Exploiting 3D information for an automated semantic analysis of underwater images

自动图像分析是一种常见的方法，但在水下图像中应用很少。这种特殊的环境对模式识别和机器学习方法提出了挑战。水下成像容易出现衰减、浑浊和光照变化等问题。总之，几项研究表明，在利用2D信息对水下图像进行语义解释方面取得了成功。然而，存在仅有2D信息不足以从图像中提取知识的情况。在这些情况下，3D信息可能会有所帮助。在此，应分析现有的3D数据集。它将得到在受控条件下获得的3D图像的补充。附加的形态信息应结合到自动2D模式识别方法中。将与海洋生物学家进行联合研究。自动从图像中提取的语义数据将用于更好地理解图像中的栖息地。

Automated image analysis is a common methodology, but has seen little application for underwater imagery. Pattern recognition and machine learning methods are challenged by the peculiar environment. Imaging under water is prone to attenuation, turbidity and varying illumination. Anyhow, several studies have shown success in exploiting 2D information for a semantic interpretation of underwater images. There exist, however, cases where 2D information alone is not sufficient to extract knowledge from the images. In those cases 3D, information could help.Here, an existing 3D data set shall be analysed. It will be complimented by 3D imagery acquired under controlled conditions. Additional morphological information shall be incorporated into automated 2D pattern recognition methods. Joint research with marine biologists will be conducted. Semantic data, automatically extracted from the imagery, will be used for a better spatial understanding of the imaged habitats.