SA-IQ
Please cite the following paper in any published work if you use this dataset.
E. Siahaan, A. Hanjalic, and J.A. Redi, "Augmenting Blind Image Quality Assessment using Image Semantics", in Multimedia (ISM), 2016 IEEE Int. Symposium on, pp.307-312, Dec 2016.
If you have any questions about the dataset, please feel free to email E.Siahaan@tudelft.nl.
Dataset Description
This dataset was constructed to analyze the influence of image semantic categories (scene categories and object categories) on visual Quality of Experience (QoE).
79 reference images were selected from the LabelMe image annotation dataset, such that they included a balanced number of different scene and object categories.
The scene categories include indoor, outdoor natural, and outdoor manmade. The object categories include animate objects (human and animals), and inanimate objects (natural and manmade objects other than humans and animals).
The reference images were impaired with Gaussian blur and JPEG compression, resulting in 158 images of medium and low quality for each of the impairment type. We conducted subjective experiments asking users to evaluate the visual quality of these images. The JPEG images were evaluated with a 5-point discrete ACR-labeled rating scale in a controlled laboratory setup, while the blur images were evaluated with a 100-point continuous ACR-labeled rating scale in a crowdsourcing setup.
We provide the image files, as well as the collected mean opinion scores (MOS), standard deviation of opinion scores (SOS), corresponding scene and object category labels, and impairment level of every image.
Download information
The database can be downloaded here. To obtain the password for accessing the files, you can contact Ernestasia Siahaan (E.Siahaan@tudelft.nl)
|
