Optimizing Submarine Photo Mosaicking by Filtering Redundant Images*

This paper presents an innovative methodology to filter out redundant images initially selected to form seabed photo mosaics. Underwater robot missions often capture hundreds or thousands of images, especially in large or long-duration surveys. These images typically contain overlapping areas that must be carefully removed to prevent redundant information, questionable pixel aggregations, unrealistic color blending, and excessive computational load during mosaic construction. Experiments using datasets collected by an Autonomous Underwater Vehicle in ecologically significant marine habitats demonstrate substantial reductions in visual data without losing essential information, significantly optimizing computational efforts and reducing considerably processing times.