Single Image Super Resolution Approach to the Signatures and Symbols Hidden in Buddhist Manuscript Sutras Written in Gold and Silver Inks on Indigo-Dyed Papers

poster / demo / art installation
  1. 1. Toshiaki Aida

    Okayama University

  2. 2. Aiko Aida

    National Museum of Japanese History - Ritsumeikan University

Work text
This plain text was ingested for the purpose of full-text search, not to preserve original formatting or readability. For the most complete copy, refer to the original conference program.

Image super-resolution can obtain a high-resolution image from multiple low-resolution ones. Conventional image super-resolution is realized by summing the information included in low-resolution images of the same scene. Therefore, it has the advantage of exact replication of details in the resulting high-resolution image from the original, although it necessarily requires many low-resolution images with various shifts of subpixel order. However, single-image super-resolution allows the inference of high-resolution images from relations to their low-resolution ones. Following the method of Yang et al., we determined the relations among more than 100,000 image patches (Yang et al., 2010). More concretely, various high-resolution images were first prepared and transformed to low-resolution ones by blurring and down-sampling to one-third of their original sizes for threefold super-resolution. Thus, we obtained pairs of high- and low-resolution images. Then, corresponding 9 × 9 and 3 × 3 image patches were extracted respectively from them and combined into coupled image patches. The method of Yang et al. enables construction of their basis vectors, called a coupled dictionary, under which the corresponding high- and low-resolution image patches can be represented by their common coefficient vectors. The coupled dictionary is the essence of the relation between low- and high-resolution images. With the obtained dictionary and a low-resolution image, the estimation of a coefficient vector for each 3 × 3 patch directly permitted inference of the corresponding 9 × 9 high-resolution patch. In this way, a high-resolution image can be inferred from a single low-resolution one, although the details of the resulting high-resolution image are not identical to those of the original.
In this presentation, we will report the results of our analysis of the signatures and symbols hidden in Buddhist manuscript sutras written in gold and silver inks on indigo-dyed papers during the late Heian period in Japan. Conventional image super-resolution requires original subjects of fine texture and cannot obtain clear images of cultural assets that have been damaged or experienced deterioration. Single-image super-resolution can overcome the above difficulties and assist in research of cultural assets, transforming their images into fine-textured ones. We applied super-resolution to low-resolution infrared images of Buddhist manuscript sutras written on indigo-dyed papers to facilitate the investigation of hidden signatures or symbols behind them. Because the purpose of our study is not archiving cultural assets but deciphering their meaning, the inference of details by single-image super-resolution can effectively increase visibility. Here, characters not originally present could possibly be reconstructed from low-resolution images, which could cause misinterpretation. Usually, the contribution from visual noise on paper is small-sized and high-frequency. Therefore, it can be easily removed by decreasing the number of dispatched basis vectors of the coupled dictionary in order to represent low-resolution image patches.
In East Asia since ancient times, respected Buddhist sutras have been written in gold and silver inks on deep-blue papers dyed with natural indigo. Digital infrared cameras have detected various characters, pictures, and signs under the dyes on some deep-blue papers handed down in Japan. Researchers have recently proposed that some of these sutras were made using recycled papers, generally public documents.
We focus on black signatures and stumps (or drawings of stump-like designs) hidden on the blue papers on which Buddhist manuscript sutras were copied in gold and silver inks during the Heian period (794-1185) in Japan. We detected similar marks on indigo-dyed papers from several scrolls of one of the most famous Japanese Tripitaka, formerly called Jingoji-kyo, planned and made by the retired emperor Toba (1103-1156) at the contemporary capital city of Kyoto. Recent researches have shown that the Chusonji-kyo, another famous Japanese Tripitaka written on deep-blue papers, also bears many signatures and signs under the indigo dye. The marks on the Jingoji-kyo closely resemble those on the Chusonji-kyo, which was made by the Northern Fujiwara clan who had ruled northeast Japan. Therefore, we naturally presume that these hidden signatures and pictures are connected somehow with the central culture or public power. Furthermore, the hidden signatures and symbols detected by digital infrared cameras, including those on the Jingoji-kyo, are apparently limited to only decades in the Heian period. At this time, the production of luxury Japanese paper was entering a transition from centrally authorized paper (so-called Kamiya-gami) to local production. After attempting analysis and classification of these signatures and marks, we concluded that they indicate that some paper studios, aristocrats, or noble priests drew signatures on the papers to show possession.


Yang et al., (2010). Image Super-Resolution via Sparse Representation.
IEEE Transactions on Image Processing,
19(11): 2861–2873.

If this content appears in violation of your intellectual property rights, or you see errors or omissions, please reach out to Scott B. Weingart to discuss removing or amending the materials.