A Framework for the Evolutionary Generation of 2 D-Lookup Based Texture Filters

A framework is presented, which allows for the automated generation of texture filters by the exploitation of the 2D-Lookup algorithm and its optimization by means of evolutionary algorithms. To use the framework, one has to give an original image, containing the structural property-of-interest (e.g. a surface fault), and a binary image (goal image), wherein each position of the structural property-ofinterest is labeled with the foreground color. Doing so, the framework becomes capable of evolving the specification of the 2D-Lookup algorithm, i.e. the two image processing operations and the 2D-Lookup matrix, towards a filter for the structural property-ofinterest. Two versions of the framework will be considered, one using the genetic algorithm (GA), and one using the genetic programming (GP). For the GA approach, the filter generator is given access to a fixed set of image processing operations. The decoded bitstring specifies, which operations are selected out. Also, the decoded bitstring specifies the entries in the 2D-Lookup matrix. For the GP approach, the filter generator uses the expression tree of an individual to design two operations based on formal superoperators, which are referenced within the expression tree. The specification of the 2D-Lookup matrix is performed by a relaxation technique. It will be shown by some texture fault examples, that the GP approach performs better than the GA approach.