2.4 Materials/Procedures

In order to exercise the greatest possible control over the experiment, allmaterials were integrated into the program used for the experiment. Thisincludes:

Subject Questionnaire
Before the experiment begins, a form will be be given to the subject to fill out. It is assumed that the subject has sufficient computer experience to properly finish this task.

A screen describing the instructions will appear after the subject has completed the questionnaire. There will be three screens before the actual experiment begins:

1. Main Instruction screen (subject will see this prior to starting each task).
2. Practice screen (subject will only see this for the first time per widget type).
The subject will be presented instructions that pertain to the next task; the task's widget along with a dummy database will be presented. The purpose of this practice screen is to ensure that the subject understands how to use the widget before moving onto the experiment portion. After completing the practice task at least once, the subject may go on to the next stage at his/her discretion.
3. Task description screen
There are six unique tasks. The subject will be presented a text descriptionof a target image; the task is to find that image. Each text descriptionwill be unambiguous and have only one target. Specifically, eachtarget image has a particular object that is readily visible which does notappear in any other image in the database.The recorded time will be the duration between the moment when the subjectleaves this screen until the moment when proper image isselected.
User Satisfaction Survey
After the subject completes the current task, a screen will appear asking thesubject a number of questions. The user will then go onto the next of six tasks.

The image database used in the experiment was the Library Of Congress' Coolidge Collection, which consistsof 176 black and white photographs from the Coolidge era. The motivation for choosing such a database was primarily to use a realistic scenario as opposed to a generic collection which may give less useful results.In the case of the Coolidge Collection, and in many other real-world situations, the degree of similarityamong the images in the database is quite high. In particular, the lack of color to differentiate objectsas well as the small range of variety of objects in the images (most all of the images in the collectiondepict various combinations of people, cars, and buildings). This implies that the user must really examineeach image individually. Additionally, a text descscription of the target image was given, as opposed to theactual image since users rarely know the exact image to look for when searching through a database.

Between the treatments of thumbnail size, the area occupied by images was kept at an arbitrarily chosen constant of 512x512 pixels. This means that for the 3x3 treatment, the resolution of each image was about 170x170. Forthe 5x5 treatment, the resolution of each images was about 100x100. Given the nature of the database used,a denser grid of images (for example, 6x6) would have made it nearly impossible to discern the contents ofeach image. Since all the images were greyscale, thecolor depth used was 8 bit (256 levels of grey).