A dynamic crash loading experiment is performed on a polypropylene foam. Several interrupted shocks are conducted, in between which microtomographic acquisitions are made, showing the evolution of the sample during its compression. This data can help construct and validate predictive models, although, because this material is multiscale (consitutive grains at the mesoscopic scale are made of microscopic closed cells), image processing is required to extract useful quantitative measures. Such processing is described here, so as to determine a representative volume for each grain of the sample, in order to associate to each grain and to each stage of the compression values such as grain density. This can help build a predictive model at the mesoscopic scale.