Publications

Detailed Information

Constrained Optimization for Translucent Hindrance Removal from a Single Image : 단일 이미지에서 반투명 방해 요소를 제거하기 위한 제약 최적화

Cited 0 time in Web of Science Cited 0 time in Scopus
Authors

Tushar Sandhan

Advisor
Jin Young Choi
Major
공과대학 전기·컴퓨터공학부
Issue Date
2018-08
Publisher
서울대학교 대학원
Description
학위논문 (박사)-- 서울대학교 대학원 : 공과대학 전기·컴퓨터공학부, 2018. 8. Jin Young Choi.
Abstract
Interstellar, satellite and microscopic imaging methods produce high-resolution images which carry vital information even at a pixel level. Translucent hindrance from reflection, cloud or dusty layer is unavoidable in real-world imaging scenarios. It camouflages the vital image details by altering color and brightness. So the captured image consists of superposition of underlying true object (foreground) and the hindrance (background). In this work we have proposed the optimization model to unravel this superposition from a single input image without losing any image information.



We have identified and formulated three novel translucent hindrance removal tasks namely: eyeglass reflection, high-altitude cloud and microscopic dust removal from a single image. These hindrances seem to be unrelated each other, but after analyzing their characteristics in detail we have managed to relate them statistically. So we were able to develop unified constrained optimization framework, which is flexible enough

to handle all translucent hindrance removal tasks and can easily accommodate any new as well as our proposed image priors. Its Fourier transform and look-up table based iterative solution removes hindrances quickly without using any GPU. It can process high-resolution images and produces state-of-the-art results.



To the best of our knowledge, the proposed method is the first attempt on formulating and addressing all three important hindrances removal tasks. Our dataset images and source codes are available publicly. This solution has a vital importance in biometrics, satellite and microscopic imaging
Language
English
URI
https://hdl.handle.net/10371/143042
Files in This Item:
Appears in Collections:

Altmetrics

Item View & Download Count

  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Share