Loading...

Detection and Localization of Image Splicing Manipulation by Deep Learning

Abdolrahimi Zarnagh, Ali | 2022

187 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55588 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Ghaemmaghami, Shahrokh
  7. Abstract:
  8. Today, with the increasing proliferation of digital tools, manipulating digital images has become a simple matter. Therefore, in many cases related to Forensics, issues related to the intellectual property, we need to verify the authenticity of the images. There are different types of image manipulation, but image splicing manipulation is the most frequent among the types of manipulations due to its simplicity and availability.In many applications, in addition to detection, the localization of the manipulated part, which is considered segmentation at the pixel level, is also of great importance.In this project, by using a structure based on deep encoder networks, a method for improving the detection and localization of image splicing manipulation is presented. To achieve this goal, maximum pooling alternative blocks, encoder with trained weights, atrous convolutional encoder and attention in the network have been used.In the designed network, the main goal is to use the details and residual history of tampering to detect tampering.For this purpose, the designed network is divided into three parts: feature extraction, feature fusion, and attention part, in each part, attention is paid to the details of feature maps. The emphasis of this research is to design a deep learning structure that, in addition to proper accuracy in detection, has fewer learning parameters than similar designs.The simulations show that the F1, Precision and Recall criteria of the presented method have been obtained as 91.7%, 93.7% and 90.5%, respectively, which has improved compared to the superior methods.In addition, the number of network parameters has reached 18 million parameters with a decrease of 8 million parameters compared to the superior methods.Also, the proposed method has shown good resistance to adding Gaussian noise to the input. So that in the peak signal-to-noise ratio of 19.98 dB, only 19% of the detection accuracy has been reduced.

  9. Keywords:
  10. Attention Mechanism ; Tamper Detection ; Autoencoder Neural Networks ; Deep Neural Networks ; Image Splicing Manipulation ; Atrous Convolution Layer ; Image Manipulation Localization

 Digital Object List

 Bookmark

No TOC