As a first step towards objective quality estimation metric we identified the artefacts, which could occur in mobile 3DTV content. In our project we plan to perform subjective quality experiments, in which human observers would grade the perceptual quality of a variety of content. We created a framework which could introduce a set of stereoscopic artefacts to a given 3D video, thus ensuring repeatability of such experiments. The framework is freely downloadable in the “download” section of our website.
Using the framework, we created a database of impaired stereo-videos. In the database we included the original video stream for reference, and versions of the same stream impaired by various 3D artefacts with different level. We selected the artefacts which are most likely to affect stereoscopic video transmitted over a DVB-H channel. During the course of research on modelling the human visual system evolves, we might update the database to include more video streams with various levels of impairment. The video streams currently included in the database are described below.
The reference video used for creating our database is based on three sequences shot by KUK Filmproduktion – Flower3, Horse and Car. The original movies are available in the “stereo-video database” section of our website. The sequences were resized by FhG-HHI to resolution of 480x270px, suitable for mobile display. Finally, we padded the streams to height of 272px adding black pixels on top and bottom, and concatenated the three streams, resulting in the reference video used in our database.
We created a collection of impaired stereoscopic video streams, by processing the reference video throught our framework and using the following configuration file - process-all-v1.txt (text file, 1KB).
Crosstalk is an artefact caused by imperfect separation of video channels. Typically, it is introduced in the visualization stage. Most autostereoscopic displays exhibit pronounced crosstalk, caused either by their optical characteristics of due to the observer not being in the optimal observation position. In this video stream exhibits symmetric crosstalk with coefficient of 0.2 - the typical value for an autostereoscopic display using parallax barrier.
TThis video stream exhibits symmetric crosstalk with coefficient of 0.6. In this case the ghosting artefacts are quite pronounced and perceiving this video in 3D is barely possible.
Blocking artefacts are created when encoding video using block-based compression, which involves quantization of the results. Same level of DCT quantization might result in different perceptual quality, based on the depth cues present in a stereo image. In this video blocking is introduced by using 8x8 DCT compression with quantization coefficient q=50 for each frame separately. There is some “blockiness” in the video, but it is not very pronounced.
This video stream is processed with 8x8 DCT compression with q=15, which produces visible blocking artefacts in each channel. The “blockiness” iof this video is comparable the one exhibited by 2D video with “moderate” quality.
This stream has very pronounced blocking artefacts as a result from using DCT compression with q=10. For a 2D video stream, this would be the equivalent to “low” quality.
Some authors propose that blocking might be considered as several, visually separate artefacts. One of them is “block-edge discontinuities” which we simulate by introducing luminance distortions inside of a block while keeping the mean luminance constant. This video stream has artefacts introduced with a “strength” coefficient s=15, which results on block-edge discontinuities mostly visible in uniform areas.
This stream has block-edge artefacts introduced with “strength” coefficient s=30, which produces very noticeable artefacts, visible everywhere in the video.
Colour mismatch is most commonly caused by white balance done separately in each camera. In this video stream, we introduce colour mismatch between channels by mimicking automatic white balancing algorithms.
Colour bleeding is another artefact caused by block-based encoding. Since chrominance is typically sub-sampled, bleeding can occur beyond the range of a block. In this video stream we introduced colour bleeding by applying different levels of quantization to chrominance and luminance channels.
In DVB-H transmission most common are burst errors, which results in packet losses distributed in tight groups. The presence of artifacts depends very much on the coding algorithms used and how the decoder copes with the channel errors. We compressed the original stream using h.264 encoder, applied the DVB-H error patterns collected by our project and decompressed the video.