7845

7310

كارشناسي ارشد

هوش مصنوعي و رباتيك

اصفهان: دانشگاه صنعتي اصفهان، دانشكده برق و كامپيوتر

1391

نه،81ص.: مصور،جدول،نمودار

ص.ع.به فارسي و انگليسي

نيلوفر قيصري، علي فانيان

22/4/92

شادرخ سماوي، نادر كريمي

مهندسي برق و كامپيوتر

ID7310

به فارسي وانگليسي: قابل رويت در نسخه ديجيتالي

Indoor Scene Recognition Elham Seifossadat e seifossadat@ec iut ac ir Date of Submission 2013 1 23 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language FarsiSupervisor Niloofar gheissari n gheissari@cc iut ac ir Ali Fanian a fanian@cc iut ac irAbstractA scene is a view of a real world environment that contains multiples surfaces and objects organized in ameaningful way The Aim of Scene recognition system is understand Scene type that shown in the imageand then assign appropriate semantic tags to it Creating such a system is one of the most important issuesin the field of computer vision and robotics and has many applications in object recognition semanticimage and video retrieval motion detection locating and guiding a robot Scene can be divided into twocategories Indoor and outdoor Indoor scene includes enviroments such as kitchens hospitals and evenwithin the vehicle and the outdoor scene including open spaces such as streets beach and mountains There are some previous works devoted to the task of scene recognition but they usually only perform wellon outdoors In contrast recognition indoor images have remained a further challenging task due to themore difficult nature of the problem The intra class variations and inter class similarities of indoor scenesare the biggest barriers for many recognition algorithms to achieve satisfactory performance on images thathave never been seen i e test data Therefore trying to create a system with satisfactorily performance inrecognition of indoor images is essential In this thesis we are designe a system to recognize indoor images The algorithm contains two basic steps extract data of images and scene modeling using this data To extract local features of image are used set ofSIFT HOG and LBP descriptors Also to extract information on the spatial relationships in image regionsare calculated Co occurrence matrixs in several angels Then we use this information to semantic imagesegmentation and graph cut energy minimization to image labeling The proposed method is invarianttoward changes of scale rotation transformation contrast view point and occlusion and clutter in images Another important feature of the proposed method is the ability to generalize to test images Keywords Computer vision Scene Recognition Semantic Segmentation Local Semantic Concepts

نيلوفر قيصري، علي فانيان

شادرخ سماوي، نادر كريمي