3D printing became increasingly popular due to the recent price drop of 3D printers. It also brought concerns about copyright infringements of physical objects that can be easily duplicated and criminal usages such as keys reproduction from picture or weapons printing. Similar to other media, watermarking technology provides ways to investigate these illegal usages. This thesis focuses on blind watermarking for 3D printed objects by applying small geometric modification on the surface. It consists of slightly modifying the surface geometry to covertly insert a marker in a 3D printed object and being able to extract it from the object without requiring any information from the original model. We proposed two new methods.
Our first method uses a surface norm histogram to encode the watermark bits and 3D moments for the object alignment. It can embed around 16-32 bits, works with any 3D printer and scanner, and resists to a reprint from a scanned model.
Our second method uses the printing layer thickness as a 1D signal carrier. It works with FDM printers, can embed more than 64 bits and can be decoded from a single shot using a 2D document scanner. The decoding method is limited to flat surfaces due to the usage of a document scanner but could be extended to curved surface using high-resolution 3D scanning method. The motivation behind this method was to be easily deployable in real-life situations due to the availability and low cost of the equipment.
Both methods produce low distortion and are complementary by fitting different scenarios. Our first method fits copyright protection scenarios in which the input for decoding is a standard 3D scan or a counterfeited object. Our second method fits metadata embedding scenarios in which high capacity or fast decoding is required, but resistance to reprint is not.