Identification of manipulation in digital images through hybrid block-based and key points algorithm

• Autores: Mohammad N. Hossain, Azam Bastan fard
• Localización: QUID: Investigación, Ciencia y Tecnología, ISSN-e 2462-9006, ISSN 1692-343X, Nº. 28, 2017, págs. 78-92
• Idioma: español
• Enlaces
  • Texto completo (pdf)
• Resumen

  • Nowadays, digital images in many legal centers are considered as a source of information and request to determine the authenticity of an image has increased dramatically. In this paper, an efficient algorithm to check and identify manipulation in which a combination of block-based methods and key points for the extraction of forged parts has been implemented. In the proposed algorithm, the input image is taken at First.  After compliance with the test target which is based on database, it is recognized that whether the image has been manipulated or not. In case of observing a positive result, it is concluded that that forgery has been made.  First, the input image is divided into irregular and non-overlapping blocks using simple clustering algorithm (SLIC)[1]. Then, feature points as the characteristics of the blocks are extracted using local binary method with several resolutions. Block attributes are adapted with each other to identify Areas suspected of forgery. In the second stage, for more accurate diagnosis of forging parts, characteristic points were replaced with small super-pixels as characteristic blocks and adjacent Features of blocks are replaced with the characteristics of positional color which are similar to feature blocks to produce consolidated areas. Finally, RANSAC[2]algorithm on integrated areas is used to remove false matches. Experimental results using a test database and forgery rotation methods, blurring, jpeg compression and etc., show that the proposed algorithm in the field of detection of copy-transfer forgery has reached to 97 percent and has also achieved recall rate of 98 percent.it has been improved   3 percent compared to other valid methods in terms of recalling and precision. This algorithm can even identify rotation methods, blur and jpeg compression by calculation which have less complexity.[1] Simple Linear Iterative Clustering[2] random sample consensus