Papers by Jean Marie Augustin
IEEE Transactions on Geoscience and Remote Sensing, 2009
In this paper, we propose and compare two super-4 vised algorithms for the segmentation of textur... more In this paper, we propose and compare two super-4 vised algorithms for the segmentation of textured sonar images, 5 with respect to seafloor types. We characterize seafloors by a set of 6 empirical distributions estimated on texture responses to a set of 7 different filters. Moreover, we introduce a novel similarity measure 8 between sonar textures in this feature space. Our similarity mea-9 sure is defined as a weighted sum of Kullback-Leibler divergences 10 between texture features. The weight setting is twofold. First, 11 each filter is weighted according to its discrimination power: The 12 computation of these weights are issued from a margin maxi-13 mization criterion. Second, an additional weight, evaluated as an 14 angular distance between the incidence angles of the compared 15 texture samples, is considered to take into account sonar-image 16 acquisition process that leads to a variability of the backscattered 17 value and of the texture aspect with the incidence-angle range. 18 A Bayesian framework is used in the first algorithm where the 19 conditional likelihoods are expressed using the proposed similarity 20 measure between local pixel statistics and the seafloor prototype 21 statistics. The second method is based on a variational framework 22
Journal of The Acoustical Society of America, 2008
A quantitative analysis was conducted over sonar backscatter data collected on the Cook Strait re... more A quantitative analysis was conducted over sonar backscatter data collected on the Cook Strait region, central New Zealand, featuring multibeam (∼ 30 kHz) bathymetry and backscatter data, groundtruthed by an extensive geological database (photographs, seabed samples, high-resolution seismics). A first processing step removes the effects of the sounder, seafloor topography, and water column. A second step includes sonar image mosaicing, signal calibration and compensation, speckle noise filtering, image segmentation and textural analysis. Backscatter angular dependence is then extracted from the raw data accounting for the co-registered multibeam bathymetry ; it is linked to the various facies of this geologically very active region, forming a catalogue usable for future investigation. Some local features are analysed in details, referring to the geological local context. Also the backscatter data from the Haungaroa volcano were used for a proofof-concept biodiversity mapping exercise. Ecological theory was utilised to predict biodiversity from the seabed substrate heterogeneity, derived from the segmentation of the backscatter data properly pre-processed. The backscatter analysis resulted in the identification of local features with geological, sedimentological, topographic, and possibly biological significance, otherwise not recognised with conventional surveying. This emphasises the potential of backscatter data in submarine seismic hazard studies and large-scale biodiversity mapping.
Continental Shelf Research, 2011
A comprehensive EM300 multibeam echo-sounder dataset acquired from Cook Strait, New Zealand, is u... more A comprehensive EM300 multibeam echo-sounder dataset acquired from Cook Strait, New Zealand, is used to develop a regional-scale objective characterisation of the seafloor. Sediment samples and highresolution seismic data are used for ground-truthing. SonarScope s software is used to process the data, including signal corrections from sensor bias, specular reflection compensation and speckle noise filtering aiming at attenuating the effects of recording equipment, seafloor topography, and water column. The processing is completed by correlating a quantitative description (the Generic Seafloor Acoustic Backscatter-GSAB model) with the backscatter data. The calibrated Backscattering Strength (BS) is used to provide information on the physical characteristics of the seafloor. The imagery obtained from the BS statistical compensation is used for qualitative interpretation only; it helps characterizing sediment facies variations as well as geological and topographic features such as sediment waves and erosional bedforms, otherwise not recognised with the same level of detail using conventional surveying. The physical BS angular response is a good indicator of the sediment grain size and provides a first-order interpretation of the substrate composition. BS angular response for eight reference areas in the Narrows Basin are selected and parameterised using the GSAB model, and BS angular profiles for gravelly, sandy, and muddy seafloors are used as references for inferring the grain size in the reference areas. We propose to use the calibrated BS at 451 incidence angle (BS 45 ) and the Specular-To-Oblique Contrast (STOC) as main global descriptors of the seafloor type. These two parameters enable global backscatter studies by opposition to compensated imagery whose intensity is not comparable from one zone to the other. The results obtained highlight the interest of BS measurements for seafloor remote sensing in a context of habitat-mapping applications.
IEEE Transactions on Image Processing, 2010
This paper investigates variational region-level criterion for supervised and unsupervised textur... more This paper investigates variational region-level criterion for supervised and unsupervised texture-based image segmentation. The focus is given to the demonstration of the effectiveness and robustness of this region-based formulation compared to most common variational approaches. The main contributions of this global criterion are twofold. First, the proposed methods circumvent a major problem related to classical texture based segmentation approaches. Existing methods, even if they use different and various texture features, are mainly stated as the optimization of a criterion evaluating punctual pixel likelihoods or similarity measure computed within a local neighborhood. These approaches require sufficient dissimilarity between the considered texture features. An additional limitation is the choice of the neighborhood size and shape. These two parameters and especially the neighborhood size significantly influence the classification performances: the neighborhood must be large enough to capture texture structures and small enough to guarantee segmentation accuracy. These parameters are often set experimentally. These limitations are mitigated with the proposed variational methods stated at the region-level. It resorts to an energy criterion defined on image where regions are characterized by nonparametric distributions of their responses to a set of filters. In the supervised case, the segmentation algorithm consists in the minimization of a similarity measure between region-level statistics and texture prototypes and a boundary based functional that imposes smoothness and regularity on region boundaries. In the unsupervised case, the data-driven term involves the maximization of the dissimilarity between regions. The proposed similarity measure is generic and permits optimally fusing various types of texture features. It is defined as a weighted sum of Kullback-Leibler divergences between feature distributions. The optimization of the proposed variational crite- - ria is carried out using a level-set formulation. The effectiveness and the robustness of this formulation at region-level, compared to classical active contour methods, are evaluated for various Brodatz and natural images.
A new method for the segmentation of textured backscattering strength (BS) sonar images is presen... more A new method for the segmentation of textured backscattering strength (BS) sonar images is presented. The method is based on the analysis of joint wavelet statistics by using the whole information brought by cooccurrence distributions. After the wavelet transform of the image, on the most informative frequency bands of the wavelet transform, we discriminate between textures by directly measuring the similarity between co-occurrence statistics. Then, we fuse the different segmentations according to the weighted voting rule. Results on real sonar images and textures from the Brodatz album illustrate the effectiveness of the scheme. Finally, performances and results are discussed.
Uploads
Papers by Jean Marie Augustin