A device and a process for compressing video documents. The process includes decomposition of the video document into shots, extraction of images representative of the said shots, measurement of the distances between the said representative images, grouping of the images whose distance is less than a predetermined threshold into a group of images, calculation of the relative cost of coding of each image with respect to the other images of the group of images. The process further includes calculation of a minimum spanning tree for each group of images minimizing the cost of coding of each group of images and coding of each group of images by browsing the images in the order given by the said spanning tree.