Research On Detection Of Content-Aware Image Resizing

In digital era,the rapid development and widespread use of image editing software enables people to change images simply and quickly.Thus it results in a large number of tampered images appearing in our lives.These perceptually realistic images not only affect people's cognitive of 'seeing is believing',but also bring great negative impacts on news,politics,culture and other fields.Therefore,carrying out the evidence reveal study to the fake or tampered images has an important significance for ensuring the authenticity of the news,maintaining judicial justice and ensuring the public trust order.In this thesis,we research on the detection method of digital images which are tampered by content-aware resizing operation.First of all,the thesis investigates the overseas and domestic research status for content-aware resizing.And then we deeply study the content-aware resizing technology.Two algorithms are proposed to detect the content-aware resizing operation.One algorithm based on the statistic characteristics of probability map,the other use the features of modified Markov.The primary work of this thesis includes:(1)For the problem of distinguishing whether the image is tampered by content-aware resizing operation or not,we propose a content-aware resizing detection method based on probability map of statistic features.By detecting the CFA(Color filter Array)characteristics of original images,content-aware resizing images and scaling images,respectively,we can draw a probability map.Then according to the difference between different types of map,four new statistical features are constructed to describe the testing image,and combine to classify images;(2)We put forward the high order markov features detection method to detect the content-aware resizing operation.Under the analysis of the relationship between pixels in the original image and the content-aware resizing image,this proposed method can better describe the relations among pixels and can distinguish whether the image undergoes the content-aware resizing manipulation.Therefore,we use higher order markov transaction probability features to detect and classify images more precisely.Experiments show that this method has good detection results.