Abstract
Spatial resolution is a crucial parameter in industrial barcode scanning applications, where high frequencies can be easily achieved in the image space. To address this, zoom lenses that can magnify and increase the image resolution of the object are gaining in popularity. However, due to the design and manufacturing complexity and high cost of zoom lenses, companies designing them need to make tradeoffs to offer a competitive product. This thesis introduces a joint blind super-resolution (SR) and lateral chromatic aberration (LCA) correction model to handle unknown real-world degradations in images captured by barcode reading cameras. Specifically, a large-scale paired dataset is collected for the barcode SR and LCA correction task using a zoom lens for supervised learning. The Real-ESRGAN is optimized with an additional Gradient Profile (GP) loss to achieve accurate barcode upscaling and restoration, named the BarcodeSR. Finally, extensive optical experiments are performed to assess the model's performance and robustness in various barcode reading application scenarios, demonstrating that BarcodeSR enhances barcode reading performance and contrast in images captured by the zoom lens across different magnification levels, field of view (FOV) positions, working distances, illumination colors, sensors, lenses, and optical alignments.
