190 likes | 304 Views
WebQuality2012. A Deformation Analysis Method for Artificial Maps Based on Geographical Accuracy and Its Applications. University of Hyogo, Japan Daisuke Kitayama , Kazutoshi Sumiya. Background. Artificial maps on the real space and the Web space
E N D
WebQuality2012 A Deformation Analysis Method for Artificial Maps Based on Geographical Accuracy and Its Applications University of Hyogo, Japan Daisuke Kitayama, KazutoshiSumiya
Background • Artificial maps onthe real space and the Web space • Signs on the city, leaflets for sightseeing, guiding information on the Web, ... • Emphasized and weaken geographical information for a certain purpose • Confirming information easily and quickly
Deformation Analysis and Geographical Accuracy • Including excess deformations in artificial maps • Too long or too short distances • Inconsistent positional relations • Finding the target position difficult to detect by excess deformations • Detecting the current position on this artificial map • Showing retrieved restaurants on this artificial map Distance is too long Distance is too short Blue station is not on the left of red station
Enhanced artificial map • Taking a artificial map from signboard or Web page • Detecting excess modifications • Showing the current position • Adding retrieved geographical objects You are here Distance is too long
Our approach • Recognizing geographical objects on a map • Reading maps by OCR techniques and detecting mistaken OCR results by spatial features • Translating coordinate to an artificial map from a real map based on geographical accuracy Real map Artificial map Gazetteer OCR 1. Recognizing geographical objects 2. Translating coordinate based on geographical accuracy
1. Recognizing geographical objects on a map • 1-1. Reading characters using OCR • 1-2. Extracting candidate geographical objects for read characters by edit distance • 1-3. Narrowing down candidate objects • 1-4. Detecting mistaken corresponded objects • Using a statistics method of outlier detection
1-1. Reading Characters and 1-2. Candidate objects • 1-1. Reading characters using OCR • There are some noises of characters • 1-2. Extracting candidate geographical objects for read characters using edit distance Objects names A: PART-DIEU B: PERRACHE C: FOCH OCR results A: PABI-DYEN B: PEPPACME C: FOCM C1: PARIS-LYON C2: SAB-LINE C3: PART-DIEU B1: PARIS-LYON B2: SAB-LINE B3: PART-DIEU A1: PARIS-LYON A2: SAB-LINE A3: PART-DIEU A: PABI-DYEN B: PEPPACME C: FOCM Gazetteer
1-3. Narrowing down candidate objects • Showing objects in a certain region • Concentrating correct objects in a certain region • Deleting a candidate object that is the furthest from the center of gravity of candidate objects • Repeating until a read character related to a candidate object Artificial map Positions of candidate objects A1 A3 A2 B3 C1 C2 OCR results A: PABI-DYEN B: PEPPACME C: FOCM B2 B1
1-4. Detecting mistaken objects’ information • Mistaken information of objects are included in 1-3 results • Checking distances between a object and other objects • Detecting outliers by statistical method as Smirnov-Grubbs test using distance difference between artificial maps and real maps Positions of candidate objects Artificial map
2.Translating coordinate using geographical accuracy (1/2) • Finding corresponding points by geometrical calculations using any three objects and a certain point • Drawing a perpendicular line to a edge from the target point • Mapping point of intersections to the artificial map • Drawing lines between each point of intersection Artificial map Real map
2.Translating coordinate using geographical accuracy (2/2) • Using a waited average of corresponding points for detecting the corresponding point on the artificial map • Calculating similarity of distance and direction between an artificial map and a real map as a weight of candidate point Artificial map Artificial map Real map Real map △ABD = 0.8 △ACD = 0.5 ・・・ ・・・ ・・・
Evaluation • Recognizing artificial maps • Number of data set: 20 • Access information of universities and sightseeing information of cities • Checking results using correct answer made by two participants • Translating positions to artificial maps • Number of data set: 10 • Using a half of experiment “Recognizing artificial maps” • Number of times: 220 • Determining correct result by three participants
Experimental result: Recognizing artificial maps • Extracted characters can be corresponded to correct geographical objects • A lot of object names didn’t read by OCR
Result of Kyusyu University • Most objects are corresponded correctly
Result of Hokkaido University • All of objects are corresponded wrong positions • We read wrong character from the artificial map
Experimental result: Translating positions to artificial maps • Correlation coefficient between accuracy and precision is 0.62 • When recognition is success, translating process returns good results
Results of Kyusyu university • Result • Precision of map recognition: 0.85 • Accuracy of corresponding point: 0.65 • We can estimate corresponding point on this map • We get a lot of correct geographical information
Result of Himeji City • Result • Precision of map recognition: 0.71 • Accuracy of corresponding point: 0.40 • Some wrong results have wrong positional relations with a road
Conclusion • Enhanced artificial map using translating coordinate based on geographical accuracy analysis • Recognizing artificial maps using OCR techniques • Finding corresponding points based on deformation analysis • Confirming recognition accuracy and translation accuracy • Future work • Treating line objects such as roads and rivers • Evaluating on real environment with the mobile phone Recognize Translate