A blind image inpainting method and system based on self-learning features and hierarchical similarity

Abstract

The application discloses a blind image restoration method and system based on self-learning features and hierarchical similarity, and belongs to the field of blind image restoration. The method solves the problem that existing blind image restoration processes still introduce human intervention mechanism to different degrees and fail to truly realize that a network autonomously learns and restores required features from a damaged image. The method comprises the following steps: automatically extracting semantic features from the damaged image through a self-learning module; introducing a hierarchical similarity block into the self-learning module; designing a self-learning semantic loss function to guide semantic feature extraction through a feature-level soft constraint; inputting the semantic features extracted by the self-learning module and the damaged image after splicing into a restoration module; and adopting a comprehensive loss function to jointly train the self-learning module and the restoration module. The application is used for image adaptive restoration in a pre-binding algorithm selection and image adaptive restoration difficult or unknown scene during damaged image restoration.

Images