Accurate knowledge of the current crack length is crucial to evaluate fracture mechanical tests. At the sub-micron to micron scale, the crack length is directly accessible via observation during in-situ experiments in electron microscopes, or indirectly via calculation from sample stiffness. In the current work, image processing techniques were used to introduce a semi-automatic technique to measure crack lengths at the micron scale from image sequences. The technique utilizes manually defined filters and searches for contours near the previous crack tip locating the new one according to the previous position. To demonstrate validity and capability, three micron-sized notched cantilevers were prepared for bending experiments and tested in-situ by partial unloading. Comparison of crack lengths determined by the proposed method, manual measurement and sample stiffness revealed a reasonable agreement, while occasional deviations allow further insights into the crack behaviour. Thus, our new approach enables more in-depth investigation of small-scale fracture processes.