Contribución a la segmentación de la boca en imágenes orientada al reconocimiento automático de gestos bucales

Abstract

En este trabajo se presenta una nueva metodología para el reconocimiento automático de gestos de la boca orientada al desarrollo de una interfaz hombre-máquina para el comando de endoscopios. Dicha metodología comprende etapas comunes a la mayoría de sistemas de visión artificial, como lo son el tratamiento de la imagen y la segmentación, además de un método para el mejoramiento progresivo del etiquetado resultante de la segmentación inicial. A diferencia de otras aproximaciones, la metodología propuesta se adecua a gestos bucales y que no implican movimientos de la cabeza. A lo largo del documento se presta especial interés a la etapa de segmentación, ya que es ésta la que presenta mayores retos en el reconocimiento de gestos / Abstract: This document presents a series of elements for approaching the task of segmenting mouth structures in facial images, particularly focused in frames from video sequences. Each stage is treated separately in different chapters, starting from image pre-processing and going up to segmentation labeling post-processing, discussing the technique selection and development in every case. The methodological approach suggests the use of a color based pixel classification strategy as the basis of the mouth structure segmentation scheme, complemented by a smart pre-processing and a later label renement. The main contribution of this work, along with the segmentation methodology itself, is based in the development of a color-independent label renement technique. The technique, which is similar to a linear low pass filter in the segmentation labeling space followed by a non-linear selection operation, improves the image labeling iteratively by filling small gaps and eliminating spurious regions resulting from a prior pixel classification stage. Results presented in this document suggest that the rener is complementary to image pre-processing, hence achieving a cumulative effect in segmentation quality. At the end, the segmentation methodology comprised by input color transformation, pre-processing, pixel classification and label renement, is put to test in the case of mouth gesture detection in images aimed to command three degrees of freedom of an endoscope holder.