Semantic Segmentation for Visually Adverse Images - A Critical Review
Semantic Segmentation is one of the high-end visual tasks that has remained a topic of interest in various domains. Segmentation of visual scenes was confined to the extraction of object boundaries present in the image data. However, with the progressive developments in technology, machines are expe...
Saved in:
Main Authors: | , , |
---|---|
Format: | Conference or Workshop Item |
Published: |
Institute of Electrical and Electronics Engineers Inc.
2020
|
Online Access: | https://www.scopus.com/inward/record.uri?eid=2-s2.0-85097545728&doi=10.1109%2fICCI51257.2020.9247758&partnerID=40&md5=d441650260e9bbf18af89a522320f07f http://eprints.utp.edu.my/29855/ |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Semantic Segmentation is one of the high-end visual tasks that has remained a topic of interest in various domains. Segmentation of visual scenes was confined to the extraction of object boundaries present in the image data. However, with the progressive developments in technology, machines are expected to produce assistive decisions to aid versatile tasks. Subsequently, these assistive decisions are dependent on efficient results and must project information on a granular level from the visual scenes. The visual scenes are usually of vast variety depending on the scenarios in which the image data is captured. As per recent trends, semantic segmentation is still an open area of research, one of its worth mentioning challenges is to handle the visually adverse images. These visually adverse images are the result of low light/ high light, rain, fog and sometimes in the form of too many objects present in the scene. The study sheds light on the non-trivial problem and diverts attention to the gaps present in literature by providing in-depth critical analysis. This study comprehensively presents unidentified problems prevailing in existing semantic segmentation techniques. A critical literary study is conducted to examine the working mechanics of existing solutions to identify their limitations to produce accurate results for the visually adverse scenarios. The study discusses some of the possible reasons which result in erroneous semantic segmentation results for visually adverse images. Finally, the problems and challenges to be tackled are concluded which highlight the future direction of analysis. © 2020 IEEE. |
---|