Quantification and abstraction: low level tokens for object extraction

摘要

A paradigm for the abstraction of digital data into symbolic tokens that take part in object extraction is described. The low level processes of abstraction deal with the construction of tokens, and with knowledgebased choice of thresholds for all quantities that are involved in object extraction. These processes that transform digital data into symbolic, abstract tokens and descriptions are controlled by a higher level symbolic process that uses reasoning. A detailed description is presented of the generation and correction of two kinds of intermediate level tokens: line segments and uniform blobs. The construction of contextual segments from these tokens in the experiment is based on their symbolic description, as well as on the adaptive and symbolic description of all relevant quantities.