Fuzzy matching of edge and curvature based features from range images for 3D face recognition. Issue 1 (2nd January 2017)

• Record Type: Journal Article
• Authors: Ganguly, Suranjan; Bhattacharjee, Debotosh; Nasipuri, Mita
• Is Part Of: Intelligent automation & soft computing. Volume 23:Issue 1(2017)
• Page Start: 51
• Page End: 62
• Publication Date: 2017-01-02
• Subjects: Range image -- edge -- curvature -- hausdorff distance -- fuzzy rule -- face registration -- face recognition -- frav3D and gavabDB databases; Artificial intelligence -- Periodicals; Intelligent control systems -- Periodicals; 003.5
• DOI: 10.1080/10798587.2015.1121616
• Languages: English
• ISSNs: 1079-8587

Abstract

Abstract: Automatic human face recognition is already in research from some decades due to its application in different fields. But there is no unique technique that is very much worthwhile for robust automatic human face recognition, suitable for all possible situations. In this paper, a new technique is proposed, which is a holistic approach, and it is based on 'one to all' comparison method. Along with the edge, four different types of curvatures are computed from face image profile to capture both the local features and surface features from 3D face image. Then, a new feature space, EC (Edge_Curvature) image, is generated for feature estimation during final recognition purpose. The similarities among intra-class members are carried out using fuzzy rule derived from the computed distance vectors by Hausdorff, distance that is used to match the probe images for the classification purpose automatically. For the validation of the algorithm, the algorithm is experimented on Frav3D and GavabDB databases with two sets of investigations. One is synthesized data-set, consisting of frontal range images (i.e. expression, illumination and neutral) and registered range face images. The other set is the original range face images. It does not include the registered faces. These investigations highlight the robustness of the proposed methodology. The success rates of acceptance of the probe images from two synthesized datasets are 98.87% for Frav3D and 87.20% from GavabDB. On the other hand, classification rate from original data-set for GavabDB is 79.78% and 91.69% for Frav3D.