User Tools

Site Tools


16.01 - Platform Invariant Low-level Image Processing

Project - Summary


  • To research novel Neural Network structures that outperform the state-of-the-art in low-level image processing problems.
  • To collect a database in order to enable evaluation of platform invariance.
  • To test the platform invariance of the developed Neural Network based image processing algorithms.

Project - Team

Team Member Role Email Phone Number Academic Site/IAB
Moncef Gabbouj PI +358 (400) 736613 Tampere University
Serkan Kiranyaz Co-PI Not available Not available Tampere University
Caglar Aytekin PhD Student Not available Not available Tampere University

Project - Impact and Uses/Benefits

A supervised camera invariant color constancy algorithm is immediately applicable to camera industry. with such method, the dependency on unsupervised methods somewhat general but inferior performance can be eliminated. Moreover, the need of training a machine for each specific camera is also unnecessary thanks to color-conversion approach. With this method, one can only train a color constancy algorithm for one camera and immediately apply this CNN for other methods. The only need is the color conversion pre-processing step which loads almost no computational complexity on top of the color constancy algorithm. Although since this method is also supervised, one needs the camera spectral sensitivity of the cameras used. This information is very easy to be obtained compared to the very heavy task of collecting a new database of each camera and training a network for each camera.

Project - Deep Dive

Project - Documents

FilenameFilesizeLast modified
16.01_year_5_poster.pdf291.8 KiB2019/08/22 13:21
16.01_year_5_presentation.pdf467.1 KiB2019/08/22 13:21
16.01_year_5_final_report.pdf832.7 KiB2019/08/22 10:47
projects/year5/16.01.txt · Last modified: 2019/08/22 10:47 by sally.johnson