In this paper we investigate an auto face annotation technique by mining weak facial images which does not consist any data. These images can be collected from various sources like videos, random images, internet, CCTV, etc. However this is a challenging problem because the images collected have incomplete data, they do not have proper label and are random as well as noisy. Such images are called weak images. To improve the quality of these weak images we build a database by extracting various facial features of these images. Given a weak query image for identification, the auto face annotation technique annotates the query image and obtains resultant values. These values are then matched with the values of images stored in the database to retrieve the top ranked similar facial images. We detect and align the facial image extracted from a weak image with help of Haar cascade classifiers and extract the GIST features from the detected facial image. Different clusters are then formed for improved scalability of these images which significantly boosts up the performance of this technique.

Keywords Gabor filter, cluster, Haar cascade classifiers, Kirsch detector, auto face annotation.