文物图像的超分辨率重建算法研究

Research on Super-Resolution Reconstruction Algorithm of Cultural Relic Images

文物的数字化保护与分类识别是当前图像处理研究的热点之一.针对常规超分辨率算法不能充分描述现实世界中文物图像复杂纹理结构的问题,本文提出一种基于回归环金字塔型生成对抗网络的文物图像超分辨率算法(Closed-loop Pyramid Information Generative Adversarial Network,CPIGAN).考虑文物图像的噪声等不定因素,本文采用不同的降采样方式构建了两种文物数据集且探索了一种改进信息块提取策略,提高了原始高分辨率文物图像中高频信息的利用率.本文进一步设计了一种金字塔型生成对抗网络并融入回归环结构,增强了网络从低分辨率图像到高分辨率图像映射的能力.基于自建文物图像数据集,本文算法与多种算法进行了实验对比分析,多个客观指标均有所提升且重建图像主观上更符合人类视觉标准.

The digital conservation and classification of cultural relics is one of the hotspots of current image processing research. In view of the problem that conventional super-resolution algorithms cannot fully describe the complex texture structure of cultural relics images in the real world, this paper proposes a super-resolution algorithm for cultural relics images based on a closed-loop pyramid information generative adversarial network(CPIGAN). Considering the noise of real cultural relics and other uncertain factors, this paper uses different down-sampling methods to construct two cultural relics datasets and explores an improved information block extraction. This strategy improves the utilization of high-frequency information in the original high-resolution cultural relic images. This paper further designs a pyramid-shaped generative confrontation network and incorporates the regression loop structure to enhance the network’s ability to map from low-resolution images to high-resolution images. Based on the self-built cultural relic image data set, the algorithm in this paper has been compared and analyzed experimentally with a variety of algorithms. Several objective indicators have been improved, and the reconstructed images are subjectively more in line with human visual standards.