Automatic quantification of iris color

Publication: Research - peer-reviewConference abstract in proceedings – Annual report year: 2012

View graph of relations

An automatic algorithm to quantify the eye colour and structural information from standard hi-resolution photos of the human iris has been developed. Initially, the major structures in the eye region are identified including the pupil, iris, sclera, and eyelashes. Based on this segmentation, the iris is resampled into a standardized quadratic coordinate system, where occluded and invalid regions are masked out. Secondly, a pixel classification approach has been evaluated with good results. It is based on a so-called Markov Random Field spatial classification into dominantly brown and blue regions. The result is a blue-brown ratio for each eye.
Furthermore, an image clustering approach has been used with promising results. The approach is based on using a sparse dictionary of feature vectors learned from a training set of iris regions. The feature vectors contain both local structural information and colour information. For each iris an explanatory histogram is build, containing information about the weighted occurrence of each visual word. A hierarchical agglomerative clustering of the entire set of photos is performed using the distance between the explanatory histograms. The approach is completely data driven and it can divide a group of eye images into classes based on structure, colour or a combination of the two. The methods have been tested on a large set of photos with promising results.
Original languageEnglish
Title of host publicationMeeting of the English Speaking Working Group (ESWG) of the International Society of Forensic Genetics (ISFG) : Programme
Publication date2012
Pages20
StatePublished

Conference

ConferenceMeeting of the English Speaking Working Group (ESWG) of the International Society of Forensic Genetics (ISFG)
CountryDenmark
CityCopenhagen
Period30/05/1202/06/12
Internet addresshttp://www.welcomehome.dk/Default.aspx?ID=2204
Download as:
Download as PDF
Select render style:
APAAuthorCBEHarvardMLAStandardVancouverShortLong
PDF
Download as HTML
Select render style:
APAAuthorCBEHarvardMLAStandardVancouverShortLong
HTML
Download as Word
Select render style:
APAAuthorCBEHarvardMLAStandardVancouverShortLong
Word

ID: 9681700