Most of the easyTouch tools available in MVTec MERLIC are optimized for processing byte images and generally do not work as expected for other pixel types, such as int1', 'int2', 'int4', 'int8', 'real', and 'uint2'. This is due to the extreme gray value ranges of non-byte images, and usually leads to a black or black and white visualization.
In order to apply MERLIC’s standard tools to different pixel types, it is highly recommended to convert 3D-height images to byte images. In order to do so, two pre-processing tools are available: Remove Outlier Pixels and Scale Gray Range to 8 Bit.
The conversion of non-byte images to byte images by appropriate scaling should generally be based on the minimal gray value range that describes the working depth. This ensures the preservation of as much information as possible.
Depending on the use case, different approaches for adequate scaling can be used. On the right, two different use cases are considered: an alignment task and reading embossed text.
For standard alignment tasks, first just determine the region that describes the object of interest. This region can be obtained by applying the tool Remove Outlier Pixels. Then, use the returned region as Region for Scaling in the Scale Gray Range to 8 Bit.
After the image has been converted successfully to byte format, one of the easyTouch-based alignment tools can be applied.
Reading embossed text
However, in order to find smaller defects on the surface or to read the embossed text successfully, a smaller gray value range should be used. In the case of reading embossed text, this minimal gray value range can be determined by drawing a ROI around the text.
After converting non-byte 3D-height images to byte images, usually smaller structures in the height image will become visible due to the reduced range of gray values.
Finally, after the 3D-height image has been converted to a byte image, the text can be read by Read Text and Numbers.
Besides the method shown in our article, two more 3D pre-processing tools are available: Convert Disparity to Height Image, which converts disparity images from stereo cameras to calibrated height images and Level Surface, which levels tilted or curved surfaces.