Digital Heritage Reconstruction Using
Deep Learning-Based Super-Resolution
Prathmesh R. Madhu and Manjunath V. Joshi
1 Introduction
Heritage and archaeological sites across the world are a major source of information
which acquaint us with our social history delineating the advancement of mankind.
They are invaluable assets of cultural heritage. Such places serve as an excellent
attraction for tourists which indirectly impacts the gross domestic product. This is one
of the major reasons for government agencies globally for taking keen interest toward
preserving these sites. Over a period of time, a number of natural calamities such
as weathering, man-made hazards like pollution etc., have sabotaged these heritage
sites. Fearing from any further damages to these sites by the tourists, access to many
heritage sites is now restricted. One such example is the mandapa with musical pillars
in Vithala temple at Hampi in India, where the visitors are not allowed to touch and
experience the melodious sound from the musical pillars. A way to preserve the
heritage sites is to physically renovate them. However, it requires a prolific amount
of historical information in order to renovate them. It also poses a danger to the
undamaged monuments and may fail to mimic the skill-full historical work.
One may overcome the above problems by using restoration in digital domain
An image can be acquired by sampling a continuous scene using a camera and the
restoration can be carried out by using in-painting and super-resolution methods. In
this work, we restrict our degradation in spatial domain only and propose a super-
resolution approach for spatial resolution enhancement. When the sampling rate of
the acquired scene is less than the Nyquist rate which is often the case with mobile
cameras used by the tourists, it results in aliasing effect leading to distortion in the
captured image. In addition to this, sensor point spread function and the motion
of the camera introduce the blur that degrades the quality of the image being cap-
tured. Image super-resolution is an algorithmic approach for increasing the spatial
P. R. Madhu (B ) · M. V. Joshi
Dhirubhai Ambani Institute of Information and Communication Technology,
Gandhinagar, India
e-mail: prmadhu@daiict.ac.in
M. V. Joshi
e-mail: mv_joshi@daiict.ac.in
© Springer Nature Singapore Pte Ltd. 2018
B. Chanda et al. (eds.), Heritage Preservation,
https://doi.org/10.1007/978-981-10-7221-5_4
67