SVM-based triaging and color transfer for underwater mosaic enhancement

Abstract

Underwater videos taken by auto white balancing cameras sometimes appear to have bluish or greenish tints which can be caused by physical, chemical, and instrument-specific factors. We present an approach for effectively classifying the video frames that can be color-corrected by employing a triaging step using Support Vector Machines (SVM). The SVM-based triaging successfully classified the mosaics into three categories: off-balanced blue, off-balanced green, and white-balanced. This is followed by a color transfer method, where histograms of the mosaics with blue-green tint were matched with the histograms of the white-balanced mosaics. This method effectively removed the tint resulting to enhanced mosaics.