Development of image analysis application processes: Completion, forgery and enhancement

Nowadays, with the widespread use of image processing and analysis techniques, the risk of confusing manipulated content with true and misleading content has increased. This situation can lead to significant harm on social, political, and personal levels, as it perpetuates the spread of fake news and misleading video and audio recordings. The malicious use of deep editing technology, one of the most effective image processing techniques in our lives, can lead to serious issues, including privacy concerns. Therefore, the development and correct use of protective measures such as deep editing modifications are extremely important. Artificial intelligence and machine development techniques detect deep editing errors and manipulations. This detection employs innovations like deep learning software, continuous analysis, and specialized applications. Simultaneously, image completion and merging technologies, which account for a significant portion of deep editing processes, have become an integral part of our daily lives. This technology serves a diverse range of users, including cyber security experts, criminal detection teams, Photoshop experts, and social media users. There are many artificial intelligence-supported image completion and deletion systems in this intended field of computer science. In addition to image completion and deletion techniques, there are also image fusion techniques that offer significant advantages. Deep learning models can retune and select low values or corrupted data. High accuracy rates have been achieved on the artificial intelligence methods that present this thesis proposal: CelebA-HQ for image completion, FaceForencisc++ for forgery errors, and FlickR2K data modules for image fusion. The results obtained from the improvements made to the CelebA-HQ dataset for image completion show that the detail image completion technology works efficiently. Depending on the mask size of the imaging processes, the training and testing results showed that the best PSNR value was 35.25 and the best SSIM value was 0.96. Recording for fraud: The FaceForencisc++ dataset yielded an accuracy rate of 98.82%. The configuration on the FlickR2K dataset for image studies yielded 29.16 PSNR and 0.95 SSIM values. Thus, in-depth editing of simulation applications, technologies used for image completion and image merging, and artificial intelligence are successfully combined, and effective results are achieved. These results show that possible artificial intelligence models are effective in predicting, creating, and analyzing deep editing techniques.