Building a large-scale natural image quality database

Description

Image Quality Assessment (IQA) has attracted increasing attention due to a number of potential image processing applications.  The project objective is to build a large-scale natural image quality database. We intend to collect 10,000 images to make up the database, and all the images will be extracted from the YFCC100M dataset, which provides almost 100 million images and 800,000 videos. Some attributes will be considered to guarantee its diversity, including blur, sharpness, brightness, content, etc. To conduct subjective quality assessment on the large-scale image dataset, crowdsourcing will be adopted. 

Tasks

  • Prefiltering YFCC100M dataset.
  • Computing different attributes on the filtered dataset, uniform sampling 10,000 images using those attributes. 
  • Subjective study on the 10,000 images by crowdsouring.
  • Collect existing IQA algorithms and compare their performance on the database.

Prerequisites

  • Basic math skills are required.
  • Basic programming skills, e.g., Matlab, C/C++, or Python.
  • Some statistical knowledge is a plus.

References

  1. Winkler, Stefan. "Analysis of public image and video databases for quality assessment." IEEE Journal of Selected Topics in Signal Processing 6.6 (2012): 616-625.
  2. Ghadiyaram, Deepti, and Alan C. Bovik. "Massive online crowdsourced study of subjective and objective picture quality." IEEE Transactions on Image Processing 25.1 (2016): 372-387.