Computer vision is vital for various applications like object tracking for autonomous driving or quality assurance. Hence, assuring that computer vision fulfills given quality criteria is essential and requires sufficient testing. In previous work, authors introduced a testing method relying on image modifications for a photometric stereo application. Image modifications include pixel errors or the rotation of images to be analyzed, revealing a substantial impact on the computed outcome of the photometric stereo application, depending on the applied modification. This paper focuses on whether we can reproduce the impact of image modifications in a real-world setup. In particular, we compare the impact of the rotation of the analyzed sample with the rotation modification applied to the image of the sample. The comparison indicates a similar effect when using rotation, showing that testing based on image modifications is valuable for verifying computer vision applications.
|Lecture Notes in Computer Science
|ICTSS 2023 : The 34th IFIP International Conference on Testing Software and Systems
|18/09/23 → 20/09/23
- High-Performance Vision Systems