M.A. Anjum, M.Y. Javed, A. Nadeem, and A. Basit (Pakistan)
Normalized image , image size, scale invariant algorithm, characteristic matrix, model training, face recognition,
The high speed computing, database, networking technologies and sophisticated image processing methodologies have increased the topical significance of face recognition. A number of automated and semi automated strategies/techniques like Hidden Markov Model (HMM) based face recognition approach , top down HMM with pseudo 2-dimensional model and face recognition using eigen faces have modeled and classified faces based on normalized distances and ratio among feature points. Such approaches have proven difficult to extend to multiple views, especially variable sizes of image and have been fragile in nature. The proposed system has advantages over other face recognition schemes in its speed, simplicity, learning capacity, varying poses, intensity level and particularly relative insensitivity to small or gradual changes of size in the face images. This model consists of two parts; the first part is conversion of RGB into gray image and preprocessing of image. In the second part, the scale invariant algorithm is applied on the face images to achieve the insensitivity of face image to size variations. Training of model is done by transforming face images into a small set of characteristic features of arbitrary size images using scale invariant algorithm methodology, then recognition is performed by projecting a test image to the face space spanned by arbitrary reduced size face images.
Important Links:
Go Back
