Information Technology Reference
In-Depth Information
Fig. 5
Display format of Simultaneous Double Stimulus
The reference image is always shown on the left side of the image and the dis-
torted image is shown on the right side. Test subjects grade the quality of the dis-
torted image on the right hand side by comparing it to the reference image on the
left.
The quality scale is the tool that the human subjects utilize to judge and to re-
port on the quality of the tested images. One of the most popular quality scales in
the subjective quality assessment research field is the 5 point quality level. Here, a
10 point quality scale was chosen, because the pilot test had shown that eight dif-
ferent quality levels could be clearly differentiated. Also, selecting a finer scale
seemed to be advantageous due to the higher quality of test images used, in which
a finer differentiating quality is suitable. The test used a discrete quality grading
scale, which implies that the subjects are forced to choose one of the ten values
and nothing in between. The quality grading scale, which is illustrated in Fig. 6,
refers to “how good the picture is”.
The test was conducted as a single session. Each of the 48 processed images
and the 6 references were presented for a period of 10 seconds; subjects evaluate
each presented image once. Subjects then needed to vote on their questionnaire
sheet before the next image was presented, and they were given 5 seconds to cast
their vote. The presentation structure of the test is illustrated in Fig. 6. The total
session length was 15 minutes. Prior to the main session, a training session was
conducted. Subjects were informed about the procedure of the test, how to use the
quality grading scale, and the meaning of the designated English term related to
the distortion scale of the image. During the training session, a short pre-session
was run in which 19 images were shown to illustrate the range of distortions to be
expected. The order of the main session was randomized, meaning that the six im-
ages and eight processing levels were randomized completely. Four to five sub-
jects participated at the same time, and six such rounds were needed to include all
subjects (see next section). The images presentation orders for each six rounds
were different.
