Abstract: Measuring the components of carbon/carbon (C/C) composites is an effective way of performance analysis and processing optimization. In this study, a new self-adaptive algorithm of multi-object image segmentation was proposed based on the characteristic of the polarized light microscopic images of C/C composites with pure pyrocarbon fabricated by the chemistry vapor infiltration and the principle of pattern recognition. The optimal thresholds between pores, fibers and pyrocarbons were automatically computed using the improved Otsus method according to the rule of the maximal variance between-class. The experimental results show that the method is effective to separate C/C composites, no matter whether the proportion or the distribution of the components is high or low, massive or scattered. The segmentation quality is fit for the further measurement of C/C composites components.