In this paper, we jointly study the construction and visualisation of a classification structure of large image collection. Our main goal is to assure the fact that the built structure will be viewable in a web interface while highlighting the topology of the data. To do so, we combine the BIRCH algorithm tree (Balanced Iterative Reducing and Clustering using Hierarchies) with the construction of proximity graphs: such a graph is built and incrementally updated to represent the children of each node in the tree. Furthermore, images representatives are pulled up to internal nodes to help the user during the tree exploration. All the structures have been built to allow dynamic updates regarding the insertion of new images in the collection. The first experiments allow to structure thousands of images in a minute and provides a fluid and intuitive navigation.