Research On Image Tamper Recognition Technology On Deep Learning

As one of the most commonly used information carriers in People’s Daily life,digital image plays a very important role in daily information exchange.However,with the rapid popularization of cameras and photo editing software,ordinary people can manipulate the contents of the images at will.Nowadays,these tampered images will be spread more quickly and widely by the Internet,which not only has a great impact on People’s Daily life order,but also has serious adverse consequences in the media,politics and even military fields.At present,the proposed digital image tampering recognition methods can be divided into the traditional feature extraction method and the deep learning method.Recognition methods based on traditional feature extraction are usually effective only for a specific type of tampering and have limited adaptability in the current big data environment.A new image tampering recognition method based on deep learning is proposed to solve the problems of traditional image tampering recognition methods.The main work content is as follows:The design of deep neural network structure,loss function and model optimization methods are analyzed and summarized,and the traditional image tamper recognition technology based on single feature extraction is introduced,which provides a solid theoretical basis for the later experimental part.The mainstream tamper image data set is integrated,from which the images suitable for neural network training are selected,and then the image annotation tool is used to label the tamper images.Because of the excellent performance of Faster RCNN in the field of target detection,an image tampering recognition method is proposed by adding multi-level feature fusion technology to Faster RCNN.The improved network structure integrates multi-level feature information.After the extracted features pass through the RPN layer,candidate boxes of tampering positions are generated for future classification and regression operation.Experimental results show that the proposed method based on feature fusion can identify various tampering types,The recognition rate is higher than three traditional algorithms and two deep learning algorithms,with an average recognition accuracy of50.56%,and tamper areas can be located.In order to improve the recognition efficiency of tamper images,a popular singlestage target detection algorithm,YOLOv5 s models,was combined with the attention mechanism.The single-stage target detection algorithm has the advantage of recognition speed,and the spatial attention mechanism and channel attention mechanism are combined to improve the recognition accuracy,with the average recognition accuracy reaching 53.92%.The recognition speed is up to 13.89 images per second,and the experimental results show that the method can accurately and quickly identify the tampering traces of digital images.