Our aim is to provide a comprehensive image dataset consisting of undistorted, naturally-blurred and artificially-blurred images for image quality assessment purposes.
The CERTH image blur dataset consists of 2450 digital images, 1850 out of which are photographs captured by various camera models in different shooting conditions that have not been altered in any way following their capture. The remaining 600 are artificially-blurred images. For their creation 60 undistorted images were randomly selected and then several types of Gaussian, motion and circular averaging filters were applied to them.
The CERTH image blur dataset is organized as follows:
630 undistorted images
220 naturally-blurred images
150 artificially-blurred images
Evaluation set consisting of the “natural blur” set and of the “artificial blur” set.
Natural blur set
589 undistorted images
411 naturally-blurred images
Artificial blur set
30 undistorted images
450 artificially-blurred images
Download the CERTH image blur dataset and the corresponding ground-truth here .
The images distributed as part of the CERTH image blur dataset were captured by members of the ITI team. The copyright of each image remains with the corresponding photographer. They are provided to the scientific community as part of the CERTH image blur dataset for use in scientific benchmarking and evaluation of research works.
If you use the CERTH image blur dataset in your research work, please cite the following paper:
E. Mavridaki, V. Mezaris, "No-Reference blur assessment in natural images using Fourier transform and spatial pyramids", Proc. IEEE International Conference on Image Processing (ICIP 2014), Paris, France, October 2014.