 |
The Effects of Congruency Between Structural & Contextual Dominance in Image Processing |
|
The Effects of Congruency Between Structural & Contextual Dominance in Image Processing |
A total of 21 subjects were used in this study drawn from a graduate class in research design at Indiana University. 13 subjects across two experimental groups received events with the same sequence of images, the verbal context being the only difference between groups. 8 subjects made-up a third control group exposed to images only.
The verbal contextual dominance variable was defined as a spoken phrase immediately preceding exposure to an image. For each stimulus event, contextual dominance was derived from the nouns within each phrase and then given points that were interpreted in terms of the image. Due to the need to compare the verbal contextual variable to the image structural variable, the contextual dominance was described in terms of the referents appearing in the image sectors.
The image was described in terms of two criteria, structural dominance and degree of complexity. The first level of image description, or structural dominance, was defined in terms of prioritized sectors of the image as assigned by a panel of visual design experts.
The second level of analysis was that of complexity. Complexity was determined simply be tabulating the number of structural elements visible in the image, e.g. 10 lines, 20 circles, 6 shapes, etc. In this study images were grouped into simple, medium, and complex categories. The criteria for inclusion in the simple category was having less than 10 structural elements, in the medium category by having more than 10, but less than 50 elements, and in the complex category by having over 50 structural elements in the image.
Identifying the degree of congruency between two groups was a statistical process of shared dominance comparisons. Based on the Experts' rankings of structural dominance, phrases were constructed which either referred to this dominance or referred to less dominant sectors of the image. Congruency was defined as an independent variable with two conditions. A "Congruent Condition" existed when the contextual phrase that was heard referred to elements identified as part of the image structural dominance. The phrase which referred to less dominant elements within the image was assigned to the non-congruent condition.
Fifteen images were divided into three groups of five according to the above complexity criteria. Structural dominance was determined and validated by a panel of visual design experts. Two contextual phrases were written for each image with one phrase written to match the structural dominance of the image and the second phrase focusing on weaker structural elements of the image. A sample from each of the three levels of complexity follow:
 |
 |
 |
Subjects were asked to sit in a specialized "head-stabilization" chair (Fig. 5) A small light was place adjacent to their right eye and a video camera was trained and focused on the eyeball and a portion of the light. The strategy was to record the reflection of the light on the cornea of the subject relative to the stationary light bulb. Following a registration process, the stimulus materials were presented.
Fig 5
The videotape was subsequently played-back one frame at a time and the motion of the reflection of the light off of the subject's cornea was documented s being in one of nine sectors of the frame.
Contextual dominance was determined by identifying the nouns within the contextual phrases and assigning points in relation to:
The points accumulated for each noun were then assigned to the sectors of the image to which the noun referred. The sectors to which the nouns referred represented 100% of the contextual dominance and were represented in terms of the percentage which each sector claimed of the dominance.
A "Congruent Group" was defined experimentally as a group which received an audible contextual stimulus which contained a referent to a structurally dominant element within the image. The assignment of an experimental group to either the Congruent data group (CG-S) or the Non Congruent data group (NCG-S) was based on a strategic decision by the experimenter to match the contextual referents heard verbally with the structurally dominant attributes within the image.
The Degree of Congruency (°ree;CG) was a post-hoc statistic used to describe the relationship between data groups in terms of the overall similarity of their shared dominance. This statistic produced a number, from 0 to 100, with the most congruent condition being 100. This procedure was executed as follows (see Fig. 7):
Corresponding sectors within each of the two data groups being compared were sequentially added together producing a "Sector Sum". Also these same numbers were subtracted from each other to produce a "Sector Difference". The process was repeated for each sector as long as there were values present within the sectors being compared. The sector sums and the sector differences were totaled, then the difference between these totals was found, and this figure was divided by 2 to produce the final result. The formula for determining the Degree of Congruency (°CG) was as follows:
For example, comparing the EXP group with the CG-S group below:
38 +89=127 & 31+11=42 (Sector Sum)=127+42=169
89-38=51 & 31-11=20 (Sector Difference)=51+20=71
(169-71)÷2= 98÷2= 49 = °CG
Fig. 7
A similar approach was utilized to determine the structural effect. The sector of primary dominance for the Expert data group determined which sector of the Congruent or Non-Congruent data group represented the structural effect for each group.
The images were grouped into 3 levels of complexity -- Simple (images 1-5), Moderately Complex (images 6-10), and Complex (images 11-15). By plotting the contextual effects for each image, an examination of the interaction effects of complexity was undertaken.
http://silver.ucs.indiana.edu/~appelman/D_THREE.html
|
 |