Research On Software Birthmark Selection

Software Birthmark is the unique characteristics that are inherent to a program toidentify the program in the event of suspected theft. Related work and the state-of-art birthmarkalgorithms are summarized. Direct at the problem of the current birthmark algorithm has greatredundancy, the birthmark selection mechanism is brought in, and a software birthmark selectionalgorithm based on mutual information is proposed. To solve the problem of fixed length k-grambirthmark lost many valuable grams, two variable length k-gram birthmark selection algorithmsare proposed. The contents in the thesis are as follows:(1) The concept of birthmark selection referring to the feature selection in the documentclassification of information retrieval is put forward in this thesis. A software birthmark selectionalgorithm based on mutual information (MI) is proposed, two software sets are constructed andMI value is calculated to measure the amount of information between gram’s and softwarecategory, the higher MI value is chosen for the standard gram set. The experiment shows that asthe birthmark data scale decreased， the performance compared to the original k-gram birthmarkis highly increased, and it is the first birthmark algorithm both improve the credibility andresilience.(2) As the software set expands, birthmark selection can cope with more semantictransformation attacks, but the standard gram set will decrease, meanwhile the fixed lengthk-gram lost many valuable grams. To solve this problem, a variable length k-gram birthmarkselection algorithm is designed, this design is relatively close to the semantic segmentation， andenlarge the standard gram set as well. The experiment shows that the design not only perform asgood as previous selection, and improve the resilience to new semantic attack in certain degree.(3) Referring the idea of multiword terms extraction in natural language processing, thek-gram that can be a section is assessed in a certain software sample space utilizing gluecalculating and LocalMaxs algorithm, this can make birthmark extraction more approximate tothe semantic. After section the birthmark selection is used, the experiment show theseimprovement is has a positive significance to detect new semantic attacks.Finally, conclusions are drawn, and the further researches are put forward.