SSID - Saarbrücken Synthetic Image Database

The Saarbrücken Synthetic Image Database (SSID) is an image database with images rendered synthetically from 3D sources. The goal of this database is to create a bases for image quality evaluation of images impaired by sources of error which can be found in 3D scenes.

Content

• Description of the Database
• Subjective Quality Evaluations
• MOS Survey
• Further Data
• Downloads
• References
• Contact

Description of the Database

Figure 1: Scenes contained in Image Database

The database contains 8 different scenes rendered from synthetic sources (see Figure 1). Each of the renderings is impaired by one of the following 7 error sources:

• Gaussian Blur
• JPEG Coding Artefacts
• JPEG 2000 Coding Artefacts
• Object Rotation
• Object Scaling
• Object Translation
• Gaussian White Noise

For each of the errors 30 error images with different error magnitudes are contained in the database. The amount of error as well as the reference image are stored in an info.txt-file per error. The structure of the database is given in Figure 2. Download links for the database are provided here.

Figure 2: Outline of Database Directories and Files

Subjective Quality Evaluations

Currently ~200 participants have done the MOS Survey, and ~17.000 opinions have been gathered. You can find an overview of the Mean Opinion Scores (MOS) per error here.

MOS Survey

Obtaining Mean Opinion Scores (MOS) for subjective image qualities requires a lot of user contributions. To further enhance the obtained scores we can never have enough contributions. The following link leads to a 15min survey of which the results are included in our research. We highly appreciate your participation!
Link to Survey

Further Data

For the scenes presented above additional data can be easily generated using Blender. Two scripts provided in the Download-Section allow the generation of depth maps and segmentation masks for the different scenes. Examples of these additional information channels are presented in Figure 3.

Figure 3: Original Image, Depth Map and Segmentation Mask

Downloads

All downloads are zip-packed and password protected. For the password please do not hesitate to contact us. We appreciate a brief information about your research interests with this database accompanied with your password request.

• Whole Database
• Database Parts
  • Reference Images
  • Gaussian Blur
  • JPEG Artefacts
  • JPEG 2000 Artefacts
  • Object Rotation
  • Object Scaling
  • Object Translation
  • White Noise
• Additional Data
  • Subjective Evaluation Scores
  • Original Scenes
  • Transformed Objects Only
• Blender Scripts
  • Object Transformations
  • Depth Maps and Segmentation Masks

Contact

Contributions to the database are very much welcome! If you would like to have a novel scene evaluated for the existing errors, would like a new kind of error to be added and evaluated, or have any other contributions please contact us! Comments, questions, or interest is also very welcome!
Christopher Haccius: haccius[at]nt[dot]uni-saarland[dot]de

Telecommunications Lab: www.nt.uni-saarland.de
Intel Visual Computing Institute: www.intel-vci.uni-saarland.de