Colour correction of underwater images

Abstract

Objects look very different in the underwater environment compared to their appearance in sunlight. High quality images with correct colouring simplify the detection of underwater objects and may allow the use of visual SLAM algorithms developed for land-based robots underwater. Hence, image processing is required to obtain images of high quality and correct colouring. Current algorithms focus on the colour reconstruction of scenery at diving depth which has the advantage that a significant part of sunlight is still present and different colours can still be distinguished. At greater depth the filtering is much stronger such that this is no longer possible. In this study it is investigated whether machine learning can be used to transform image data. In order to obtain images under underwater lighting conditions in a controlled environment a special light source with a defined wavelength is used for illumination of test objects in a laboratory setup. The images are then fed through statistical learning algorithms with or without pre-filters. It is shown that k-nearest neighbour and support vector machines are most suitable for the given task and yield excellent results. Most difficult are the distinction between red and black / dark grey objects as well as coping with reflections.