Applications of neural networks in human shape visual perception

Area of Science:

• Optics and Photonics
• Computer Vision
• Human Visual Perception

Background:

• Optical and electronic advancements reduce image noise but do not fully overcome human visual limitations with low-resolution thermal images.
• Psychological, physiological factors, and monitor variations hinder accurate identification of thermal imagery.
• Accurate recognition of long-range targets is crucial for military applications, especially with degraded image quality.

Purpose of the Study:

• To develop an effective method for enhancing the recognition of low-resolution thermal images.
• To assist the military in refining shape parameters for long-range target identification.
• To improve the accuracy of human visual recognition in challenging thermal imaging scenarios.

Main Methods:

• Extended human visual model characteristics using the modulation transfer function.
• Incorporated novel shape parameters into the human visual model.
• Integrated the enhanced human visual model with a backpropagation neural network for image recognition.

Main Results:

• Experimental validation demonstrated the efficacy of the new recognition model.
• The developed system achieved a consistent recognition accuracy rate exceeding 95%.
• The method effectively addresses limitations in identifying low-resolution thermal images.

Conclusions:

• The novel human visual model significantly enhances the recognition of low-resolution thermal images.
• The backpropagation neural network integration provides a robust solution for military target identification.
• This approach offers a promising tool for improving visual recognition in degraded imaging conditions.