Glare Detection Algorithm in Underwater Images via Pixel Clustering Using Image Color Space Fusion

Robots designed to assist and ensure safety during underwater diving activities utilize underwater cameras to gather real-time diver information. This study proposes a breath bubble detection algorithm based on unsupervised k-means clustering to overcome the high accuracy demand of deep learning models and the challenge of constructing training datasets. By integrating color and relative coordinate information from underwater images and applying CLAHE to reduce noise, pixel clustering is performed to extract reflective regions. Experimental results demonstrate the effectiveness of the proposed algorithm in detecting breath bubble regions in underwater images. Improved detection accuracy is achieved through the combination of RGB, LAB, and HSV color spaces. This research provides a foundation for the monitoring of diver status and equipment malfunctions in underwater environments.