As indicated above, the images in each scheme were classified into non-overlapping categories. In Face, there were twelve categories: typical Asian males, typical Asian females, typical black males, typical black females, typical white males, typical white females, Asian male models, Asian female models, black male models, black female models, white male models and white female models. In the Story scheme, there were nine categories: animals, cars, women, food, children, men, objects, nature, and sports.
The images used for each category were carefully selected from a number of sources. ``Typical male'' and ``typical female'' subjects include faces selected from (i) the Asian face database [26] which contains color frontal face images of 103 people and (ii) the AR Face database [17] which contains well over 4000 color images corresponding to 126 people. For the AR database we used images in angle 2 only, i.e, frontal images in the smile position. These databases were collected under controlled conditions and are made public primarily for use in evaluating face recognition technologies. For the most part, the subjects in these databases are students, and we believe provide a good representative population for our study. Additional images for typical male subjects were derived from a random sampling of images from the Sports IllustratedNBA gallery.
Images of ``female models'' were gathered from a myriad of pageant sites including Miss USA, Miss Universe, Miss NY Chinese, and fashion modeling sites. Images of ``male models'' were gathered from various online modeling sources including FordModels.com and StormModels.com.
For the Story scheme, the ``men'' and ``women'' categories were the same as the male and female models in our Face experiment. All other images were chosen from PicturesOf.NET and span the previously mentioned categories.
To lessen the effect that an image's intensity, hue, and background color may have on influencing a user choice, we used the ImageMagick library (see www.imagemagick.org) to set image backgrounds to a light pastel color at reduced intensity. Additionally, images with bright or distracting backgrounds, or of low quality, were deleted. All remaining images were resized to have similar aspect ratios. Of course, it is always possible that differences in such secondary factors influenced the results of our experiment, though we went to significant effort to avoid this and have found little to support a hypothesis of such influence.