Thesis
Extracción de rasgos de imágenes de tortillas de maíz de diferentes niveles de productores para su caracterización inductiva
Autor
Ing. Rojas Padilla, Oscar Manuel
Institución
Resumen
In this thesis using Artificial Intelligence techniques such as computer vision and inductive learning was developed a method to quantify visual organoleptic characteristics of three different tortilla producers. Using characteristic features, extracted from tortilla’s images, solution rules were constructed for characterize each producer; with this rules founded is possible to establish a quality assessment method based on visual features, which due to its subjectivity nature are not currently used in quality analysis and much less are considered in setting standards of this product.
This research proposes the extraction of border píxels, using a morphological Hit-Miss transformation, and with a unique operation extract all of them no matter the value of the neighbors. By other hand for the inductive learning the BOUNDSTAR technique is implemented. For constructing the initial knowledge rules three type of characteristics where proposed: color, shape and texture. For detect the shape characteristics an analysis for circular shapes method is proposed, the results are invariant to rotation, growth and translation. Note that this method can be used for measure characteristics of other foods or even of shapes not necessarily circular.
With the analysis of another 300 images were found the learning rules, which characterize the three producer’s production. The evaluation of these rules was made using a new 300 images batch (100 images per producer), which were classified with respect to the learnt rule; the obtained results report efficiencies from 93% to 100% in the characterization of each producer.