Falsification of someone’s handwritten signature is very common nowadays because it is one of the most frequently used mean of authorization. Therefore, handwriting experts have more and more work, so we can help their work with giving a software to them which increases their work efficiency. This software helps them to decide whether the signature in question belongs to the person, who is believed that he or she has written it.
In my work I have joined a software project being developed for this aim. Experts often use images of signatures, which have something that distracts the authorization, e.g. seals or stamps near the signature. One of my exercises was to implement a feature to remove these from the images.
Another common problem is, that the user has only the scanned version of the signature, so that they can not measure the true dimensions of the picture, or one of the features of handwriting directly. In these cases, they use some ad-hoc ways to define these lengths, areas or angles. So, I had to give a solution for this problem, to define these values of these geometric shapes’ attributes with the knowledge of the properties of the image.
In my thesis, I present the application and its architecture, technologies and frameworks I have been using during by work. Then, I give a potential solution for extracting the seals from the images of these signatures. In the second part of my thesis, I show how to solve the second problem, and how I implemented the features which calculate the quantitative properties of a linear segment or a loop.