We propose a multi resolution method for automatic detection of eddies on both simulated and satellite Sea Surface Temperature (SST) or ocean color images. Our approach consists in a curvelet based analysis of image fronts along circles with specified radius values. The main interests of the proposed method are: It is carried out on image gradient, and is thus not sensitive to the choice of binarization threshold; With the multi-resolution aspect and the anisotropic spatial support shape of curvelet elements, we can deal with no perfectly circular eddies (curvelet width) and we also take into account the image regularity to distinguish between real eddies and noise (curvelet length). We have tested the method on several modeled and satellite images in the North Western Mediterranean sea and the results prove the relevance of the proposed approach.