To characterize the structure of TATB-based polymer bonded explosives（PBX） on the meso-scale， an automatic granule boundary extraction method based on the two-dimensional（2D） reconstructed CT images with low quality is presented. Binarization was first performed according to the gray level distribution of the original CT images of TATB-based PBX. Then a proposed elastic capsule algorithm was applied to the binary images and an approximate location of the granule boundary was obtained. With a combination of morphological operations and the marker-based watershed algorithm， a more accurate segmentation of TATB-based PBX granules was realized. The present granule boundary extraction method for 2D CT images of TATB-based PBX shows improved accuracy and portability， as compared to Canny algorithm， iterative phase congruency， SLIC superpixels and other feature extraction techniques.