categorization and reorientation of images based on low level features分类和重新定位基于低层次的图像特征.pdfVIP

Journal of Intelligent Learnning Sy stems and Applicatioons, 2011, 3, ***-** 1 doi:10.4236/jilssa.2011.31001 PPublished Onlinee February 201 11 (http://www.ScciRP.org/journaal/jilsa) Categorizatiion and Reorrientattion off Imagges Bassed on Loww Level Feattures 1 1 1 1 2 Rajen Bhattt , Gaurav Sharma , Abhhinav Dhall , Naresh Kuumar , Santanu Chaudhuury 1Samsung Indiaa Software Centeer, Noida, India; 2Electrical Enggineering Deparrtment, Indian Innstitution of Tecchnology Delhi, New Delhi, India. Email: rajen.bhhatt@, schaudhury@ Received June 222nd, 2010; revissed July 29th, 20010; accepted Seeptember 23rd, 22010 ABSTRACCT A hierarchicaal system to peerform automaatic categorizaation and reorientation of immages using coontent analysiiss is pre- sented. The prroposed systemm first categorizzes images to ssome a priori ddefined categoories using rotaation invariantt features. o o o oo At the second stage, it detectts their correctt orientation ouut of {0 , 90 , 180 , and 270 } using categoory specific moodel. The system has be een specially deesigned for emmbedded devicees applicationss using only loww level color aand edge featuures. Ma- chine learningg algorithms ooptimized to ssuit the embeddded implemenntation like suupport vector machines (S VMVMs) and scalable boosting have beenn used to deveelop classifiersrs for categorizzation and oriientation detecction. Results are pre- sented on a coollection of abbout 7000 conssumer images collected fromm open resourcces. The propoosed system finnds it ap- pp lications to vvarious digital media productts and brings pp attern recogniition solutions to the consummer electronics d