| Today the amount of information is growing geometrically.In this era,as a carrier of information,digital image is closely related to our life.With the frequent occurrence of criminal cases such as the use of digital image fraud and the dissemination of illegal and criminal contents,how to protect the originality,authenticity and integrity of digital images has become a hotspot in the field of information security study.Image source identification technology is a key technology in the field of digital image forensics.By identifying the source of imaging device,the authenticity of the image in the content and source can be ensured.At the same time,digital image is a kind of important open source intelligence.The application of image source identification technology to police intelligence work can play an important role in deeply digging information clues and cracking down on digital crime.At present,the image source identification technology based on PRNU(Photo-Response Non-Uniformity,PRNU)device fingerprint has been widely studied and highly recognized,but in practical application,this kind of technology has some problems,such as low identification accuracy in low-resolution images,high computational complexity of high-dimensional PRNU device fingerprint,high storage cost and so on.Therefore,it is of great practical significance to study the image source identification technology with higher identification accuracy at the level of low-dimensional PRNU device fingerprint to meet the practical needs of police.In order to improve the performance of PRNU device fingerprint in image source identification,a PRNU device fingerprint purification scheme based on cross-matching sample training is proposed in this paper.First of all,based on the auto-encoder network(Auto-Encoder,AE),an improved deep stacked auto-encoder model(Stacked Auto-Encoder,SAE)is proposed.Then,a cross-matching sample training technique is designed to train the model,and the corresponding purified model of PRNU device fingerprint is obtained.Finally,the fingerprint of the PRNU device is inputted into the purified model that has been trained,and the PRNU device fingerprint is purified through the compression-decompression process of the model.The comparative experiments on Dresden image dataset show that the purified scheme in this paper can improve the accuracy of image source identification on three small size PRNU device fingerprints,and is better than similar algorithms.At the same time,it has a better effect of purification for PRNU device fingerprint extracted by different denoising algorithm.This paper designs the application process of image source identification technology based on purification of PRNU device fingerprint in police intelligence work.First of all,in view of new criminal cases in recent years,through the analysis of typical cases from three different aspects which are internet public opinion,internet pornography and information network fraud,it is concluded that there are key points for police intelligence departments in analyzing and judging such cases.Secondly,it is discussed that the image source identification technology based on purification of PRNU device fingerprint can provide technical support for police intelligence departments in image intelligence source verification,intelligence clue expansion and serial case analysis.Finally,the application process of this technology in corresponding cases is designed to guide the rapid application of police intelligence departments in police practice. |
