1 / 21

Searching by shape in heterogeneous databases

Searching by shape in heterogeneous databases. Introduction Algorithms Methodology Experiments Conclusions and future works. Searching criteria. Colour. Texture. Spatial relationships. Shape. Searching by shape. Features : Rotation invariant Translation invariant

lassie
Download Presentation

Searching by shape in heterogeneous databases

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Searching by shape in heterogeneousdatabases • Introduction • Algorithms • Methodology • Experiments • Conclusions and future works

  2. Searchingcriteria Colour Texture Spatialrelationships Shape

  3. Searching by shape • Features: • Rotationinvariant • Translationinvariant • Scalinginvariant • Fast • Notadaptive

  4. First algorithm • Features: • Itworks on contours • Itis scalar • No feedback • Requiredinvariantsassured

  5. d1 C d2 Parameters f(x)= ax3+ bx2+ cx+d Ampiezza d1 d2 Distanze ordinate

  6. Considerations • Advantages: • Good result in a few cases • Very fast (only 4 parameters) • Rotation transaltion scaling invariance • Disadvantages: • Sensitivity to little local variations • Symmetric shapes make the algorithm collapse

  7. Second algorithm 1 ) Mass center is computed 60° 1 0° 0.75 2) Inertial axisi are computed 0.5 120° 0.25 C 3) 4 annulus are plotted 300° 4) 6 sector are plotted 180° 240°

  8. 0°-60° 0°-60° 60°-120° 60°-120° 120°-180° 120°-180° 180°-240° 180°-240° 240°-300° 240°-300° 300°-360° 300°-360° 0 0.25 0 0.25 0.25 0.5 0.25 0.5 60° 1 0° 0.75 0.5 0.75 0.5 0.75 0.5 0.75 1 0.75 1 120° 0.25 C 300° 180° 240° Matrix generation d1 d1

  9. Matrix comparison Matrix image 11 Images are ranked according to similarity Query matrix Matrix image 2 Matrix image N

  10. Precision & Recall Performance Problem When an elementisrelevant? Weneed a classification in the database

  11. The database • Database of 4553 images by Corel Draw • Heterogeneous images for size, subject and colour • Wedefine 22 categories of differentcardinality (from 54 to 400)

  12. Subdivisionbasingupon the shape of the object • I. e. simboli poligonali =Polygonalsimbols • Subdivisionbasingupon the semanticmeaning of tehobjects (i.e. flyingobjects) Choice of categories A trade off between:

  13. Experiments • Differentlevel of resolution (wavelets) • 20 query for eachcategory and eachresolutionlevel Thereisnot a priviledgedlevel of resolution for allclasses

  14. Simboli Tondi Simboli a Scudo Cardinalità Cardinalità Precision 5 Precision 5 Precision 10 Precision 10 Precision 15 Precision 15 Precision 20 Precision 20 Precision 25 Precision 25 Ideale Ideale 301 374 1 1 1 1 1 1 1 1 1 1 Imm. Base Imm. Base 301 374 0,544 0,644 0,595 0,464 0,421 0,523 0,398 0,533 0,541 0,370 Livello 1 Livello 1 374 301 0,608 0,660 0,488 0,639 0,453 0,602 0,422 0,557 0,529 0,403 Livello 2 Livello 2 374 301 0,688 0,540 0,592 0,441 0,432 0,533 0,454 0,486 0,459 0,455 Livello 3 Livello 3 374 301 0,550 0,600 0,426 0,484 0,434 0,456 0,426 0,416 0,386 0,424 Experiments

  15. Experiments • Analysis of the results for eachcategory • More 20 queries for eachcategoryat the best resolution Precision > 60%

  16. Categorie Cardinalità Precision 5 Precision 10 Precision 15 Precision 20 Precision 25 A. Reali 300 0,330 0,270 0,213 0,200 0,180 A. Stilizzati 131 0,268 0,194 0,152 0,120 0,110 Automezzi 54 0,320 0,205 0,157 0,130 0,116 Case 81 0,240 0,153 0,110 0,093 0,090 Composizioni 247 0,330 0,265 0,223 0,190 0,176 Dinosauri 95 0,470 0,360 0,300 0,268 0,244 F. Atipiche 400 0,360 0,260 0,223 0,190 0,176 Frasi 101 0,240 0,145 0,123 0,103 0,094 Insetti 132 0,310 0,190 0,157 0,140 0,128 O. Allungati 145 0,660 0,520 0,427 0,380 0,360 O. Poligonali 391 0,490 0,355 0,294 0,273 0,242 O. Curvilinei 197 0,280 0,160 0,140 0,113 0,098 O. Volanti 332 0,490 0,335 0,294 0,273 0,242 Pers. Reali 200 0,460 0,330 0,287 0,283 0,276 Pers. Stilizzate 161 0,408 0,240 0,205 0,222 0,195 Pesci 164 0,376 0,248 0,207 0,188 0,178 Scene 123 0,240 0,133 0,107 0,095 0,088 S. Poligonali 361 0,454 0,304 0,276 0,238 0,220 S. Tondi 374 0,660 0,639 0,602 0,557 0,529 S. a Scudo 301 0,688 0,592 0,533 0,486 0,459 Uccelli 208 0,333 0,200 0,156 0,139 0,120 Visi 55 0,260 0,145 0,127 0,110 0,096 Experiments

  17. Query: 1° 2° 3° 4° 5° 6° 7° 8° 9° 10° Experiments Distanze 1°- 0 2°- 0,0803 3°- 0,0896 4°- 0,0909 5°- 0,1006 6°- 0,1039 7°- 0,1041 8°- 0,1070 9°- 0,1087 10°- 0,1117

  18. Query: 1° 2° 3° 4° 5° 6° 7° 8° 9° 10° Experiments Distanze 1°- 0 2°- 0,0483 3°- 0,0710 4°- 0,0813 5°- 0,0871 6°- 0,0922 7°- 0,0927 8°- 0,0936 9°- 0,0938 10°- 0,0952

  19. Query: 1° 2° 3° 4° 5° 6° 7° 8° 9° 10° Experiments Distanze 1°- 0 2°- 0,1289 3°- 0,1433 4°- 0,1506 5°- 0,1520 6°- 0,1545 7°- 0,1546 8°- 0,1578 9°- 0,1585 10°- 0,1594

  20. 0.25 0.2 0.15 Distances 0.1 0.05 0 0 50 100 150 200 250 300 350 400 450 500 Number of images Distances

  21. Fast • Acceptableprecision for some classes • Upgrade of thealgorithm • Fusion with colour or texturemethods Conclusions The methodis: Future works

More Related