Composite training images for synthetic discriminant functions.

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Synthetic discriminant functions (SDFs) are crucial for pattern recognition.
• Increasing training data cardinality can lead to higher computational overhead.
• Existing methods may face limitations in incorporating diverse training information.

Purpose of the Study:

• To introduce a novel method for constructing synthetic discriminant functions using composite images.
• To reduce computational overhead in SDF construction.
• To enable the inclusion of more training information within existing computational constraints.

Main Methods:

• Utilizing composite images to represent multiple training samples.
• Constructing SDFs with increased effective training data cardinality.
• Demonstrating the procedure with the minimum-average-correlation-energy (MACE) SDF as an example.

Main Results:

• The proposed method allows for more images in SDF construction without increasing training-set cardinality.
• This leads to lower computational overhead.
• Alternatively, it allows for more training information inclusion for the same computational load.

Conclusions:

• Composite images offer an efficient approach to enhance synthetic discriminant functions.
• The method provides a practical solution for managing computational resources in pattern recognition.
• This technique can improve the performance and robustness of SDF-based systems.