In this paper we present a method to define signatures for colour object based on the Zernike moments. The aim of these signatures is to characterize simultaneously the object shape and texture. We propose two kinds of signatures that are invariant to image scale, rotation and translation.The first one is based on a marginal approach that processes each RGB component separately. The Zernike moments are computed considering the greyscale value of the pixels and the used method provides two maps of moments for each component, one with real values and one with imaginary values. An adapted comparison procedure is used to compare the moment maps of two distinct objects.The second signature is based on a vectorial approach where each pixel is characterized by a 3D vector defined in the RGB cube. The Zernike moments are now computed considering the colour variation between two adjacent pixels. This second approach is faster and provides only 2 moment maps that represent the colour variations of the object.These signatures have been used to track exotic fishes in an aquarium in the framework of the Aqu@theque program.