Fractal Texture Enhancement of Simulated Infrared Images Using a CNN-Based Neural Style Transfer Algorithm with a...

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Simulated infrared images often lack realistic texture and fractal characteristics.
• Enhancing synthetic infrared data is crucial for accurate analysis and applications.

Purpose of the Study:

• To develop a CNN-based method for transforming simulated infrared images into realistic ones.
• To improve the fractal characteristics and texture of synthetic infrared images.

Main Methods:

• Utilizing pretrained VGG-19 networks for infrared feature extraction.
• Applying neural style transfer to imbue simulated images with real infrared fractal features.
• Conducting fractal analysis and evaluating image quality using PSNR, SSIM, and NIQE.

Main Results:

• The proposed algorithm successfully enhanced simulated infrared images.
• Enhanced images demonstrated improved NIQE and SSIM scores, indicating closer similarity to real infrared images.
• The method effectively transferred brightness and fractal characteristics.

Conclusions:

• The CNN-based approach effectively enhances simulated infrared images by incorporating real infrared texture and fractal features.
• The technique offers a valuable tool for improving synthetic infrared and general synthetic images.
• This method significantly advances the realism of simulated infrared imagery.