13 September 2017 – A team of four from NUS Computing won the Best Paper Award last month, at the 17th International Conference on Computer Analysis of Images and Patterns (CAIP).
The paper, Multi-View Separation of Background and Reflection by Coupled Low-Rank Decomposition, was authored by Research Fellow Dr. Lai Jian, Associate Professor Leow Wee Kheng, Associate Professor Terence Sim, and Master’s graduate Mr. Li Guodong.
Explaining their research, Prof. Leow said, “Images captured by a camera through a piece of glass have a reflection superimposed on the transmitted background. It affects the image content and is difficult to remove. In this paper, we propose a multi-view method for separating the background and the reflection of the images.”
He added that their proposed multi-view method is the most convenient and their solution only requires five images of the desired background, captured at varying viewing angles. It solves the multi-view separation problem with two main components: alignment and recovery.
The suggested method uses a hybrid method for alignment which combines both rigid and non-linear registration, which is more flexible than rigid registration and overcomes the overfitting problem of non-linear registration. For recovery, the team modeled the background and reflection as low-rank data, and applied the Robust Principled Component method to separate them from noise data. “Our approach can be used to separate non-planar background scene and global reflection, which is a very difficult and challenging task,” said Prof Leow.
CAIP 2017 is part of a series of biennial conference devoted on all aspects of computer vision, image analysis and processing, pattern recognition, and related fields. This year’s conference was held in Ystad, Sweden, in late August.