摘要：Important structures in a large area of a damaged image cannot be satisfactorily repaired by traditional inpainting algorithms. Here, an image completion algorithm based on structure reconstruction and constraint (SRC) is presented to improve the structural coherence of the damaged image. First, the damaged structure of the target image is detected and located. Then, different missing structures are respectively reconstructed. The edge structures are reconstructed by the Euler spiral, which satisfies energy minimization. The corner structure is reconstructed using the intersection of two extended Euler spirals. Finally, the reconstructed structure is used as a constraint condition to modify the priority of the image completion and to guide the texture propagation within the damaged part. The proposed method thereby resolves the problem of the corner structure being unable to be adequately repaired for current image inpainting methods. In addition to preserving the structural continuity, it also effectively avoids texture inconsistency. Experimental results show that the peak-signal-to-noise-ratio values of images recovered by the proposed method increased 3.43 to 12.85 dB. Moreover, compared to the content-aware fill algorithm and Criminisi algorithm, its mean square values decrease by 42.16% to 94.61%. The structure consistency, neighbor texture information coherence, neighbor and visual effect are better than those of the other algorithms. The presented algorithm is thus suitable to repair minor damage, such as a straight line or curving scratch, as well as large area damage, such as object removal in nature scenes and cultural relic images.
