Automatically Identifying Automatically Identifying and Georeferencing Georeferencing and Street Maps on the Web Street Maps on the Web Sneha Desai, Craig A. Desai, Craig A. Knoblock Knoblock, , Sneha Yao- -Yi Chiang, Yi Chiang, Ching Ching- -Chien Chien Chen Chen and and Yao Kandarp Desai Desai Kandarp University of Southern California University of Southern California Department of Computer Science and Department of Computer Science and Information Sciences Institute Information Sciences Institute
Outline Outline • Introduction and Motivation • Overall Approach and Algorithms • Experimental Results • Related Work • Conclusion and Future Work
Introduction and Motivation Introduction and Motivation � Various street maps are available on the Various street maps are available on the � web, but many of them web, but many of them � cannot be easily distinguished with other cannot be easily distinguished with other � images images � lack of the metadata that describes the lack of the metadata that describes the � geocoordinates and scales and scales geocoordinates
Introduction and Motivation Introduction and Motivation Street Maps Scanned Documents Photographs Political, state, area maps • Non street maps : • Irrelevant for the applications that seek only street maps • Street maps : • lack of metadata
Introduction and Motivation Introduction and Motivation � In this work, we In this work, we � � identify the street maps among different identify the street maps among different � images images � a apply our previous work pply our previous work to automatically � extract road intersections from the street maps (Chiang et al.) � apply conflation techniques to find the geocoordinates and align the streets on the maps with imagery (Chen et al.)
Outline Outline • Introduction and Motivation • Overall Approach and Algorithms • Experimental Results • Related Work • Conclusion and Future Work
Overall Approach Overall Approach City name (Query String) Module 1: Automatic classification of street maps Phase 1: Retrieving images from different sources Google Images Yahoo images Images Phase 2: Identifying street maps Map Filter Street maps of the city queried Module 2 : Automatic extraction of intersections Intersections on the street Maps Module 3 : Automatic georeferencing street maps s Geocoordinates and scales of the street maps
Identifying Street Maps Identifying Street Maps � Law’s Texture Classification Algorithm � (K. Laws. 1980) � Street maps have the unique textures � lines, labels, characters � Generate 75 different attributes (25R,25G,25B) to distinguish these textures on the images.
Identifying Street Maps Identifying Street Maps � (note) (note) SVM light V2.0 Support Vector Machine � (T. Joachims, 1999) � Training : � We provided 1150 different positive and negative examples of images � 75 attributes per image � Classification: � Using the trained SVM model to classify test images
Identifying Street Maps Identifying Street Maps Different types of Images from Image Search Filter 1 Street Maps Filter 2 Photographs, scanned docs, Political, area climate maps, icons (Non Street Maps) Dense and Sparse Street Maps Detailed Street Maps
Georeferencing Street Maps Street Maps Georeferencing � In our previous work: In our previous work: � � Automatically and Accurately Conflating Automatically and Accurately Conflating � Orthoimagery and Street Maps (Chen et al.) and Street Maps (Chen et al.) Orthoimagery � Integrate raster map and other sources. Integrate raster map and other sources. � � Utilize the Utilize the layout of the road intersections � within a local area to determine the map to determine the map’ ’s s location. location. Vector data with geocoordinate information Imagery with geocoordinate information Raster map without geocoordinate information
Outline Outline • Introduction and Motivation • Overall Approach and Algorithms • Experimental Results • Related Work • Conclusion and Future Work
Experimental Results Experimental Results (name of the, p, r ) 5 Total retrieved image URLs from image sources Remove nonworking and duplicate URLs 198 Nonworking URLs + Working URLs Duplicate URLs 152 Filter-1 46 Non-Street Maps Street Maps 113 (R=100%, P=97.35%) 39 (R=92.86%, P=100%) Filter-2 Dense and sparse street maps Street maps, found by Filter2 17 (R=94.44%, P=100%) 22 (R=100%, P=95.45%) Automatically georeferenced street Street maps not of the city queried maps of the city queried 15 (R=88.24%, P=100%) 7 (R=100%, P=71.43%)
Experimental Results Experimental Results • On the stage of • Identifying street maps, 100% recall, 95.45% precision • Georeferencing, 100% recall, 71.43% precision • The average computation time for identifying one street map 29.65 seconds
Outline Outline • Introduction and Motivation • Overall Approach and Algorithms • Experimental Results • Related Work • Conclusion and Future Work
Related Work Related Work • “Functionality Based Web Image Categorization.” Hu et al. - Focus on frequency domain and image features like uniformity, size and aspect ratio. (put the difference) • “Webseer: an image search engine for the world wide web.” Frankel et al. - Searching images by image context (file name-type-size and color depth) and by content based tests (put the difference)
Outline Outline • Introduction and Motivation • Overall Approach and Algorithms • Experimental Results • Related Work • Conclusion and Future Work
Conclusion and Future Work Conclusion and Future Work Main Contribution: � Identification of the street maps (precision Identification of the street maps (precision � = 95.45%) = 95.45%) � Automatically Automatically georeferencing georeferencing street maps street maps � (precision = 71.43%) (precision = 71.43%) � determine the determine the geocoordinates geocoordinates, scales , scales � � align the map with satellite imagery align the map with satellite imagery �
Conclusion and Future Work Conclusion and Future Work We plan to : � Classify the images into categories � political maps � weather maps � etc. � Reduce the number of feature dimensions � Combine OCR-related techniques
Thank you Thank you
Recommend
More recommend
