Spatial-domain implementation of optimal multicriteria correlation filters

Area of Science:

• Optics and Photonics
• Image Processing
• Computer Vision

Background:

• Correlation filters are crucial for pattern recognition in optical systems.
• Frequency domain methods for defining filter support regions are computationally intensive.
• Existing methods face limitations with certain spatial light modulators.

Purpose of the Study:

• To introduce a novel spatial domain approach for approximating optimal regions of support for correlation filters.
• To develop efficient methods for generating these regions for optical correlators.
• To demonstrate the performance and advantages of the proposed convolution-based method.

Main Methods:

• Approximating frequency domain regions of support using small spatial domain convolution kernels.
• Developing an optimal and a fast nonoptimal approach for generating these kernels.
• Implementing the convolution kernels on low-cost arithmetic frame grabbers.

Main Results:

• The convolution-based approach yields performance comparable to optimal frequency-domain methods.
• The resulting input images for the correlator are always positive-valued due to low-pass filtering characteristics.
• The method is compatible with spatial light modulators that cannot produce a zero state.

Conclusions:

• Spatial domain convolution offers an efficient and effective alternative for defining correlation filter regions of support.
• This approach facilitates real-time, low-cost implementation in optical correlators.
• The method enhances the applicability of correlation filters in various optical processing systems.