Indian Movie Face database (IMFDB) is a large unconstrained face database consisting of 34512 images of 100 Indian actors collected from more than 100 videos. All the images are manually selected and cropped from the video frames resulting in a high degree of variability interms of scale, pose, expression, illumination, age, resolution, occlusion, and makeup. IMFDB is the first face database that provides a detailed annotation of every image in terms of age, pose, gender, expression and type of occlusion that may help other face related applications.

IMFDB is created with similar goals as Labeled Faces in the Wild (LFW) and Public Figures (PubFig) face datasets complementing them in many ways:

• Faces in IMFDB are collected from Indian movie videos unlike LFW and PubFigs where the images are collected from Internet sources. Faces in movie videos arguably have large variations in illumination, resolution, blurr, etc.
• Videos collected from the last two decades contain large diversity in age variations compared to the images collected from Internet through a search query.
• Public figures often retain certain kind of appearance, dress patterns and expressions when appearing in public while movie videos offer greater variations.
• LFW and PubFigs were built on images detected by Viola-Jones face detector, thereby focusing only on end-to-end systems that have automatic face detection followed by recognition. IMFDB is built by manual selection and cropping of video frames resulting in a large spectrum of poses develop efficient algorithms to handle pose.

We have kept the following design guidelines in mind while building the database.

• Selection of movies and actors: To ensure diversity in appearance of actors, movies are selected from 5 Indian languages namely, Hindi, Telugu, Kannada, Malayalam, and Bengali. For each actor, movies are selected such that they give wide variations in age.
• Selection of frames: Often, a single frame is selected from a shot (scene) unless there is a another frame with significant difference with the previously selected frame. If there are multiple variations available in a shot, face with occlusion and pose variation, were preferred. Small faces that are difficult to recognize manually are not considered.
• Cropping of faces: Faces are cropped with a tight bounding box. In order to maintain consistency across images, we followed a heuristic of cropping the face from forehead to chin.
• Annotation: For every image, annotation is provided for following attributes:
  1. Expressions :   Anger, Happiness, Sadness, Surprise, Fear, Disgust
  2. Illumination  :   Bad, Medium, High
  3. Pose            :   Frontal, Left, Right, Up, Down
  4. Occlusion    :   Glasses, Beard, Ornaments, Hair, Hand, None, Others
  5. Age             :   Child, Young, Middle and Old
  6. Makeup       :   Partial makeup, Over-makeup
  7. Gender        :   Male, Female

Reference

If you use this dataset, please cite the following paper:

Shankar Setty, Moula Husain, Parisa Beham, Jyothi Gudavalli, Menaka Kandasamy, Radhesyam Vaddi, Vidyagouri Hemadri, J C Karure, Raja Raju, Rajan, Vijay Kumar and C V Jawahar. "Indian Movie Face Database: A Benchmark for Face Recognition Under Wide Variations"

National Conference on Computer Vision, Pattern Recognition, Image Processing and Graphics (NCVPRIPG), 2013.

Contact