Deep-sea image processing

High-resolution seafloor mapping often requires optical methods of sensing, to confirm interpretations made from sonar data. Optical digital imagery of seafloor sites can now provide very high resolution and also provides additional cues, such as color information for sediments, biota and divers rock types. During the cruise AT11-7 of the Woods Hole Oceanographic Institution (WHOI) vessel R/V Atlantis (February 2004, East Pacific Rise) visual imagery was acquired from three sources: (1) a digital still down-looking camera mounted on the submersible Alvin, (2) observer-operated 1-and 3-chip video cameras with tilt and pan capabilities mounted on the front of Alvin, and (3) a digital still camera on the WHOI TowCam (Fornari, 2003). Imagery from the first source collected on a previous cruise (AT7-13) to the Galapagos Rift at 86°W was successfully processed and mosaicked post-cruise, resulting in a single image covering area of about 2000 sq.m, with the resolution of 3 mm per pixel (Rzhanov et al., 2003). This paper addresses the issues of the optimal acquisition of visual imagery in deep-sea conditions, and requirements for on-board processing. Shipboard processing of digital imagery allows for reviewing collected imagery immediately after the dive, evaluating its importance and optimizing acquisition parameters, and augmenting acquisition of data over specific sites on subsequent dives. Images from the deepsea power and light (DSPL) digital camera offer the best resolution (3.3 Mega pixels) and are taken at an interval of 10 seconds (determined by the strobe´s recharge rate). This makes images suitable for mosaicking only when Alvin moves slowly (≪1/4 kt), which is not always possible for time-critical missions. Video cameras provided a source of imagery more suitable for mosaicking, despite its inferiority in resolution. We discuss required pre-processing and image enhancement techniques and their influence on the interpretation of mosaic content. An algorithm for determination of camera tilt parameters from acquired imagery is proposed and robustness conditions are discussed.