Research On Digital Comic Copyright Recognition Technology Based On ASIFT And Image Hash

Digital comics are computer comics made by using digital media technology,which are mainly made by using the digital technology of computers unlike traditional manual comics,most 3D animations belong to the scope of digital comics,and the rapid rise of digital technology has injected a brand new vitality into the rapid development of China’s comics industry.However,the problem of copyright infringement of comics has also gradually become a major obstacle to their future development.The current foundation of China’s law has many theoretical flaws,and the concealment of forms of infringement and the low cost of breach of contract are important reasons for this situation being in a difficult position.Therefore,there is an urgent need to conduct an in-depth study on the issue of identifying comics copyright.The current protection of copyright for digital comics is not perfect,and it was controversial before whether the comic character modeling could constitute an independent work and thus obtain the protection of copyright law.Plagiarism is both a serious violation of their copyright rights and a kind of behavior that is difficult to be recognized in the trial and enforcement practice of copyright.The recognition of plagiarism should usually be formally distinguished from similar acts: plagiarism and use of opinions,ideas and views of copyrighted works;casual plagiarism and use of background,statistical information and objective facts within the works of others;plagiarism and fair use;plagiarism and coincidence,etc.Based on this,this dissertation proposes a recognition method based on the features of comic images.It mainly includes two parts: comic feature recognition based on scale invariant feature transformation and comic feature recognition method based on depth decomposition and hashing.These methods can not only compress the amount of feature information,but also retain the significant information in the image so that the recognition rate and matching performance can be improved.Firstly,starting from the cartoon character portrait,we perform pre-processing operations such as normalization,Radon transform and logarithmic polar transform on the cartoon portrait,so as to achieve image uniformity and facilitate the subsequent feature recognition and matching process.Secondly,the ASIFT feature extraction and matching method based on image localization and the portrait feature extraction technique based on multi-step decomposition of quaternion mixed domain are proposed.In the first method,the cartoon image is first processed with color feature descriptors and Canny edge detection method,and then the ASIFT algorithm is used to extract and match feature points;in the second method,in the RGB color space of the pre-processed cartoon image,the R,G and B components are used as coefficients of the quaternion matrix i,j and k,respectively,to make an independent color part Y,and then the calculated luminance values are added to the matrix after The weighted quaternion is performed,and the quaternion Fourier phase spectrum significance detection is performed on it.Finally,the transformed images are compared for hash similarity.Finally,the analysis compares the feature recognition method based on ASIFT and the feature extraction method based on multi-step decomposition of quaternion mixed domain,summarizes the range values for judging whether there is a possibility of plagiarism in cartoons by the combination of the two methods,and formulates a complementary judgment standard.The experimental results show that the feature recognition method based on ASIFT and the feature recognition method based on depth decomposition and hash algorithm proposed in this paper can play a good matching role in matching comic images,and the accuracy of matching similar parts can be greatly improved compared with similar image recognition methods,and the mean hash can reach more than 60%,the difference hash more than 50%,and the perceptual hash in terms of similarity In terms of similarity,the mean hash can reach more than 60%,the difference hash more than 50%,the perceptual hash more than 60%,and the image anti-geometric transformation is also improved.