A developed Criminisi algorithm based on particle swarm optimization (PSO-CA) for image inpainting

• Published in: The Journal of supercomputing 2024, Vol.80 (11), p.16611-16629
• Main Authors: Li, Fang-Fang, Zuo, Hui-Min, Jia, Ying-Hui, Qiu, Jun
• Format: Article
• Language: English
• Subjects: Algorithms; Compilers; Computer Science; Digital imaging; Interpreters; Matching; Optimization; Particle swarm optimization; Processor Architectures; Programming Languages; Searching; Signal to noise ratio
• ISSN: 0920-8542; 1573-0484
• DOI: 10.1007/s11227-024-06099-5

As a robust digital image inpainting technology, the Criminisi algorithm (CA) has been widely used. However, its high running time that it needs to search in the entire undamaged area of the image to determine an optimal matching block presents a challenge. To address this issue, this study proposes an improved version of CA, named PSO-CA, which incorporates the particle swarm optimization algorithm (PSO) with CA. The running time of the CA is significantly reduced benefiting from the parallel optimization capability of the PSO. In addition, the search space is restricted to the neighbouring region of the block that needs to be filled. The availability of the proposed PSO-CA algorithm is assessed in the laboratory colour model by the running time and three matching indices, such as the peak signal-to-noise ratio (PSNR). The experimental results indicate that PSO-CA significantly enhances the inpainting speed and produces the same or better results compared with the initial CA and the Criminisi with search space algorithm (CWSS).