Perceptual-Based Picture Quality Measurements
Following creation of the perceptual contrast difference map, the next step is to convert this information into Picture Quality Rating (PQR) and Difference Mean Opinion Score (DMOS) measurements, the same scoring systems used in subjective evaluations.
To compute objective picture quality measurements for each frame in the test video and the overall test video sequence, it is necessary to “pool” the individual perceptual contrast differences at every pixel. Potential pooling methods include averaging the perceptual contrast differences of the pixels in the video frame, or computing the root mean square (RMS) of these values. Experimental work in visual sciences has shown that pooling these values using a more generalized mean, often called the Minkowski metric, yields the best results for both PQR and DMOS evaluations.
Engineering, verification, and quality assurance teams can apply these objective picture quality measurements to a wide range of applications. Examples include designing and optimizing video codecs, qualifying video equipment for deployment, configuring video systems to achieve optimal quality with minimal bandwidth, and verifying quality of re-purposed video content.
Over a wide range of impairments and viewing conditions, the DMOS measurement helps these teams determine how differences between the reference and test videos affect subjective quality ratings. The PQR measurement helps these teams determine how much viewers will notice differences between the reference and test videos, especially in the critical case of high-quality video when differences are near the visibility threshold.
Conclusion
In most cases, engineering and quality assurance teams that need to assess picture quality cannot afford the time and expense of recruiting viewers, configuring tests, and conducting subjective assessments. They need fast, accurate and repeatable objective picture quality measurements. However, to be of much use objective measurements need to closely match subjective evaluations.
Viewers are not noise detectors. They will perceive some differences between the reference and test videos, but not others. Full-reference picture quality measurements based on detecting the noise differences alone between the reference and test videos fail to account for the characteristics of human perception and have broad limitations as a result.
Perceptual-based full-reference picture quality measurements take a new and different approach. Using a human vision system model, they directly compute the perceptual contrast differences between the reference and test videos. They use these perceptual contrast differences to produce results that match subjective viewers’ ratings of video quality, thus significantly reducing the need for subjective measurements and giving video professionals a powerful new tool for evaluating picture quality throughout design, production and transmission processes.
About the author:
Greg Hoffman is a product marketing manager for Tektronix Video Product Line in Beaverton Oregon, focused on the development of picture quality analysis products. He has previously been a product marketing manager for waveform monitors, RF monitoring products, and MPEG monitoring and measurement tools. In his 25-year tenure at Tektronix Hoffman has also held positions in engineering and engineering management in Tektronix’ advanced research group, and logic analyzer and oscilloscope product lines. He has an MS in Physics from the University of Utah and holds patents in the fields of color printing and instrument control software.
