Loading...

Analysis of Blind Algorithms for Image and Video Quality Assessment

Otroshi Shahreza, Hatef | 2018

58 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 51754 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Behroozi, Hamid; Amini, Arash
  7. Abstract:
  8. The quality assessment of image and video data has became an important and challenging problem in the fields of image/video processing. Oftentimes, the images/videos files are designed to be viewed by a human observer. Further, the number and size of such files
    are growing everyday. Thus, computer-assisted monitoring of the quality has emerged as the only possible solution for this challenge. In this thesis, we shall first review the existing techniques for no-reference(blind) quality assessment of image and video data. Next, we propose some new approaches for the no-reference quality assessment of image and video data in-the-wild. In the case of images, our contributions consist of two methods based on convolution neural networks. In one of the proposed methods, a transfer learning approach was used. After training on KonIQ-10k Database the trained neural network achieved 0.70 for the Pearson Correlation Coefficient (PLCC) between the predicted quality and human opinion score on test data. This performance is competitive with previous methods which were not using deep neural networks. In the second method all the parameters of the neural network were updated during training, which achieved 0.81 for PLCC. Similarly, we introduce two feature-based methods for the case of video assessment; one with a traditional machine learning approach, and the other with using recurrent neural networks. These methods achieved PLCC of 0.73 and 0.76, respectively, between the predicted quality and human opinion score on test data of KonVid-1k Database. Note that this thesis is the first research in which the recurrent neural networks are used for quality assessment of video. Furthermore, a comprehensive study on the available databases for image/video quality assessment is provided. The result of this study can be used as a reference for the researchers in related fields
  9. Keywords:
  10. Quality Evaluation ; Image Quality ; Video Quality ; Deep Learning ; Machine Learning ; Database ; No-Reference (Blind)Algorithm

 Digital Object List

 Bookmark

...see more