An image sequence analysis procedure is developed to quantitatively analyze complex multiple crack patterns in tension tests of fiber-reinforced composite specimens. Planar textured surfaces of such specimens can be observed with a monocular image sequence using a camera of suitable spatial and temporal resolution. Due to the narrow crack paths, a dense high-precision displacement vector field is computed applying least-squares image matching techniques. Some uniformly distributed matching points are triangulated into a mesh. To measure deformations, principal strains and crack widths are computed for each face. Stretched triangles presumably containing one or multiple cracks are subdivided into three new triangles in order to densify the mesh in critical regions. The subdivision is repeated for some iterations. The crack width computation of the triangles requires at least three vertices and its displacements. Due to the dense displacement vector field, there are more points available. In this paper, an algorithm for the crack width computation in a least-squares fit is presented.