In order to perform autonomous manipulation in underwater surveys, a robust seabed type classification technique is crucial. Seabed images convey a lot of information about seabed types and various image segmentation methods have been implemented to classify seabed types by analyzing the features of images such as contour and region. However, these strategies are not robust for diverse underwater environments. Therefore, this paper proposes a novel method based on multifractal spectrum to descript and classify the deep seabed types by analyzing the textures. The applicability of multifractal approach to seabed type classification is verified by different sample images of deep seabed.