JPEG image resampling automatic detection method based on deep random forest

Abstract

The invention belongs to the technical field of information safety and discloses a JPEG image resampling automatic detection method based on a deep random forest. The method comprises extracting a feature vector describing the recompression characteristic of a JPEG image: describing texture features presented by a local periodic correlation affected by a resampling operation; describing the Benford features of differences affected by the resampling operations in R, G, and B channels; describing an adjacent coefficient difference feature of a relationship between a DCT coefficient affected by the resampling operation and its surrounding coefficient; describing a blocking effect feature that appears after the JPEG image resampling. The method fuses the four feature vectors by using typical correlation analysis, learns and detects the fuses feature vectors by using the deep random forest, effectively recognizes the image resampling operation, and quantifies the relationship between the four feature vectors to greatly reduce the number of feature dimensionality, reduce the amount of calculation, improve the correlation between features, and enhance the detection accuracy.

Description

technical field [0001] The invention belongs to the technical fields of information security, pattern recognition and digital image processing, and in particular relates to a JPEG image resampling automatic detection method based on deep random forest. Background technique [0002] At present, the existing technologies commonly used in the industry are as follows: [0003] Detection of resampling operations is an auxiliary forensic tool for recovering image processing history. When a tamperer stitches two or more images together, geometric transformation operations (such as scaling, rotation, or skewing) are almost always required in order to create a “seamless” fake image. The process of geometric resampling transformation usually requires two steps of resampling and interpolation, so it can be considered to be able to detect traces of resampling to identify whether a given image or some parts thereof have been resampled. Second, resampling detection methods are crucial ...

AI Technical Summary

Problems solved by technology

[0012] (1) At present, there are many kinds of digital image resampling forensics methods, and the forensics process also has its own emphasis. The development of diversity has also led to the increase of non-standard methods in the field of resampling forensics, wasting a lot of unnecessary manpower and material resources. To judge the correctness and effectiveness of the method, there is an urgent need for a unified forensics framework to standardize the process of resampling forensics

[0013] (2) Most of the existing detection methods only use one feature for forensics. The feature cannot fully describe the characteristics of the image after resampling, so there are many disadvantages. For example, the method based on the EM algorithm is obviously dependent on the initial value, and many algorithms It is not sensitive to detection when the sampling factor is around 1.0, etc.

[0014] (3) None of the above methods perform fusion or dimensionality reduction operations on features, and there are a lot of redundant information and irrelevant information at the image feature level

[0016] (1) Researchers have proposed many methods in the field of digital image resampling forensics technology. It is necessary to scientifically classify and summarize the existing methods, and verify the effectiveness of the methods through experiments according to specific methods, which requires a huge workload.

Images