Research On Source Identification Technology For Digital Image

With the rapid development of computer software functions and the rapid development of multimedia networks,the phenomenon of digital image being forged and tampered with is common in today's society,and it has adversely affected many aspects such as public trust media,national politics,and legal evidence collection.It also jeopardizes the information security of citizens.Digital image forensics technology emerged in this environment.Digital image source identification is an important branch of digital image forensics technology.Its research purpose is to ensure the originality,integrity and security of digital images during storage and propagation.Different types of digital cameras generate different digital image features,so this feature is used as the basis for digital image source identification.The research work of this paper is as follows:(1)At present,digital image source identification technology is mostly based on supervised learning method,which uses a large number of camera model samples to train the classifier,so as to achieve the purpose of accurately classifying image samples.When samples of unknown camera models appear in the training samples,samples of the unknown models cannot be accurately classified because the lack of training samples of unknown camera models.Aiming at the problem of unknown model,a method of parameter optimization and extended training is proposed.Firstly,two indicators are proposed,and the optimal parameter K in the K-nearest neighbor algorithm is obtained through the correlation calculation model.Then,it is detected by the K-nearest neighbor algorithm to obtain a partial unknown model sample.In the extended training,the training set obtained by the K-nearest neighbor algorithm is used to train the multi-class support vector machine combined with the reduced clustering algorithm.The stopping rules of iterative training are also studied.The experimental results show that the proposed method outperforms the four commonly used methods.(2)The effect of using the local binary pattern feature as the image source identification task under the large sample condition is better,but the identification rate is not high under the small sample.Therefore,an ensemble learning method under a condition of small sample is proposed.Based on the idea that some features represent part of the sample,one sample is mapped into multiple samples.Then we use the ensemble learning method to classify samples.Base on the idea of selective integration method to add an optimization base classifier number step between the two steps of generating and integrating the base classifier,and solve the problem that the base classifier is too much and the calculation amount is redundant.Experiments show that the performance of the proposed method is better than the other four methods.