Continuous vs. discrete scale stereoscopic video subjective evaluation: case study on robust watermarking

This paper establishes the theoretical relation between the scores assigned by observers on continuous and discrete (q unevenly distributed classes) scales. The experimental results validate this theoretical relation and bring to light the impact of the semantic labels, generally associated to the quality levels, in the evaluation process. They consider image quality, depth perception and visual comfort according to a Single Stimulus Continuous Quality Evaluation procedure. The full reference content is represented by 2 hours 11 minutes of HD 3DTV content. The modified content (the content to be evaluated) is obtained by watermarking the latter. A total of 110 observers, randomly assigned to three independent panels (reference, cross-checking and a posteriori validation panels) are inquired. The statistical investigations include outlier detection, Chisquare goodness-on fit test, Gaussian mixture probability density function estimation, Student's t-test and binomial test.