Towards automatic image segmentation using optimised region growing technique
- Authors: Nicholson, Ann , Li, Xiaodong , Alazab, Mamoun , Islam, Mofakharul , Venkatraman, Sitalakshmi
- Date: 2009
- Type: Text , Conference paper
- Relation: Paper presented at 22nd Australasian Joint Conference, AI 2009: Advances in Artificial Intelligence, Melbourne, Victoria : 1st-4th December 2009 Vol. 5866, p. 131-139
- Full Text: false
- Description: Image analysis is being adopted extensively in many applications such as digital forensics, medical treatment, industrial inspection, etc. primarily for diagnostic purposes. Hence, there is a growing interest among researches in developing new segmentation techniques to aid the diagnosis process. Manual segmentation of images is labour intensive, extremely time consuming and prone to human errors and hence an automated real-time technique is warranted in such applications. There is no universally applicable automated segmentation technique that will work for all images as the image segmentation is quite complex and unique depending upon the domain application. Hence, to fill the gap, this paper presents an efficient segmentation algorithm that can segment a digital image of interest into a more meaningful arrangement of regions and objects. Our algorithm combines region growing approach with optimised elimination of false boundaries to arrive at more meaningful segments automatically. We demonstrate this using X-ray teeth images that were taken for real-life dental diagnosis.
- Description: 2003007514
Stochastic model based approach for biometric identification
- Authors: Islam, Mofakharul , Venkatraman, Sitalakshmi , Alazab, Mamoun
- Date: 2010
- Type: Text , Conference proceedings
- Full Text: false
- Description: In this paper, we present a new stochastic model based approach for enhanced image segmentation in biometric identification systems. Biometric features such as fingerprint, face, iris, hand geometry and more recently dental features are being used for human identification. Image analysis of each of these biometric features has various challenges to overcome. To address such contemporary problems of image segmentation, we provide a novel approach based on maximum a posteriori (MAP) fitting Gaussian mixture model using Expectation-Minimization (EM) algorithm within the Bayesian framework. Our new algorithm captures the pixel intensity by the likelihood term in Bayesian Networks, and a priori biasing term of the spatial location information with the help of Markov Random Fields (MRF) model. We have employed a novel approach of using Daubechies wavelet transform for texture feature extraction that uses MRF model and a robust technique of determining the number of pixel classes based on Cluster Ensembles for a reliable segmentation of dental X-ray images. We present how our approach could be applied in dental biometrics to achieve very fast and reliable human identification. Experiments show that our new unsupervised image segmentation technique provides accurate feature extraction and teeth segmentation for effective biometric identification.