The following research elaborates on some of the 'semantic' and 'algorithmic' aspects of the categorization process for thc colour domain. The structure of colour categories is argued to resemble the structure of Ihe distribution of colours in the perceived world. This distribution can be represented as colour statistics in some perceptual and approximately uniform colour space (e.g., the CIELUV colour space). We propose that the process of colour categorization is determined by a trade-off between (1) accuracy in representation of perceived colours and (2) simplicity of the category system. Colour categorization can be represented through the grouping of colour statistics by clustering algorithms (e.g., K-means). These assumptions are analysed on the basis of colour statistics of 630 natural images in the CIELUV colour space.
|Number of pages||9|
|Journal||IPO Annual Progress Report|
|Publication status||Published - 1998|