Research On Image Splicing Detection Based On Composite Texture Features

Recently,digital images are widely used as factual evidence to depict events.Because of the dominance of computers in business,education and other industries,digital images are frequently regarded as authoritative evidence.At the same time,with the development of various software tools and low-cost hardware,image tampering has also become very accessible and virtually invisible to the naked eyes.Therefore,we cannot fully trust the authenticity and integrity of digital images,and it is urgent to explore more advanced detection methods to deal with the challenges brought by digital image tampering to the forensics field.In recent years,many algorithms for detecting splicing images have been proposed by scholars at home and abroad,and good results have been achieved,but the robustness of the algorithm and the accuracy of detection still need to be improved.In order to solve the problem that traditional image splicing detection algorithms are not robust to image content and illumination change,this paper proposes an image splicing detection algorithm based on composite texture features is proposed.The specific research content is as follows:1.Image feature description based on multi-scale and multi-direction decomposition.In this paper,a non-down-sampling contour wave transform(NSCT)is performed on twodimensional gray images to obtain a series of sub-band images containing image texture features(including 1 low-frequency sub-band image and 4 high-frequency sub-band images obtained through binary tree decomposition).Then the low-frequency sub-band image is processed by high-pass filtering and four high-frequency sub-band images,respectively,to obtain the webber local descriptor(WLD)texture features and local three-valued mode(LTP)texture features.2.Extraction and combination of multiple texture featuresIn this paper,based on the classical image splicing detection algorithm,the feature extraction algorithm combining local binary mode(LBP)and gray co-occurrence matrix is improved.On the one hand,in the description of texture features,a feature extraction algorithm combining local three-valued mode(LTP)and gray level co-occurrence matrix is proposed to obtain the gray level variation features of adjacent pixel points without noise and under the influence of light.On this basis,another new idea is proposed,which combines WLD texture with gray scale symbiosis matrix to obtain the characteristics of intensity and direction change of gradient of adjacent pixels.On the other hand,in terms of feature composition,this paper successively calculates the LTP value of two adjacent pixel points in the image,and takes the adjacent LTP value as the data pair to calculate the symbiosis matrix,and finds out the relationship between the gray level changes of the adjacent pixel points,so as to detect tiny splicing traces.At the same time,WLD differential excitation and direction feature are added in this paper.By calculating the co-occurrence matrix of adjacent WLD characteristic values,the intensity and direction change features of image pixel points are fully described,so as to accurately capture the image stitching traces.At last,the five features of WLD value cooccurrence matrix and LTP value co-occurrence matrix,including contrast,correlation,dissimilarity,entropy and energy,were extracted and fused into feature vectors,and RBF neural network was used for classification.For each image,with 50 d feature vectors to describe,at last in Colombia color splicing dataset,180 real images and 183 splicing images are divided into training set and test set,and for the new composite texture feature using RBF neural network to distinguish whether for splicing image classification,detection accuracy reached 95.41%.