Detecting gestures in medieval images
By Joseph Schlecht, Bernd Carque and Bjorn Ommer
Published in IEEE International Conference on Image Processing (ICIP) (2011)
Abstract: We present a template-based detector for gestures visualized in legal manuscripts of the Middle Ages. Depicted persons possess gestures with speciﬁc semantic meaning from the perspective of legal history. The hand drawn gestures exhibit noticeable variation in artistic style, size and orientation. They follow a distinct visual pattern, however, without any perspective effects. We present a method to learn a small set of templates representative of the gesture variability. We apply an efﬁcient version of normalized cross-correlation to vote for gesture position, scale and orientation. Non-parametric kernel density estimation is used to identify hypotheses in voting space, and a discriminative veriﬁcation step ranks the detections. We demonstrate our method on four types of gestures and show promising detection results.
Introduction: We present an automatic method to ﬁnd gestures in the illustrations of medieval manuscripts. Our focus on gestures in the visual arts of the Middle Ages is the ﬁrst step in a longterm interdisciplinary project to gain deeper insight into the nature of embodied communication in medieval culture . We base our approach on four illustrated manuscripts of Eike von Repgow’s Mirror of the Saxons. The detector described in this paper lays the groundwork to compare corresponding scenes from each copy automatically with regard to the depicted gestures.
Our goal is to detect multiple types of gesture at different scales and orientations in the digitized manuscripts. The fact that the gestures are drawn by hand introduces a significant challenge due to artistic variation. A positive aspect of these man-made images, however, is that they follow simple 2-D patterns without perspective. We take advantage of this drawing style with a template driven detection strategy. Given labeled instances of a particular type of gesture, our approach centers on learning a subset that spans its appearance variation. We cast votes for detections based on an efficient version of normalized cross-correlation, followed by a veriﬁcation stage to rank the hypotheses.