Effects of Post-Processing on Stereo Video
September 22, 2011, user experience technical Summit, Hollywood—Sin Lin Wu from Arizona State University and Jorge Caviedes from Intel described the effects of post processing on stereo quality. All of the techniques that have been used and researched for years for 2-D are not all applicable to 3-D post production.
Usually, post production just works with straight forward modifications like correction, enhancement, and transitions in post production. Unfortunately, the changes for 2-D are not the same type and effectiveness as those for 3-D. Color and contrast differences matter much more for 3-D images.
An experiment, they compared viewers’ responses for the same image displayed in 2-D and 3-D. The display used active shutter glasses for the 3-D so you lose some brightness in viewing the 3-D images compared to the 2-D ones. The stereo view was displayed as separate left and right side-by-side frames in 1080P.
Their findings were that the image content has a major impact on the perceived quality. Artifacts are more noticeable in 3-D than in 2-D and any lettering in the images quickly becomes illegible in 3-D due to ringing. There is little difference between 2-D and 3-D in bright scenes in terms of perceived quality, but 3-D suffers in dark scenes.
They determined that the algorithms for 3-D need more enhancements in the settings with a much lower thresholds for artifacts. The 3-D effects are greatly reduced for low intensity scenes. If the tasks change while viewing, the image transforms need different quality levels. Ringing of the images is not correlated with the left-right frames and the editor must correct for the image compression noise.
Editors need to look at the effects of the video processing to check for issues with simultaneous L/R, separate L/R, 2-D and depth. If there is a large difference between left and right images, you need to know if this can be adjusted at the display or needs to be corrected before release. The industry needs some type of control references for standardization in these issues. Testing is harder with live test people because the issues are relative and subjective, so most experiments are only able to check for deltas.
