Sonar image processing for underwater object detection based on high resolution system Imen Mandhouj, Hamid Amiri Frederic Maussang, Basel Solaiman Systems and Signal Processing (LSTS) Laboratory ENST Bretagne, ITI laboratory National School of Engineering of Tunis (ENIT) Technopole Brest-Iroise, 29238 Brest Cedex Imen.mandhouj@gmail.com, hamidlamiri@yahoo.com {Frederic.maussang;Basel.Solaiman}@telecom- bretagne.eu Abstract—This paper is concerned with the problem of within the sonar image. But for the reason of the complexity recognition of objects laying on the sea-bed and presented on of oceanic environment and the particular optical properties sonar images. Considering that high resolution sonar system of light in water, the sonar image obtained from sonar provides acoustic images of high-quality, several researches instrument is polluted by the noise. Therefore, it became an have been interested in Synthetic Aperture Sonar (SAS) and Sides can sonar images for underwater objects. This work important research field to remove the noise of sonar image presents recent detection algorithms targeting their main [10] before the application of various approaches of image specificity and innovations during the different steps of sonar processing. Furthermore, according to the operated image processing. approach, pre-or post-processing are generally used to make each step of processing more robust [10]. Thus, from the Keywords-component; objects recognition; high-quality preprocessing to the classification process, recent system; image processing; sonar image. approaches present certain variability that we will explain in this work still within the context of sonar image processing I. INTRODUCTION for the underwater objects detection using high frequency systems. Thanks to advances in digital electronics, many new sonar systems have appeared. Today we found out the acoustic camera, interferometric sonar, synthetic aperture This paper is organized as follows. In Section II, various sonar (SAS), the synthesis incoherent sonar, parametric methods of pre-processing sonar images are represented. In sonar, Side scan sonar, etc [1]. Sonar image obtained from Section III, the principle, the utility and the different recent such sonar instruments are used in different fields to realize approaches of sonar image textural analysis are cited. seafloor task, such as navigation, seabed mapping, fishing, Finally, different segmentation and classification methods ocean drilling barrier, oil exploration, mines’ detection, and for underwater objects are represented in Section IV. so on [2]. II. SONAR IMAGE PRE-PROCESSING METHODS In view of the interest in the field of mining since hispanic times and nowadays, several research studies have Because of the particular optical properties of light in been interested in the mining and have been developed for water and the presence of suspended particles, sonar images the detection of underwater mines using high-frequency are very noisy, the lighting is not uniform, the colors are system. Both sides can sonar and Synthetic Aperture Sonar muted and the contrast is low. Most methods dedicated to (SAS) technologies provide high-resolution imagery for reduce the noise apply different filtering and are often mine hunting application. As they are characterized by their classified in two categories: the methods acting in the spatial very high performance resolution, images provided by SAS domain and those acting in the transformed domain. [4] and sides can sonar are of great interest for the detection Stéphane Bazeille and Isabelle Quidu [6] proposed an and classification of objects lying on the seabed or buried in algorithm for image preprocessing sonar which allows the sediment, and mainly, the underwater mines[3][5]. In the correction of lighting, noise and color and requires no user context of mine warfare, detected objects can be classified intervention or a priori information on the acquisition from their cast shadow or their high intensity reflection of conditions. This algorithm is a combination of four different the wave on the object. So several studies have been filters each of which aims to revise a special defect. To developed within this context and are differentiated by their eliminate the defects of non-uniformity of illumination, the approaches and methods used to obtain the search object Homomorphic filtering is used. Sonar image is decomposed SIDOP’12 : 2nd Workshop on Signal and Document Processing into the reflectance factor and the illumination intensity [11] The filtering process is then based on Fourier transform using the following equation: and uses the ‘H’ modified Gaussian filter. On the other hand, in order to attenuate noise acquisition, wavelet transform is used. Indeed, various denoising methods act in g(x,y)=i(x,y).r(x,y). (1) the wavelets domain. These methods have three steps: the computation of a wavelet transform (WT), the filtering of the detail wavelet coefficients and the computation of the g(x,y) is the sonar image, corresponding inverse WT (IWT) [7]. Fig. 2 shows the i(x, y) is the multiplication factor of illumination and r(x, y) principle of the wavelet decomposition. is the reflectance function. Applied Homomorphic filtering algorithm is summed up in the following steps: • Separation of the components of illumination and reflectance using the log of the image in (2): g(x, y) = In�f(x, y)� = In�i(x, y). r(x, y)� = In(i(x, y) ) + In(r(x, y) ). (2) • Evaluation of Fourier transform of the log-image. The resulting equation is as follows : Figure 2. Wavelets transform (WT) implementation, h- 𝐺�𝑤𝑥 , 𝑤𝑦 � = 𝐼�𝑤𝑥 , 𝑤𝑦 � + 𝑅�𝑤𝑥 , 𝑤𝑦 �. (3) lowpass filter, g-highpass filter; h and g form a pair of quadarture mirror filters. • High-pass filtering of the Fourier transform: use of the H modified Gaussian filter : G�wx , wy � = A first category of denoising methods applied in the H�wx , wy �. I�wx , wy � + H�wx , wy �. R�wx , wy �. (4) wavelets domain is based on nonparametric techniques [18] and uses the hard or the soft thresholding filters. Stéphane Where Bazeille and Isabelle Quidu [6]’s algorithm of reducing noise acquisition is based on wavelet decomposition of the w2x + w2y image corrupted by white Gaussian noise using A. F. 𝐻�wx , wy � = (rH − rL ). (1 − wx exp �− �)) + rL. 2∗δ2w Abdelmour and I. W. Selesnick’s orthogonal wavelet [15]. (5) In order to get an approximate value of the noise variance, Donoho and Johnstone’s estimator is applied [18] using the 𝑟𝐻 and 𝑟𝐿 are the maximum and minimum coefficients following estimator 𝜎𝑛 : of the filter and 𝛿𝑥 is the factor determining the cutoff frequency. 𝑚𝑒𝑑𝑖𝑎𝑛(|𝑦𝑖 |) • Evaluations of inverse Fourier transform to return in the 𝜎𝑛2 = � � ². (6) 0.6745 spatial domain. • Application of the exponential function to get the filtered image. Where the 𝑦𝑖 are the coefficients of diagonal details wavelet at the finest scale. The effects of the application of this algorithm on a Finally, to correct color mutation, the algorithm applied a sonar image captured by a SAS system are shown in Fig. 1. linear translation of RGB three histograms so as to equalize their average. Other algorithms are developed in the same context. The paper by Buades [20] proposed denoising algorithm acting in the spatial domain. The algorithm explained a very modern non parametric denoising method based on non- local (NL) averaging. The NL-means algorithm tries to take advantage of the high degree of redundancy of the sonar image and the value of pixel is estimated by the neighbors’ pixel value using the following equation: 𝑁𝐿𝑥 (𝑖) = ∑𝑗∈𝐼 𝛽(𝑖, 𝑗)𝑥(𝑗). (7) Figure 1. The effects of Homomorphic filtering on the preprocessing algorithm. In the right image, complete algorithm is applied, in the left image, the Homomorphic filtering is not used. SIDOP’12 : 2nd Workshop on Signal and Document Processing 𝑁 𝑁 For each pixel 𝑥𝑖 , the weight β(i,j) depends on the 𝐻𝑀 = ∑𝑖=1 𝐺 ∑𝑗=1 𝐺 𝐶𝛿2 (𝑖, 𝑗). (8) similarity between the pixels i and his neighbor j. And “I” defines the set of alls “i” neighbors. The paper by Katkovnik 𝐶𝛿 is the correlation matrix, whereas NG is the gray [16] discussed a denoising method for the case of additive level’s image. noise in a transform’s domain. The method proposed combined the local polynomial approximation (LPA) with The estimated contrast is given by: the maximum likelihood and quasi likelihood for the design of nonlinear filters. Another excellent denoising method for 1 𝐶𝑇 = ∑𝑁𝐺−1 𝑘 2 ∑𝑁 𝑖,𝑗=1,|𝑖−𝑗|=𝑘 𝐶𝛿 (𝑖, 𝑗). 𝐺 (10) the case of transform’s domain makes the object of Argenti 𝑁𝐺 −1 𝑘=0 [12]. In their paper, algorithms proposed for noise reducing used for estimation either a maximum a posteriori (MAP) or The entropy in 𝐶𝛿 is defined by: a linear minimum mean square error (LMMSE) filtering 𝑁 𝑁 approach [40]. The denoising procedure is based on 𝐸𝑁 = 1 − ∑𝑖=1 𝐺 ∑𝑗=1 𝐺 𝐶𝛿 (𝑖, 𝑗)log(𝐶𝛿 (𝑖, 𝑗)). (11) estimating Non-Subsampled Contourlet Transfor (NSCT) coefficients of each band by calculating their fourth order The correlation between the rows and columns of the moments. [40] matrix is given by: To validate the performance of each proposed denoising 𝑁 𝐶𝑅 = ∑𝑖=1 𝐺 ∑𝑁𝐺 𝑗=1((𝑖 − 𝜇𝑥 )(𝑗 − 𝜇𝑦 )𝐶𝛿 (𝑖, 𝑗))/(𝜎𝑥 𝜎𝑦 ). (12) algorithm, a robust criterion must be specified to note the overall quality of the image. Such a criterion is not yet defined in the literature and is still the subject of several where 𝜇𝑥 , 𝜎𝑥 , 𝜇𝑦 , 𝜎𝑦 represent respectively the averages searches. The performance of the proposed algorithm is and standard deviations of the marginal distributions of often evaluated by presenting the effect of pretreatment on elements of co-occurrence matrix. the next steps of sonar image processing and especially The directivity defined the existence of a preferred direction segmentation. of the texture and is computed by the following equation: 𝑁 III. SONAR IMAGES TEXTURAL ANALYSIS 𝐷𝑅 = ∑𝑖=1 𝐺 𝐶𝛿 (𝑖, 𝑗). (13) The analysis of textured images plays an important role in image processing, pattern recognition and particularly in The uniformity characterizes the proportion of the same sonar images classification [8], [9], [13]. There are different gray level and is given by: methods of feature extraction for image processing cited in 𝑁 [21]. As part of mine detection, the distinction between the 𝑈𝑁 = ∑𝑖=1 𝐺 𝐶𝛿 (𝑖, 𝑗)2. (14) image of a mine and an object that physically resembles a mine is very complex and is relied on the recognition in Once these parameters are evaluated, the following shapes and textures. Ph. Blondel [14]’s study combined two parameters: energy, entropy and average are calculated in different advances in sidescan sonar applications [30], [31]. each sub-image of wavelets decomposition. Also, for the In fact, sidescan sonar images in particular, are mainly analysis of image texture, a Gabor filter bank (a set of described by their tonal and textural properties. This method filters, each selecting a frequency and a particular angle in of advanced image analysis has been developed and based the image) is applied. Those parameters extracted are thus on the quantification and recognition of acoustic textures. transformed by linear (principal component analysis PCA, The method has been extensively calibrated and ground- linear discriminant analysis DLA) or nonlinear (Curvilinear truthed in complex terrains. The algorithm implicated was Component Analysis CCA) extracting methods and form then successfully applied to the detection of mines. most discriminating parameters departure. H. Laanaya’s paper [22] presented a method of sonar image K.Imen’s papers [17] propose a new region-based classification based on the process of extracting knowledge segmentation of textured sonar images with respect to from data. In Previous Laayana’s works [25], [26], [27] seafloor types. Haralick’s parameters [35], coefficients of methods for the extraction of textural attributes in images wavelet and Gabor are used as descriptors to characterize upgrades gray are quite similar. The texture is early seabed. The contribution of this method is that the choice of characterized as the statistics of the response to scale-space descriptors is not randomly but based on techniques for filters such as Gabor and wavelet analysis, co-occurrence selection of parameters and attributes of the most matrix [33] or multifractal descriptors [34]. A co-occurrence discriminating textures. The selection problem is addressed matrix is a NG * NG size matrix where NG is the number of through the definition of a similarity measure adapted to the gray levels of the image. Haralick [35] defined 14 texture characteristics of relevant textures towards a learning set of features from the co-occurrence matrix. In this work only textures. The (dis)-similarity between a seabed type T and six parameters are used: homogeneity, contrast, entropy, Tʶ texture is calculated in (15) using the Kullback-Leibler correlation, uniformity and directivity. divergence (KL) [41]: The homogeneity in the 𝛿 direction is given by: ( ) 𝐽 𝐾𝐿𝑤− �𝑄𝑘 , 𝑃𝜃 (𝑇)� + ∑𝐹𝑓=1 ∑𝑗=1 𝑤𝑓2 𝑤𝜃𝑖𝜃𝑗 𝐾𝐿 �𝑄𝑓,𝑗 𝑘 𝑃𝑓𝜃 (𝑇)�. (15) SIDOP’12 : 2nd Workshop on Signal and Document Processing Once the proposed approach based on texture analysis gives the textural features, an application to sonar texture classification is addressed. Nevertheless, when the approach of detecting underwater objects is not based on texture features, the classification step is realized by exploring other paths of segmentation. IV. SONAR IMAGE SEGMENTATION AND CLASSIFICATION METHODS Different methods of segmentation and classification of underwater objects mainly a mine are cited in [19]. Many segmentation methods are based on statistical model using first and second order statistics [28], [29]. Frederic’s work [23] presented a new method of underwater mine echoes Figure 4. Mean–standard deviation representation of the test image. detection. Segmentation echoes step is based at first on This representation is built with a 3 x 3 window size in figure (a), 5 x 5 in figure (b) and 21 x 21 in figure (c). using local statistical proprieties of sonar images. Higher Order Statistics (HOS) are then used to improve detection tool [24]. The first and second order statistics have served to a mean-standard deviation representation. Kurtosis and To automate the segmentation algorithm, the standard Skewness are evaluated on a square window for each pixel deviation threshold value is evaluated using the entropy [32] to obtain correct location of the echoes of detected criterion. The entropy value is calculated on each axis of the objects. As a test of Frederic’s algorithm, the case of sonar image using the Shannon formula as follows: image presented in Fig. 3 is taken. Corresponding mean– standard deviation representation is presented in Fig. 4. In this figure we show the result of three tests realized in order 𝐻𝑎𝑥𝑖𝑠 = − ∑𝑖∈𝐼 𝑝𝑎𝑥𝑖𝑠 (𝑖)𝑙𝑜𝑔2 𝑝𝑎𝑥𝑖𝑠 (𝑖). (18) to segment the test image using [23] the first and second image’s local statistical proprieties. For each of three Where 𝑝𝑎𝑥𝑖𝑠 (𝑖 ) is the number of segmented pixels in the calculation window, the dashed line represents the column i (respectively line i), and “I” is the set of columns proportionality coefficient between mean and standard (respectively lines). deviation (estimated with the Weibull law), and the solid lines feature the threshold values. This method is finally tested on real SAS data containing underwater and other objects, laying on the seabed or buried On each calculation window, the expression of on the seafloor [23]. Skewness (16) and Kurtosis (17) is given as a function of the proportion of pixels in the deterministic calculation The test of the segmentation method on the test SAS window (𝜌) and the signal to noise ratio (p) as follows: image and the value of entropy calculated for each axis using the above equation are presented in Fig. 5. 𝑝 (1−2𝑝)𝜌3 −3𝑝 𝑆𝐹 (𝜌, 𝑝) = 3 . (16) �1−𝑝 (𝑝𝜌2 +1)2 𝜌 �(1−6𝑝+6𝑝2 )𝜌4 −6(1−2𝑝)𝜌2 +3� 𝐾𝐹 (𝜌, 𝑝) = 1−𝑝 (𝑝𝜌2+1)2 . (17) Figure 5. Segmentation graph and repartition of segmented pixels according to the two axes. Results for a standard deviation value = 8000 and an average = 6750. Computed entropies: X-axis: 4.16; Y - axis: 8.76. Another view of classification approach is revealed in Arnaud Martin and I. Isabelle Quidu’s paper [37]. In fact, Figure 3. SAS test image of buried and proud objects at showing two for sonar images classification, A. Martin and I. Isabelle types of mine: a spherical (placed in the left end of the image) and a take advantage of Support vector machines (SVM) classifier second practically buried in the bottom (placed in the right end). based on the statistical learning theory [38]. The SIDOP’12 : 2nd Workshop on Signal and Document Processing contribution of the proposed method is the use of the belief matrix. On the strategy using the predictive phase in the functions theory for the combination of binary classifiers method presented, results are generally considered of high coming from the SVM. The modelization of the basic belief performance for both parameters calculated. assignments is then evaluated directly from the decision functions given by the SVM. Additionally, the approach Generally, in order to show the contribution of the proposed in Laayana’s paper [22] aims to make first robust proposed method, classification approach is compared with traditional classification methods, such as support vector classification on the same data, but using other strategies. machines or k-nearest neighbors then apply evidential and fuzzy SVM for regression. For finding the SVM’s hyper V. CONCLUSION plane, the proposed method uses one of two approaches one- against-one or approach one-against-rest. The choice of the The focus of this paper is the study of the invariant k nearest neighbors is labeled by a set distance. The approaches for sonar objects detection with application to individual are then placed in the class to which belongs the high frequency sonar system images. This paper explained greatest K number of neighbors. Another proposed approach considerable recent work to underwater objects detection to integrate the concept of fuzzy in SVM is to model the but also it presented a few of previous work in the same noise on the inputs by weighting. For the evaluation of filed in order to establish the mutation of accurate method. classification approaches exploited, [22] adopts the result of Several research studies for detection approach turn to apply [36] work. The comparison between the different classifiers different methods of image processing including filtering, showed that SVM gives the best classification rate. Later segmentation and objects’ classification and involved, in the Isabelle Leblond and Isabelle Quidu propose an original case of need, other technologies (principal component strategy to classify underwater objects from SAS images analysis, fusion methods, fuzzy estimation…). It is generally [39]. The method proposed in this work allows classifying difficult to make comparison between different recognition an object from multiple views via a prediction of the new algorithms, even for the same filed, since different test sets angle of visit for the additional views in order to take off the are used to evaluate performance. In fact, to validate the ambiguities on the classification. The classification performance of each proposed algorithm, some works algorithm is presented in Fig. 6 as a chain of transactions challenge to define a criterion that it considers robust to note whose extent is based on the results of the previous step. the overall quality of the image. Other works make a comparison with previous results applied on the same sonar data. Thanks to their diversity, many novel approaches are the result of a simple combination of two or more existing methods. However, more work is needed to understand a number of important factors which typically affect the sonar data. REFERENCES [1] M. Legris, K. Lebart, F. Fohanno, and B. Zerr, “Les Capteurs D’imagerie En Robotique Sous-Marine: Tendances Actuelles Et Futures,” Traitement du Signal, 20:137–164, 2003. [2] X. Lurton, “ Eléments D'analyse Des Systèmes Acoustiques Sous- Marins, ” ENSIETA, Brest, France, 2002. duplicated lecture note. Figure 6. Simplified diagram of Isabelle Leblond and [3] P. T. Gough and D. W. Hawkins, “A Short History Of Synthetic Isabelle Quidu’ classification algorithm. Aperture Sonar,” in Proc. IEEE Int. Geosci. Remote Sens. Sump, Seattle, WA, Jul. 1998, pp. 618–620. [4] M. P. Hayes and P. T. Gough, “Synthetic Aperture Sonar: A Maturing Discipline,” in Proc. 7th Eur. Conf. Underwater Acoust., Delft, The Netherlands, Jul. 2004, pp. 1101–1106. The first operation consists in the classification of the [5] M. P. Hayes and P. T. Gough, “Using Synthetic Aperture Sonar For sonar image using the method of K-nearest neighbor applied Mine Detection,” in Proc. Austral.-Amer. Joint Conf. Technol. Mines Mine Countermeas. Syst., Sydney, Australia, Jul. 1999, pp. 1.1–1.10. to the result of the ACP presentation in the plane of the [6] Stéphane Bazeille, Isabelle Quidu, Luc Jaulin and Jean-Philippe object. In case where an additional view must be chosen, the Malkasse, “Automatic Underwater Image Denoising, ” GRETSI, prediction step is performed. This primary step is to estimate Saint Martin d'Hères, France, 2008. the orientation of the object and then propose the angle of [7] Dandan Liu, “The Sonar Image Denosing Method of Surfacelet Based his next visit. The choice of the new direction is then based on Sample Matrix,” Computing, Communication, Control, and on the estimated angle and the defined rules. These rules Management, 2008. CCCM '08. ISECS International Colloquium on, Guangzhou, august 2008. specific to each type of mine are well defined for the case of [8] H. Laanaya, “Classification En Environnement Incertain : Application mine and cylinder Rockan mine and also in the case of a À La Caractérisation De Sédiment Marin,” in Ph.D. Thesis of the general class mine. Performance evaluation of the method is universities of Britany Occidental, Brest, France, and Mohamed-V evaluated by calculating two parameters: "rate of correct Agdal, Rabat, Maroc, December 2007. identification" and "false alarm rate" from the confusion SIDOP’12 : 2nd Workshop on Signal and Document Processing [9] A. Martin, “Comparative Study Of Information Fusion Methods For MTS/IEEE Oceans’03 conference, pp, 2155-2160, SanDiego, Sonar Images Classification,” in The Eighth International Conference Calofornia, USA, September 2003. on Information Fusion, Philadelphia, USA, 25-29 July 2005. [30] Ph. Blondel and O. Gómez Sichi, “Textural Analyses Of Multibeam [10] J. Chanussot, F. Maussang, and A. Hétet, “Scalar Image Processing Sonar Imagery From Stanton Banks, Northern Ireland Continental Filters For Speckle Reduction On Synthetic Aperture Sonar Images,” Shelf,” Applied Acoustics, vol. 70, no. 10, 2009. in Proc. MTS/IEEE OCEANS Conf., Biloxi, MS, Oct. 2002, pp. [31] Karasalo I., P. Skogqvist, Ph. Blondel and P.F. Dobbins, “Buried 2294–2301. Waste Inspection: Acoustical Images And Inversion From Multiple- [11] R. Garcia, T. Nicosevici, and X. Cufi, “On The Way To Solve Aspect Scattering,”in “Buried Waste In The Seabed - Acoustic Lighting Problems In Underwater Imaging,” IEEE OCEANS 02, Imaging And Bio-Toxicity (Results From The European SITAR 2002. Project),”Ph. Blondel and A. Caiti (eds.), Springer-Praxis, p. 115-126, [12] Argenti F., Bianchi T. and Alparone L. , “Multiresolution MAP 2007. Despeckling of SAR Images Based on Locally Adaptive Generalized [32] Maussang Frederic, Chanussot Jocelyn, Hétet Alain and Amate Gaussian PDF Modeling,” IEEE Trans. on Image Proc., vol. 15, no. Maud, “Mean / Standard Deviation Representation Of Sonar Images 11, Nov. 2006, pp. 3385-3399, ISSN 1057-7149. For Echo Detection : Application To SAS Images,” IEEE journal of [13] A. Martin, “Fusion for Evaluation of Image Classification in oceanic engineering, october 2007, vol. 32, n° 4, pp. 956-970, 2007. Uncertain Environments,” in The Nineth International Conference on [33] Trygve Randen and John Hakon Husy, “Filtering For Texture Information Fusion, Florence, Italy, 10-13 July 2006. Classification: A Comparative Study,” TPAMI, vol. 21, pp. 291–310, [14] Ph. Blondel, “Automatic Mine Detection By Textural Analysis Of 1999. COTS Sidescan Sonar Imagery”, International Journal of Remote [34] Yong Xu , Hui Ji and Cornelia Fermuller, “Viewpoint Invariant Sensing Vol. 21, Iss. 16, 2010. Texture Description Using Fractal Analysis,” IJCV, vol. 83(1), pp. [15] A. F. Abdelmour and I. W. Selesnick, “Symmetric Nearly Orthogonal 85–100, 2009. And Orthogonal Nearly Symmetric Wavelets,” Research Report, [35] R. Haralick, “Statistical And Structural Approaches To Texture,” 2003. Proceedings of the IEEE, Vol. 67(5), pp 786-804, 1979. [16] Katkovnik V., Egiazarian K. and Astola J., “Local Approximation [36] Arnaud Martina, Hicham Laanayaa and Andreas Arnold-Bosa, Techniques in Signal and Image Processing,” SPIE Press, ISBN 0- “Evaluation For Uncertain Image Classification And Segmentation,” 8194-6092-3, Bellingham, Washington, USA. journal Pattern Recognition archive, Volume 39 Issue 11, November, [17] Karoui Imen, Fablet Ronan, Boucher Jean-Marc and Augustin Jean- 2006, Elsevier Science Inc. New York, NY, USA. Marie, “Segmentation Texturales Des Images Sonar Des Fonds [37] Arnaud Martin and Isabelle Quidu, “Decision Support With Belief Marins Par Une Approche Variationnelle Basée Région,” gretsi, saint Functions Theory For Seabed Characterization,” Conference martin d'hères, france, 2008. Proceeding: DOI: 10.1109/ICIF.2008.4632317ISBN: 978-3-8007- [18] D. L. Donoho and I. M. Johnstone, “Ideal Spatial Adaptation Via 3092-6In proceeding of: Information Fusion, 2008. Wavelet Shrinkage,” Biometrika, 81(3):425.455, 1994. [38] V.N. Vapnik, “ Statistical Learning Theory,” John Wesley and Sons, [19] J.T.Broach, R.S Harmon and G.J. Dobeck, “Detection and 1998. Remediation Technologies for Mines and Mine like Targets VII,” [39] Leblond, Isabelle, Quidu, Isabelle, Berron and Cécile, “ Mise En conference chair, in Proc.of SPIE vol. 4742 Orland, florida, USA, Oeuvre De Stratégies Prédictives Sur Les Vues A Acquérir Pour La April 2002. Classification Multi-Vue D'Objets Immergés A Partir D'images SAS, [20] Antoni Buades Bartomeu Coll and Jean-Michel Morel, “Image And ” 17th Conference on Pattern Recognition and Artificial Intelligence Film Denoising By Non-Local Means,” Ph. D. Thesis, Universitat de (RFIA),Caen :France (2010). les Iles Baleares. [40] Argenti F., Bianchi T., Scarfizzi G. and Alparone L., “SAR Image [21] A. Martin, G. Svellec, and I. Leblond, “Characteristics Vs Decision Despeckling in the Undecimated Contourlet Domain: A Comparison Fusion For Sea-Bottom Characterization,” Colloque Caract´erisation of Lmmse and Map Approaches ,” Geoscience and Remote Sensing in-situ des fonds marins, Brest, France, 21-22, October 2004. Symposium, 2008. IGARSS 2008. IEEE International. [22] H. Laanaya, A. Martin, A. Khenchaf and D. Aboutajdin , “Extraction [41] Pedro J. Moreno, Purdy P. Ho and Nuno Vasconcelos, “A Kullback- De Connaissances A Partir De Données Pour La Classification Des Leibler Divergence Based Kernel for SVM Classification in Images Sonar, ” GRETSI, Saint Martin d'Hères,France,2008. Multimedia Applications,” Cambridge Research Laboratory, HP [23] Maussang Frederic, “Traitement D'images Et Fusion De Données Laboratories Cambridge, January 9 th, 2004. Pour La Détection D'objets Enfouis En Acoustique Sous-Marine, ”, Ph. D. Thesis , Universitat Joseph Fourier - Grenoble ,2005. [24] Maussang Frederic, Chanussot Jocelyn, Hétet Alain, Amate Maud, “Higher-Order Statistics For The Detection Of Small Objects In A Noisy Background Application On Sonar Imaging,” EURASIP journal on advances in signal processing, 2007, vol. 2007. [25] H. Laanaya, A. Martin, A. Khenchaf and D. Aboutajdine, “Dimensionality Reduction Of Sonar Images For Sediments Classification,” Brest, France, 21-22, October 2004. [26] H. Laanaya, A. Martin, A. Khenchaf and D. Aboutajdine, “Feature Selection Using Genetic Algorithm For Sonar Images Classification With Support Vector Machines,” European Conference on Propagation and Systems, Brest, France, vol. 24(11), 15-18, March 2005. [27] C. Archaux, H. Laanaya, A. Martin, and A. Khenchaf, “An SVM Based Churn Detector In Prepaid Mobile Telephony,” International Conference On Information & Communication Technologies (ICTTA), Damas, Syrie, pp. 19–23, 2004. [28] M.Mignotte, C.Collet, P.Pérez, and P.Bouthemy, “Three-Class Markovian Segmentation of High-Resolution Sonar Images”, Computer Vison and Image Understanding, vol. 76, no.3,pp.191- 204,December 1999. [29] Maussang Frederic, J.Chanussot, and A.Hétet, “Automated Segmentation of SAS Images using the Mean-Standard Devation Plane for the Detection of Underwater Mines,” In Proc. Of