Abstract:
This paper examines the problem of detecting tears on a protected document page. We present an approach based on analyzing the document image in the infrared range. It is assumed that in this case it is possible to separate the damage from the protective elements applied by
IR-transparent inks. So the problem of tears detection might be reduced to a search for thin lines of a certain length adjacent to the border of the document page. Thus, we developed a tear search algorithm based on the search for “ridge” type lines followed by checking whether the line satisfies the specified properties. We created and published a VIUR dataset with Russian banknotes in order to test the algorithm. The recall of the proposed algorithm is 0.87, the precision is 0.94.