Perceptual QA of SCIs

Instruction
Research on Screen Content Images (SCIs) becomes important 
as they are increasingly used in multi-device communication 
applications. SCIs 
render texts, graphics and natural pictures together, such as web pages, slide files and computer screens. Since it is significant to efficiently acquire, compress, store or transmit SCIs, numerous solutions have been proposed for processing SCIs, especially for SCIs compression. Lately, the MPEG/VCEG community also calls for proposals to efficiently compress screen content image/videos, which will be reported as an extension of the HEVC standard.

When processing SCIs, various distortions may be involved, such as blurring and compression artifacts. In real applications, specified objective metrics are more desired to predict quality of processed SCIs, based on which we can control the processing of SCIs more efficiently. Before releasing the objective metrics, we need to check whether these metrics are consistent with human visual perception when judging SCIs. Subjective quality scores which are directly given by human subjects should be offered as ground truths to realize the checking. Hence, it is meaningful to investigate the quality evaluation of distorted SCIs either in the subjective or objective manners.


Download
You can download the image database as well as the supporting file via the following links: Download SIQAD

 If you use these images in your research, we kindly ask that you reference this website and our paper listed below.

Huan Yang, Yuming Fang, Weisi Lin, Perceptual Quality Assessment of Screen Content Images, IEEE Transactions on Image Processing, Volume PP, Issue 99, pp: 1-1, 2015.


Reference Images
Twenty SCIs are selected to be the reference images, which are with high diversity in layout styles and contents. Some examples of the reference SCIs are shown in following. 

                                                      
                                                   


Distorted Images
Total 980 distorted images are generated based on the 20 reference SCIs. Seven distortion types, i.e., Gaussian Noise (GN), Gaussian Blurring (GB), Motion Blurring (MB), Contrast Change (CC), JPEG compression, JPEG2000 compression and Layer-Segmentation based Compression (LSC), are adopted to generate distorted images. Seven distortion levels are set to each distortion type. The detailed processes to generate these distorted images are given in the supporting file.