Archives

Ant Colony Optimization (ACO) With Three Different Classifiers for Soft Biometrics Estimation


Dr.A. Devi
Abstract

Automatic image interpretation and face recognition is a challenging problem in artificial intelligence, pattern recognition and computer vision with applications in security authentication, biometry, and biomedicine. In image processing techniques applications, most important field is biometry. In recent days, in the research field, facial images based soft biometric classification gained a huge interest. In existing work, (i) Noise is an irregular fluctuation or unwanted signal that accompanies a signal, quality of digital images can be seriously affected by noise. Images are corrupted with noise during image transmission; capture etc. which results in the degradation of image quality, (ii) unclear nature of image boundary are resulted because of low quality of image and possible factors, (iii) Less interpretable nature of independent variables: On dataset, principal components are formed from original features by implementing PCA. But, it is not interpretable as well as readable as like original features. To solve these issues, proposed a novel method to enhance soft biometric authentication’s classification accuracy. This work consists of four main stages, (i) Preprocessing, (ii) Extraction of Feature (iii) Selection of Feature and (iv) Classification. Firstly, preprocessing stage removing noise from images by using filtering algorithm, while keeping its features intact for better understanding and recognition, then edges are d etected using canny edge detector, which is a most useful image enhancement techniques for improving image analysis process quality. Secondly, features are extracted from the preprocessed image. Two multi-resolutions transform namely shearlet and waveatom are used for extracting effective features from facial images. Age, facial expression, ethnicity, gender are estimated using these features. Thirdly, Ant Colony Optimization (ACO) algorithm used for more significant features is selected to enhance the classification accuracy for training dataset. Finally, Support Vector Machine (SVM), Artificial Neural Network (ANN) and Self-Organization Map (SOM) classifiers are used for performing final estimation. Three various databases like FG-NET, Extended Cohn-Kanade, US Adult Faces are used for collecting huge database for experimentation.

Volume 12 | 05-Special Issue

Pages: 74-90

DOI: 10.5373/JARDCS/V12SP5/20201737