Deep seafloor explorations have numerous scientific applications in the fields of archeology, geology and biology et al. To obtain large-scale detailed visual information, it is needed to stitch multiple adjacent images into a photo-mosaic. Here we focus on image matching which is a key step in photo stitching application. Due to the complexity of underwater imaging environment such as light attenuation, non-uniform illumination and scattering, the images taken in deep water are degraded greatly and the performance of classical methods for image matching is reduced which seriously affect subsequent image alignment. In this paper, we try to investigate three underwater image preprocessing methods to increase the number of matched feature points. These three methods include color channel selection, lighting compensation and contrast enhancement. The experiment results show some encouraging conclusions. More feature points are matched in blue or green channel than red one. By proper inhomogenous lighting compensation and/or linear contrast enhancement, the matched points also increase significantly. When the pre-processes methods are used in combination, the number of matched features is improved 2.7 times statistically.
Software And Hardware