An Image Tampering Detection Algorithm of Qualification Certificate Based on CNN and SVM

Abstract

With the high speed development of digital image processing technology, the phenomenon of malicious tampering with certificate images is rampant. Facing the problems of difficult human eye recognition of tampering traces and poor detection and classification effect of CNN model, a tampering detection algorithm based on CNN combined with SVM is proposed. The algorithm first performs data preprocessing of image compression, difference, resise, and normalization on the image. Then it uses the CNN model to extract the tamper feature, finally completes the calculation and classification of the fully connected layer feature data through SVM, and builds a certified image CNN+SVM model with tamper detection capability. Experimental results show that comparing with the traditional human eye recognition method and the existing deep learning CNN model, the algorithm can reach a detection accuracy of 48.09%, 91.79 %, and 98.67% under the self-built data set of tampering with certificate images. The detection accuracy rate is higher than the previous two detection accuracy rates.