ShapeToGraphMapping code for publication -
This repository contains both the code for Voronoi graph construction and feature extraction, and code for boundary clustering as described in the paper.
----------------------------------------------------------------
-
EDGE EXTRACTION [vertexList, edgeList, imageStruct] = multiLayerBorder(image,0.99) This code accepts a binary image (image) and traces the borders between the black and white boundaries of pixels. This prepares the input for Voronoi diagram generation If working with a label image, please see information on multiLayerBorder_LABEL
-
DIAGRAM CONSTRUCTION [records]= Voronoi_DT[vertexList, EL, ‘ImageStruct’,imageStruct)
Uses the information from multiLayerBorder to construct a voronoi diagram on a set of input edges. Information about the Voronoi diagram will be stored in records. Records has five key fields, each of which has a set of it’s own fields .Inputs – an nx5 array of information about all inputs .edgeSup – an nx9 array with additional info about input edges. .verts – a structure with multiple fields. Contains information about output vertices. .edges – a structure with multiple fields. Contains information about output edges in the diagram. .facesFull – contains information about the Voronoi cell surrounding each input For more information about the records structure, see the section on records.
-
BRIDGE EXTRACTION [dataStruct] = processRecords(records) This process identifies all bridges, defined as vertices and half edges which exactly bisects two objects. Requires the processed output array and creates a structure with a field .bridges which holds information about the vertices and edges on this bridge.
-
LOOP EXTRACTION [dataStruct] = extractCycles(records, dataStruct) This process uses information from the records_ARRAY from cleanRecords and dataStruct from bridge extraction to find all global cycles inside each object with a hole. This adds fields .loop.object.* and .loop.hole.*, which each have fields .acyclic(n), .enclosed(n), and .enclosing(n), which provides the boundary profiles along with points on the cycle in each region. Information about a specific boundary can be extracted by calling getCycleInfo(dataStruct, color)