Human Perception in Segmentation of Sketches

Abstract

Abstract. In this paper, we study the segmentation of sketched engineering drawings into a set of straight and curved segments. Our immediate objective is to produce a benchmarking method for segmentation algorithms. The criterion is to minimise the differences between what the algorithm detects and what human beings perceive. We have created a set of sketched drawings and have asked people to segment them. By analysis of the produced segmentations, we have obtained the number and locations of the segmentation points which people perceive. Evidence collected during our experiments supports useful hypotheses, for example that not all kinds of segmentation points are equally difficult to perceive. The resulting methodology can be repeated with other drawings to obtain a set of sketches and segmentation data which could be used as a benchmark for segmentation algorithms, to evaluate their capability to emulate human perception of sketches.Keywords: Sketch recognition. Low level ink processing and pen stroke segmentation. Engineering Graphics. Segmentation Ability. PresentationOur interest is computer-based recognition of sketched engineering drawings, such as would allow automated conversion of engineering sketches into CAD representations. Segmentation of the drawing is a critical stage, and one which has received much attention over the years. Some important aspects of segmentation still remain unsolved, perhaps because (as [1] shows), segmentation is not, in fact, a single problem, but a set of similar problems. In this paper, we consider one such unsolved aspect: the benchmarking of computer-based segmentation of sketches. Recognition of the object portrayed in engineering drawings is a topical subfield of graphics recognition, which deals more generally with how computers can interpret semi-structured drawings which contain both freeform elements and symbols defined by convention. In the case of creating 2D or 3D CAD models of engineering objects from single or multiple drawings, it is the freeform elements, lines and curves, which portray the surfaces of the object, and it is these which we wish to identify. At this stage of processing, the conventional symbols (like dimensions and hatching) are clutter, and should be removed (and perhaps stored for later use). Our objective is thus to segment engineering drawings into lines, curves and clutter.When evaluating new segmentation approaches, one common strategy is simply comparing the number of segmentation points obtained by the new approach with the number of segmentation points which the "theoretical" shape possesses (by "theoretical" we mean the ideal primitives obtained from a line drawing by applying a well-defined set of topological and geometrical constraints). This strategy assumes that the new approach should detect those properties which the theoretical shape should possess, regardless of whether or not the actual drawing used as input really does possess them.In reality, we cannot assume that a sketched line drawing on paper will always contain exact...

Robust feature point detection for freehand strokes with deep learning approach

“…Therefore, to be more practical, we further ask users to label the positions of real corners and tangent points for the 1500 strokes, based on human perception [26]. These labels are also used to evaluate the accuracy of our algorithm.…”