Research On The Attention-Guided Logistics Documents Image Splicing Tampering Detection Algorithm

With the rise of various image processing software,people can modify digital image content wantonly.Nowdays,as an information carrier digital images have played a pivotal role in logistics engineering and other fields,and a large number of tampered images will have a serious negative impact on daily work and social stability.Most of the existing image forensics algorithms directly use convolutional neural networks to extract image features and then classify them.Those algorithms do not take advantage of key features such as the small proportion of the splicing area of the logistics document tampering pictures and the difference in the splicing edges.Therefore,there are weak model generalization capabilities and poor detection effects in the logistics document tampering detection task.This paper proposes an attention-guided splicing tampering detection algorithm based on the above problems,so that the focus of the algorithm model is focused on the key features in the above-mentioned logistics document images,so as to reduce the false call of the detection task and improve the generalization ability of the model.The specific research results are as follows:(1)A classification and positioning model based on self-attention mechanism is proposed.This paper calculates the correlation of various features including image statistical features,noise residuals and spatial information to build an attention weight matrix to filter feature vectors.On the one hand,this model can improve the inherent relevance of feature vectors,and on the other hand,it can focus on regional differences and edge features while ignoring irrelevant information such as image content,color,and texture.Finally,this paper sets up multiple sets of experiments on the logistics document image tampering data set to compare with several mainstream tampering detection algorithms.The experimental results show that the classification accuracy and recall rate of the model in this paper have reached 95.8%and 82.9%respectively.Compared with the basic network,this method improves the accuracy rate by 7.3%and the recall rate by 11.2%.It slows down the missed calls and false calls of traditional tampering detection algorithms in the task of detecting tampered images of logistics documents.(2)An image preprocessing method based on noise residuals is proposed.By constraining the weight initialization and update strategy of the convolutional network,it has the ability to extract the noise residuals at the edge of the target area and can target different task scenarios and can adaptively modify it for different task scenarios.At the same time,the multi-scale feature extraction module is used to extract the pre-processed image in parallel,which improves the detection accuracy of small target regions.Finally,the deep convolutional network is used to iteratively extract the feature map,and the pixel distribution characteristics and edge features of each target area are integrated into the feature map,which improves the information richness of the feature map.It is verified through experiments that this method can increase the recall rate of 4.26%and the accuracy rate of 3.79%on the experimental results of the benchmark network.(3)Finally,this paper have carried out engineering processing on the algorithm model,and built an image splicing tampering detection system based on Python programming language and Django web application framework which makes the research of this article have practical application significance and provide positive help for the secure transmission of information in the field of logistics engineering.