Obscurant Segmentation in Long Wave Infrared Images Using GLCM Textures

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Autonomous image segmentation offers benefits across diverse fields like medical diagnosis and autonomous vehicles.
• Current research predominantly focuses on visible light modalities, leaving a gap in infrared scene segmentation.

Purpose of the Study:

• To identify and improve semantic methods for infrared scene segmentation.
• To propose a novel tile-based segmentation approach for occlusion clouds in Long Wave Infrared (LWIR) images.

Main Methods:

• Utilized Gray Level Co-Occurrence Matrix (GLCM) features extracted from local image tiles.
• Trained a binary classifier using GLCM features to identify occluded tiles.
• Introduced a new texture analysis scheme for improved image segmentation.

Main Results:

• The proposed GLCM-based method demonstrated strong performance in infrared image segmentation.
• Achieved higher accuracy and better inter-region homogeneity compared to traditional pixel-based methods.
• Successfully enabled conditional processing by distinguishing between dust cloud and clear tiles.

Conclusions:

• The novel GLCM approach offers a more suitable texture analysis for infrared image segmentation.
• This method effectively segments occlusion clouds in LWIR images, complementing existing semantic segmentation algorithms.
• The findings contribute to advancing machine learning applications in thermal imaging.