The UCR vislab database has 10,038 images covering a variety of real-world scenes such as castles, cars, humans, animals, etc. Three persons are asked to make independent judgment regarding the class assignment with 56 possible class names given to them. Each image is assigned to the appropriate class when the majority of these three persons agreed with this assignment. For the images which are assigned to three distinct classes by the three persons, the labeling provided by the image processing and computer vision expert (having more than 2 years of experience in the field) is used as the groundtruth. The sample images from each class are shown in below.

The number of images in each class varies from 20 to 695. Each image in the database is represented by a 22-dimensional feature vector having 16 Gabor features (the mean and standard deviation of filtered images at 4 orientation and 2 scales) and 6 color features (the mean and standard deviation from the HSV color domain). For further information on this database please contact the system administrator at jgonzalez@vislab.ucr.edu