by Wojciech Jaśkowski, Krzysztof Krawiec, Bartosz Wieloch
Abstract:
We propose a novel method of evolutionary visual learning that uses a generative approach to assess the learner’s ability to recognize image contents. Each learner, implemented as a genetic programming (GP) individual, processes visual primitives that represent local salient features derived from the input image. The learner analyzes the visual primitives, which involves mostly their grouping and selection, eventually producing a hierarchy of visual primitives build upon the input image. Based on that, it provides partial reproduction of the shapes of the analyzed objects and is evaluated according to the quality of that reproduction. We present the method in detail and verify it experimentally on the real-world task of recognition of hand-drawn shapes. In particular, we show how GP individuals trained on examples from different decision classes can be combined to build a complete multiclass recognition system.We compare such recognition systems to reference methods, showing that our generative learning approach provides similar results. The chapter contains also detailed analysis of processing carried out by an exemplary individual.
Reference:
Genetic Programming for Generative Learning and Recognition of Hand-Drawn Shapes (Wojciech Jaśkowski, Krzysztof Krawiec, Bartosz Wieloch), Chapter in Evolutionary Image Analysis and Signal Processing (Stefano Cagnoni, ed.), Springer Berlin / Heidelberg, volume 213, 2009.
Bibtex Entry:
@InCollection{Jaskowski2009evolutionary,
Title = {Genetic Programming for Generative Learning and Recognition of Hand-Drawn Shapes},
Author = {Wojciech Jaśkowski and Krzysztof Krawiec and Bartosz Wieloch},
Booktitle = {Evolutionary Image Analysis and Signal Processing},
Publisher = {Springer Berlin / Heidelberg},
Year = {2009},
Editor = {Stefano Cagnoni},
Pages = {281-290},
Series = {Studies in Computational Intelligence},
Volume = {213},
Abstract = {We propose a novel method of evolutionary visual learning that uses a generative approach to assess the learner's ability to recognize image contents. Each learner, implemented as a genetic programming (GP) individual, processes visual primitives that represent local salient features derived from the input image. The learner analyzes the visual primitives, which involves mostly their grouping and selection, eventually producing a hierarchy of visual primitives build upon the input image. Based on that, it provides partial reproduction of the shapes of the analyzed objects and is evaluated according to the quality of that reproduction. We present the method in detail and verify it experimentally on the real-world task of recognition of hand-drawn shapes. In particular, we show how GP individuals trained on examples from different decision classes can be combined to build a complete multiclass recognition system.We compare such recognition systems to reference methods, showing that our generative learning approach provides similar results. The chapter contains also detailed analysis of processing carried out by an exemplary individual.},
Doi = {10.1007/978-3-642-01636-3_5},
Keywords = {genetic programming, visual learning, image recognition, generative learning},
Url = {http://www.springerlink.com/content/t146377701628050/}
}