Research On Median Filter Detection In Digital Forensics

With the rapid development of information technology,some powerful image editing tools such as Photoshop and Mei Tu Xiu Xiu have sprang up.However,by using these tools,criminals can easily tamper with images content to achieve illegal purposes,leading to a negative impact on individuals and even the whole society.Therefore,it is of great significance to judge the authenticity of image content.As a simple nonlinear operation,median filtering can keep the image content unchanged when removing noise,thus often be used by tamperers to eliminate the visual distortion and statistical distortion caused by tampering image to achieve the anti-forensics purpose.To reveal the history of image tampering,many researches have been made on median filtering detection.In addition,it is noted that a two-dimensional median filter can be approximated by a combination of a one-dimensional median filter.The study of one-dimensional median filtering forensics is enlightening for the two-dimensional median filtering forensics.In this paper,for digital forensics,we studied the median filtering forensics of one-dimensional signals and digital images,and the main works are as follows.(1)Study on median filtering forensics theory.To reveal the specific reasons for the stability structure generated by the median filter,two types of point relations,i.e.,sharp-point and flat-point,are defined in this paper.It is found that sharp-point will become flat-point and flat-point will not change after median filtering.The theoretical proof for this is also given.Some results such as the stability structure is composed of non sharp-point and the maximum number of times the median filter affects the signal are obtained.(2)Study on median filtering forensics.According to the mathematical statistical law presented by the number of flat- point of the signal with the increase of filtering times,this paper proposes a one-dimensional median filtering detection method based on discrete co-efficients.This method can determine whether a signal has passed the median filtering operation through a simple threshold.Inspired by the characteristics of one-dimensional stability structure,the flat-point-structure and sharp-point-structure for two-dimensional signals are further defined.In order to reflect the stability structure characteristics caused by the two-dimensional median filtering,the flat-point-structure is normalized to the nor-malized frequency characteristics.Then,by combining with block accumulation feature to resist JPEG compression.Finally,a nine-dimensional feature is generated as the pro-posed median filter detector.The method can also detect whether the local area of the large image has been tampered with by the median filter.(3)Online platform for image tamper detection.The algorithm proposed in this paper and some current mainstream image passive forensics algorithms are integrated into the platform.The platform can be mainly divided into four modules: user interaction mod-ule,business function module,database management module,and security management module.It has high practicability.