ANR PHOENIX

Workgroup SYMBOLETS

[SYMBOlic waveLETS in data mining]

 

Context and ‘SYMBOLET’ framework

 

Information discovery from huge numeric image datasets usually relies on data mining techniques. The steps involved in a data mining technique can be summarized as follows:

(i)                extraction of local and spatial features, generally consisting in symbols and

(ii)              mining (usually exhaustive) for the retrieval of frequent/occurrent and connected symbol evolutions.

The latter step (ii) is sensitive to the reliability of the description features extracted in step (i). In this respect, many strategies have been investigated for pre-processing data (image filtering techniques, oriented pyramid structuring, etc.) so as to increase feature reliability. These strategies mainly apply in the so-called numeric domain (digital image concern) where arithmetic sum and product operations imply nice algebraic properties.

 

The ‘SYMBOLET’ framework is a different approach: it consists in defining functional representations for symbolic data (feature domain). Regularity/occurrences of symbols will be embedded in a symbolic algebra framework that involves re-defining standard addition and difference operations.

 

 

Team