Presentation is loading. Please wait.

Presentation is loading. Please wait.

1 Generic Image Structures in Integrated Media Nick Rossiter & Michael Heather, University of Northumbria at Newcastle

Similar presentations


Presentation on theme: "1 Generic Image Structures in Integrated Media Nick Rossiter & Michael Heather, University of Northumbria at Newcastle"— Presentation transcript:

1 1 Generic Image Structures in Integrated Media Nick Rossiter & Michael Heather, University of Northumbria at Newcastle nick.rossiter@unn.ac.uk

2 2 Two main direction in Image Research Image contents: –modelled as set of attributes –at fairly high level of abstraction –but with little scope for free or ad hoc queries Feature extraction/object recognition subsystems: –automated object recognition –but difficult methods, computationally expensive and domain specific

3 3 Problems with Current Directions General models (e.g. databases) not used Concentration on customised methods Interoperability is difficult as it requires –more general techniques –meta and metameta data Importance of text not always recognised

4 4 Theoretical Needs Emphasis on powerobjects rather than atomic objects with flexible searching on clusters and groups Construction of universal relations for new connections intra-schema (local universe) and inter-schema (global universe) - i.e. for integration of images with text across heterogeneous databases Joining of type/domain attributes for different image representations (pixel, graph, Postscript)

5 5 Set Theoretic Approaches For example object-oriented methods –problems in unifying these Recent developments for universal description –MOF (Meta-Object Facility) –RDF (Resource Description Framework) –promising but aimed more at business data In general lack universal type system

6 6 Category Theory and Types Categories provide a theory of types. Typing is an inherent feature of every image recognition with two basic categories of data, the source and the medium –e.g. for an old master the source will be a human painter, whereas the medium may well be a painting in oils which will again import certain characteristics to the image and be specifiable in the retrieval process

7 7 Characteristic Features of Image Data Differentiating between background and foreground e.g. weather forecasting Use of colour for differentiation e.g. sunsets Texture e.g. human faces (qualia) –periodicity –directionality –randomness. Semantic interrogation - less work

8 8 S = category Source, M = category Medium, W/IMG = subcategory of W containing components of image, S X IMG M = product of category S and M over IMG

9 9

10 10

11 11 Conclusions 1 Full complexity of image recognition is shown: –Figure 2 (pullback with fuller collection of arrows) –Figure 3 (table showing nature of each pullback arrow) Binary relation between two categories seems adequate for representing image relationships The highest type of arrow, natural transformation, can represent characteristics like creativity (  s ) or image quality (  s x m )

12 12 Conclusions 2 Open systems are exposed to real-world complexity Category theory facilitates the integration of different models –universal type system –multi-level meta information –assists workers who use a selection of models (hierarchical, object-oriented) to represent aspects such as texture


Download ppt "1 Generic Image Structures in Integrated Media Nick Rossiter & Michael Heather, University of Northumbria at Newcastle"

Similar presentations


Ads by Google