and signal-to-clutter ratios.

Purpose of the Study:

• To present a novel technique for synthesizing SDF MSFs that incorporates filter modulation directly into the synthesis process.
• To evaluate the performance of the new method for phase-only filters (POFs) and binary-phase-only filters (BPOFs) using computer simulations.
• To demonstrate the capability of the developed filters in achieving high accuracy and signal-to-clutter ratios for various classification problems.

Main Methods:

• A new filter synthesis technique is introduced, integrating filter modulation into the SDF MSF design.
• The filter synthesis involves solving a system of simultaneous nonlinear equations using an iterative procedure.
• Computer simulations were conducted using thresholded images of the Space Shuttle across various aspect angles to test POFs and BPOFs.

Main Results:

• The synthesized filters successfully achieved the specified peak-correlation response with an accuracy of within 1%.
• A high signal-to-clutter ratio was obtained for one-class, two-class, and multilevel classification problems.
• Conventional projection-based SDF POFs and BPOFs failed to produce the desired peak-correlation response, highlighting the superiority of the new method.

Conclusions:

• The proposed technique for SDF MSF synthesis, including filter modulation, offers superior performance compared to conventional methods.
• This approach enables the construction of filters capable of highly accurate correlation and robust discrimination in complex scenarios.
• The method is effective for various pattern recognition tasks, including those involving challenging datasets like thresholded images at different aspect angles.