M. J. Maher (Ed.): ASIAN 2004, LNCS 3321, pp. 248–258, 2004.
© Springer-Verlag Berlin Heidelberg 2004
A Novel Texture Synthesis Based Algorithm for
Object Removal in Photographs
Feng Tang, Yiting Ying, Jin Wang, and Qunsheng Peng
State Key Lab of CAD & CG, Zhejiang University,
310027 Hangzhou, P. R. China
{tang, ytying, jwang, peng}@cad.zju.edu.cn
http://www.cad.zju.edu.cn
Abstract. Natural images and photographs sometimes may contain stains or
undesired objects covering significant portions of the images. Inpainting is a
method to fill in such portions using the information from the remaining area of
the image. In this paper, we propose a novel photograph editing framework that
utilizes texture synthesis techniques. Major contributions of our algorithm are:
1) a constraint-based candidate patch searching method which limits the
searching within neighboring region with similar texture; 2) a metric of
Coherence Confidence for selecting the best fit candidate preventing error
accumulation and propagation; 3) integration of graphcut optimization to make
the seam visually invisible. Experiments show that our system can efficiently
handle different cases especially large regions in complex background.
1 Introduction
In practice, we often meet with such problems as old photos spoiled by ink or old
paintings full of scratches after long time reservation. Besides, photos may contain
undesired large objects to be removed, for example, a passing-by person may drop
into the view when a photo is taking. Although these two types of problems originate
from different scenarios, they are commonly known as inpainting. How to robustly
and efficiently solve both problems remains a challenge.
Several works have addressed these problems. For the first type of problems,
which contains scratches or small missing regions, there are diffusion based and filter
based methods. Bertalmio [1]-[2] filled in the tiny spoiled regions by propagating
information from the outside of the masked region along level lines (isophotes).
Reference [3] proposes Total Variational (TV) inpainting algorithm that invokes
Euler-Lagrange equation and applies anisotropic diffusion [5] inside the inpainting
domain based on the contrast of the isophotes. Oliveira [6] filtered the input image
using a Gaussian convolution kernel to remove the undesired flaws. Yet such
algorithms account for only small local regions. Artifacts of blurs may occur when
they are applied to large region of inpainting.
For the second type of problems, a kind of synthesis based methods has been
proposed. Liang et al. [7] composed the texture of a large region by selecting