Lossless coding of video sequences using backward-adaptative spatio-temporal prediction