1D-mosaics have been introduced as a tool for structuring and navigation in video content. These objects can be con- sidered as the spatio-temporal signatures of the video shots. Our work aims at grouping automatically the video shots into scenes using these signatures. The original method is based on the tree-structured lattice vector quantization of the 1D-mosaics. Because of the hierarchical structure of the code-books, they can be compared progressively, and lattice use is time efficient. Indexing retrieval results are given for two video sequences, and different mosaics are successively compared to each other in order to assess the presented scheme's effectiveness.