Skip to main content

Table 2 Reviewed representations and descriptors for symbol recognition and spotting

From: Symbol spotting for architectural drawings: state-of-the-art and new industry-driven developments

Paradigm* Descriptor Robustness and invariance Features used
P \(\mathcal {F}\)-signature [11] Robust to geometric transformations and noise Image pixels
P Pixel-Level Constraint [12] Scale and rotation invariant and robust to degradation Image skeleton
P Blurred Shape Model (BSM) [13, 14] Robust to soft, rigid, and elastic deformations Skeleton points
P Circular Blurred Shape Model (CBSM) [15, 16], BSM properties + rotation invariant Contour map (flexible for other representations)
P Shape Context Descriptor (SCIP) and extensions [20, 22] Scale and rotation invariant Interest points
V Perceptual grouping, Fully Visibility Graph (FVG) [24] Rotation and scale invariant Vectorial primitives
V Structural representation and Attributed Relational Graph (ARG) [25, 26, 29, 30] Scale and rotation invariant, robust to small variations Vectors and quadrilateral primitives
V Hierarchical Plausibility Graph (HPG) [32, 33] Robust to various distortions Critical points and lines
V Shape, topology, and Region Adjacency Graph (RAG) [35, 36] Rotation and scale invariant Image regions
V Boundary and Region Adjacency Graph (RAG) [37] Rotation and scale invariant Image regions
V Convexity and Near Convex Region Adjacency Graph (NCRAG) [38] Rotation and scale invariant Oriented line segments
V Bag-of-GraphPaths (BoGP) [39] Rotation invariant Critical points
V Jacobs’ statistical grouping [45] Scale and rotation invariant Contour map
V Bag-of-Relations (BoR) [31] Scale and rotation invariant and robust to irregularities Thick (solid) components, circles, corners, and extremities
V Cassinian ovals [47] Not invariant to scaling and rotation Polylines of closed region contours
  1. *P = pixel-based, V = vector-based