Non-regular missing image restoration method based on mathematical morphology of localized multi-structural elements

A single-image restoration method that does not rely on pre-trained models is proposed to address artefacts, structural distortion, and texture blur commonly occurring in the restoration of irregularly missing image regions. The method first optimizes multi-threshold segmentation using a genetic algorithm, enhances brightness channels through image partitioning, and improves detail visibility in low-light areas; secondly, local multi-structural elements (0 °, 45 °, 90 °, 135 ° directions) are constructed, and multi-scale edge detection is performed using mathematical morphology's dilation and erosion operations to accurately extract image structural information. Finally, a matching pursuit algorithm based on sparse representation of similar structures is adopted to fill missing areas pixel by pixel along the iso-illuminance direction, enabling collaborative restoration of irregular contours and textures. Experiments on geometric image datasets with complex structures and diverse missing shapes show that the proposed method outperforms comparison methods in terms of PSNR and SSIM, without requiring large-scale training data. The repair effect is not affected by the statistical distribution of missing area size and shape.