Please use this identifier to cite or link to this item:
|Title:||Vein identification for forensic investigation||Authors:||Zhang Hengyi||Keywords:||DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision||Issue Date:||2015||Source:||Zhang, H. (2015). Vein identification for forensic investigation. Doctoral thesis, Nanyang Technological University, Singapore.||Abstract:||With newly developed methods, blood vessel patterns hidden in color images become observable. In this research, we extend traditional blood vessel recognition from high quality blood vessel patterns collected from fingers, hands and wrists to low quality blood vessel patterns collected from other body parts, where skin is thicker. Blood vessel extraction and matching algorithms with special schemes designed for noisy patterns are proved to be effective for both infrared images and color images. Two fusion rules are also designed to combine blood vessel patterns uncovered by different methods. The encouraging matching results and the genetic dependence analysis show the potential of body blood vessel patterns for forensic investigation. According to our best knowledge, no one did similar research before. For evaluation of the proposed algorithms, we have developed a skin image collection system capturing both color and NIR images from 328 individuals. The experimental results on 1900 color images and 1900 infrared images from 490 forearms and 460 thighs demonstrate that the matching performance of the blood vessel patterns from the color images is comparable with that from the infrared images. With the proposed automatic blood vessel pattern matching algorithm, we also analyze and measure the similarity between 234 genetically identical forearm pairs and 204 genetically identical thigh pairs. Experimental results indicate that genetically identical blood vessel patterns have extra similarity, but they are distinguishable. Blood vessel uncovering methods are also improved during this research in two approaches. The first approach is based on image mapping using synchronized color and NIR image pairs and a neural network (NN) with automatic adjustment in weights and image intensities. The second approach uses multiple mapping models and local parameter estimation to improve the visualization of blood vessel patterns. The uncovering methods are also applied to images collected from the Internet.||URI:||http://hdl.handle.net/10356/63320||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||SCSE Theses|
Page view(s) 5185
checked on Oct 25, 2020
checked on Oct 25, 2020
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.