1 Sea Surface Slicks Characterization in SAR Images T. F. N. Kanaa Ecole Nationale Supérieure Polytechnique (ENSP) de Yaoundé, B.P. 8390, Yaoundé, Cameroun. Email : t_kanaa@yahoo.fr . G. Mercier Ecole Nationale Supérieure des Télécommunications (ENST) de Bretagne, Technopole Brest- Iroise, 29238 Brest, France. Email : Gregoire.Mercier@enst- bretagne.fr . E. Tonye Ecole Nationale Supérieure Polytechnique (ENSP) de Yaoundé, B.P. 8390, Yaoundé, Cameroun. Email : tonyee@hotmail.com . Abstract – The authors present a new method to characterise and discriminate oil slicks and some look-alikes in ERS-2 SAR images according only to the observed sea roughness, to reduce oil spill detection and monitoring systems cost. It exploit sea wave spectrum images from the multiscale analysis based on a modified morphological pyramid. Many backscatter characteristics extracted at each level, depended on object and background features are normalized to make its spectral scales be identical. Twenty objects (spot and border) backscatter features have been measured. Eleven sea surface slicks types have been analysed, namely oil, atmospheric instability, wind front, unstable air-mass, current front, falling land wind, large gravity waves, low wind area, natural slicks, swell visible and wind sheltered area. The results presented as smoothed basic profiles and textural spectra allow to tackle oil slicks supervised classification in new images. Oil slicks and current front are discriminated. But, some ambiguities of slicks discrimination in SAR images remain persistent. I. INTRODUCTION Different investigations has approved of the synthetic aperture radar (SAR) images processing diagram for oil slick detection in three steps [1]: dark spots detection, dark spots features extraction and spots classification. Most of dark spots detection algorithms do not take to account the sea state [2]. The study of the dampening of the wave spectra energy has been done by the multiscale analysis of the remotely sense observation. Wavelets transforms are first used, and a Markov chain model is applied for the slick segmentation [3]. Then a modified morphological pyramid is included to the fuzzy c-mean algorithm for the refinement of the detection results [4]. In general, the detected spot need several information de be classify as oil slick or its look- alikes, particularly the characteristic image expression, the backscatter profiles and gradients, the geographical occurrence and the weather limitations [5]. Unfortunately, this approach costs a lot and results always to an supervised classification. The oil slick detection problem needs information at real time. This has drive some scientists to consider the geometrical features, a few surroundings synergetic data [6] by making contextual analysis, and the backscatter values [7, 8, 9, 10]. In the actual condition, auxiliary data are expensive and not available. This forces the authors to work with only textural measures of the backscatter values to reduce oil spill detection and monitoring systems cost. Oil slick floating on the sea surface has an influence on these ocean properties [11]. The physico-chemical results of the hydrophobic interaction between the water and the oil involve the mutual dissolution of the one in the other. This mixture gives place to three layers according to the intensity of the mixture agitation : floating oil, layer of conflict and sea water not polluted. The layer of conflict, located at the intermediary of both others, is called the mixture saturated solution or dispersed phase. It creates much confusion in the images and is characterized by a dynamic mixture of the two fluids. According to this hypothesis, all three corresponding regions in the sea SAR image named as spot, border and background remain interesting for slicks characterization. The multiscale approach is set about dark spot detection and objects features extraction (§II). The data used and the textural profiles obtained are then presented (§III), slicks discrimination is experimented in three new images with contents not identified (§IV), the comments and the futures improvements are described in §V. II. CHARACTERIZATION METHOD A. Dark spot detection This multiscale analysis is carried out by the use of a modified morphological pyramids on adaptive filtering [4]: ( ) ( ) [ ] j GC f j GC f j GC f j GC cv j FI j FI j FI j FI e f E f f E f - ∨ ⋅ - + ⋅ - + = 1 (2.1)