Character recognition (OCR) using Shape from Shading

Low-contrast inscriptions applied by means of scribing, stamping or mold casting pose a particular challenge to machine vision and, in particular, to illumination technology. The numbers are recessed or raised on the respective components. In addition, there are often very different surfaces. Production-related fluctuations in the application of the lettering represent a further requirement.

Using the Shape from Shading technology, a process is employed which, as the name suggests, works with light and shadow. By arranging four LED bar lights at a 90° angle to each other, image recordings are generated continuously, taking into account the angle of light incidence. These four shadow images are then calculated. The height information is displayed in a greyscale image. Additional parameters can be used to enhance contrasts and smooth interfering contours (e.g. in the surface). The resulting image is evaluated with an OCR tool. The read sequence of digits is then transferred to the higher-level system control for further processing.

Where classical machine vision reaches its limits, the Shape from Shading method can create new solutions. For the automated optical inspection of Direct Part Marks (DPM) on components, this method increases process reliability. In addition, high-contrast printing (e.g. via inkjet or laser) can be avoided during marking, which contributes to cost savings. This method also can be used to solve other tasks, such as edge inspections, placement control or surface inspections.