M.A. Thesis:

Image inpainting based on deep semi-supervised deep learning learning method

Nowadays, due to the high importance of digital images and their use in the identification and authentication of individuals, it is necessary to create a clear, clear and unaltered image. Therefore, image preprocessing is one of the important steps in the field of image processing to remove image stains and noise and make it suitable. Some images undergo undesired changes and damage, and over time some parts of the image are destroyed or some unwanted objects are captured on the image. Images received from the satellite may contain incomplete information or may have been partially altered. This lost information in the image, which is not visible to the user, will be recoverable by image disinfection and repair techniques. Image reconstruction has many uses, such as scratches, redundant entries such as names and dates, and images with undesirable spots, as well as in cases where the eyes are red in the image. In noise removal and image restoration, the distorted area is approximated and estimated based on the spatial information of the original image area. In this study, semi-supervised in-depth learning method is used to improve the efficiency of the network education in question and can be used as background information or user information that will be semi-supervised learning. Learning is done from a set of similar images, which are considered as training sets. The sample image and a spot in the form of a mask are then converged to extract the affected area in the second section, then reconstruct the image based on the spatial information of the adjacent pixels.

Keywords: Image Inpainting - deep learning - semi-supervised -Neural Networks