Quantitative analysis of seabed features in video images is of a great importance for numerous underwater applications. It is particularly interesting to produce mosaics from the seabed to create a video map whose extent is far larger than the size of a still camera image.

The MATISSE® system (Mosaicing Advanced Technologies included in a Single Software Environment developed by Ifremer) is based upon image and signal processing components and produces on-line and off-line geo-referenced mosaics of the seabed by given video input and navigation data (Allais et al., 2004; Vincent et al. 2003). The resulting mosaics can be represented in a dedicated environment as a GIS enabling a further image analyses and linkages to other georeferenced data.

The whole algorithm to produce geo-referenced mosaics is performed by different steps as lighting correction, camera self-calibration, and mosaic post-processing. Lighting correction is necessary because in case of deep underwater vehicle applications, videos are acquired using an artificial lighting. In general, a non-uniform intensity distribution is produced within the image and this can have negative effects on the mosaic processing and rendering. The intensity correction in an underwater environment was investigated in Borgetto et al. (2003).

Since the marine environment is subject to varying optical conditions and since existing uncalibrated video data can be considered, tools for automated camera calibration without external means are required. A specific method has been developed in (Pessel et al., 2003). This method needs only a sequence of few images of the seabed and is based upon the determination of the epipolar geometry between two successive images of the sequence. The epipolar geometry is represented by the fundamental matrix which links the image coordinates of the same point in two images. The method is then based upon the estimation of the fundamental matrix which is a function of the intrinsic and extrinsic parameters (Pessel et al., 2003).

At this stage, the mosaics can be built and enhanced by lighting correction. The next step introduces trajectory data into the processing of mosaic correction. Trajectory data of the camera are directly linked to vehicle navigation data provided by data from sensors such as Doppler velocity log, gyrocompass, inertial systems, which imply a long-term drift, or noisy acoustic positioning. A second method is based upon the trajectory of the image centre on the seafloor. A preliminary step allows correction of the image trajectory by blending vehicle navigation and image motion. Once the image trajectory on the seafloor is corrected, the mosaics are processed by warping algorithms (Jouffroy and Opderbecke, 2004).