Resampling image detection method based on convolutional neural network

Abstract

The invention discloses a resampling image detection method based on a convolutional neural network. The method comprises the following steps: segmenting a detection image according to alpha, and obtaining a plurality of sub-images; normalizing the color channels of all the sub-images; constructing a resampling database according to the detection image; performing training optimization on the resampling database through a convolutional neural network; screening the normalized sub-images according to the optimized resampling database; the screened sub-images are judged through a threshold method, If the sub-images are not smaller than the threshold, it is determined that resampling operation exists in the detection image; If so, determining that no resampling operation exists in the detection image. According to the method, feature extraction is carried out through the convolutional network, an overall network structure is designed by using the thought of residual errors, traditional manual initialization parameter setting is abandoned, parameters in the convolutional neural network are optimized through a momentum technical method, a globally optimal detection effect is achieved, and the situation that the convolutional neural network is locally optimal is prevented.

Description

technical field [0001] The present invention relates to the field of image recognition, more specifically, to a resampled image detection method based on convolutional neural network. Background technique [0002] Digital image forensics technology is a category in the field of information security, and resampling detection is one of the important technologies for digital image forensics. Image tampering often involves several geometric transformations, usually sampling the original image into a new sampling grid, and usually resampling the original image using nearest neighbor interpolation, bilinear interpolation, and cubic interpolation. Therefore, resampling detection plays a key role in image tampering detection. [0003] In the traditional image resampling detection, the image resampling detection method is mainly aimed at the detection of resampling period. Given some known number of resampled image samples, it is possible to find the resampling period of these resa...

AI Technical Summary

Problems solved by technology

But the period introduced by the resampling operation depends on the resampling rate, so it is impossible to uniquely determine the mode of resampling

Under the influence of image noise, the image resampling period will be seriously disturbed, and the method of estimating the resampling period of the sample image based on the maximum expectation algorithm will not work

In addition, the use of the maximum expectation algorithm has certain limitations. The maximum expectation algorithm needs to manually set the initial parameters. The final detection result is very sensitive to the initial parameters, and the initial parameter selection directly affects the detection speed and whether the ideal detection effect can be achieved.

Images