RICH Reading Images for the Cultural Heritage DIEPGRAVENDE COMPUTERS

1
(No Transcript)
2
Computer visionandArchaeology

• RICHReading Images for the Cultural Heritage

3
RICH team

• Guus Lange (ROB, Amersfoort)
• Eric Postma (MICC-IKAT, UM)
• Paul Boon (MICC-IKAT, UM)
• Hans Paijmans (MICC-IKAT, UM)
• Laurens van der Maaten (MICC-IKAT, UM)

4
RICH aims

• Development of new techniques for automatic image
  analysis
• Providing tools to archaeology to make
  classification easier, faster, and more objective
• Enrichment of knowledge on archaeological material

5
Working examples

• Content-based image retrieval of historical glass
  
• Incorporated in historical glass reference
  collection website
• Automatic coin classification

6
CBIR for historical glass

• Aids classification of glass
• Nowadays, the expert searches through entire
  books to find alike glass drawings
• This process is slow and error-prone
• Our system compares glass photographs (made by
  the expert) with glass drawings (from the books)
• Provides entry into collection website

7
(No Transcript)
8
(No Transcript)
9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
CBIR for historical glass

• Allows for knowledge enrichment
• All objects in the collection can be compared
• Visualization of this comparison allow insight in
  relations between objects
• Unsupervised learning could even be used to
  construct new typologies

18
(No Transcript)
19
Current work...
20
Automatic coin classification

• After introduction of the euro, large amounts of
  unsorted coins were collected (over 300 tons)
• Manual sorting not feasible
• We are developing a high-performance, high-speed
  system for coin classification

21
Automatic coin classification

• Example coin (1 of 109 coin classes)

22
Automatic coin classification

• Using various contour features and texture
  features
• Edge-based statistical features
• Gabor-based features
• Daubechies wavelet features

23
Automatic coin classification

• Our system achieves promising classification
  performances (currently 76)
• Rejecting unknown or unclear coins (low number of
  wrong classifications)
• Classification takes 1 second on a normal desktop
  PC
• Including image loading, segmentation, feature
  extraction and classification

24
Automatic coin classification

• Final goal classification of medieval coins

25
Conclusions

• RICH delivers useful applications to archaeology
• RICH delivers new insights
• To archaeologists
• New view on typologies and classifications
• To computer scientists
• Provides difficult, real-world data for the
  development of new techniques

26
(No Transcript)