Research On Image Source Identification Algorithm Based On Mixed Feature Extraction And Improved Support Vector Machine Model

With the popularity of digital cameras and digital technology, the digital image has been widely used in daily life and work.Accordingly,tampering with the contents of the image also becomes more and more frequent and easy.The resulting impact might interference people’s normal life,or even harm national, social and political stability.Therefore,identification of the authenticity of the image is becoming increasingly urgent,blind image forensics was mentioned as a study point and became one of the hot research nowadays.Blind image forensics mainly involves four aspects, one of which is how to confirm the image is a natural image from the camera,mobile phones and other equipment taken,or an image through computer-generated,but also an image directly generated by a scanner. The traditional image source identification algorithm include feature extraction and classification structure and many other technologies.In this paper,feature extraction and classification structure in the process of image source identification were configured as the research object,the concept of hybrid feature extraction was presented and the SVM model was improved.Finally,combined with experiments,the research applied to the image source identification system,the results proved that the proposed hybrid feature extraction and improved SVM model greatly improved the image source identification rate.Specific research work is as follows:(1) Given the high number of dimensions could lead to "the curse of dimensionality" problem,the traditional image source identification algorithm mostly use the strategy of single feature extraction.However,the content of the image is very rich,single feature basically can not generalize the characteristics of the image,which would likely lead to the final identification errors. In response to this problem, we proposed the concept of hybrid feature extraction,extracted the color features,texture features and statistical features,then performed dimensionality reduction and feature normalization operation.Finally,we also designed an experiment to verify the effect of the hybrid feature extraction.(2) In the final classification stage of image sources identification,existing methods are mostly using the existing SVM models.These models can achieve good results in some problems,but whether it is suitable for the image source identification problem has not been verified.Since the kernel function and kernel parameters of SVM model have a critical influence for classification effect,we verified the kernel function and kernel parameters foror the image source identification problem to improve the SVM model,selected the kernel function and kernel parameters that most suitable for image source identification issues.Finally,we designed an experiment to verify the effect of the improved SVM model by using the selected kernel function and kernel parameters.(3) In addition,we applied the research results to a specific image sources identification system and achieved good effect.To better illustrate the effect of the two improvements in this paper for image sources identification issues,we designed a comprehensive comparative experiment through the system.The experiment compared the identification effct of the traditional image sources identification algorithm,the algorithm using only mixed feature extraction,the algorithm using only the improved SVM model and the algorithm that using the mixed feature extraction and the improved SVM model at the same time.As described above,for all-round, multi-angle to verify the research,we designed three experiments.From the experimental results obtained the following conclusions:Firstly,under the same conditions,compared to the single feature extraction,the hybrid feature extraction can achieve better identification accuracy.Secondly,under the same conditions,compared to the corresponding parameters of the existing SVM models,the selected kernel function and kernel parameters can achieve better identification accuracy.Lastly,using the mixed feature extraction and the improved SVM model at the same time achieved better identification accuracy in all algorithms, fully verified the correctness of the concept in this paper.In summary,our research on image sources identification achieved remarkable results,greatly improved the accuracy of image sources identification.