The Database of Faces (www.cl.cam.ac.uk)

The Database of Faces (formerly the ORL Database) is a classic, publicly available benchmark released by AT&T Laboratories Cambridge containing 400 grayscale face images (10 images each for 40 subjects) captured between 1992–1994. Images are 92×112 pixels with 256 gray levels in PGM format, organized in 40 subject folders (s1…s40) with files named Y.pgm (1–10). Shots were taken against a uniform dark background in an upright, mostly frontal pose with controlled variations in lighting, facial expression (open/closed eyes, smiling), timing, and accessories (glasses/no glasses). The archive (~4.5 MB) is downloadable from the AT&T Cambridge repository and is accompanied by a canonical paper and a PhD thesis on Hidden Markov Models for face recognition. Significance: this dataset has long served as a compact, low-complexity benchmark for face-recognition research (PCA/Eigenfaces, LDA, HMMs, etc.), useful for rapid prototyping, reproducible comparisons and teaching. Technical constraints—small image size, limited pose/background diversity, and controlled acquisition—make it easy to experiment with classical algorithms but also limit its realism compared with modern, large-scale in-the-wild datasets. Users should credit AT&T Laboratories Cambridge when using the images.