Tape substrate (TS) product is a high-density circuit pattern on thin film substrate, and it requires precise and high resolution imaging system for inspection. We introduce here a TS inspection system developed, where the products are fed through a reel to reel system, and a series of inspection algorithms based on a referential method. In the system, it is so hard to achieve consistent images for such a thin and flexible materials as TS product that the images suffer from individual, local distortion during the image acquisition. Since the distortion results in relatively big discrepancy between an inspection image and the master one, direct image to image comparison approach is not available for inspection. To inspect the pattern in a more robust way in this application, we propose a graph matching method where the patterns are modeled as a collection of lines with link points as features. In the offline teaching process, the graph model is achieved from skeleton of the master image, which is collected as a data base. In the run time, a boundary tracking method is used for extracting the graph model from an inspection image instead of a skeleton process to reduce the computation time. By comparing the corresponding graph models, a line that is linked to undesired endpoints can be detected, which becomes an open or short defect. Through boundary tracking approach, we can also detect boundary defects such as pattern nick and protrusions as well.