In machine vision applications, matching is one of the most commonly used image analysis tools. While basic matching tools can be found in just about every machine vision software package on the market, HALCON offers several unique advanced matching tools for 2D images.

Find them explained here….

correlation based matchingCorrelation-based matching uses a normalized cross correlation of gray values to evaluate the correspondence between a model and a search image. It is significantly faster than classical gray-value-based matching and can handle linear variations in illumination as well as rotation relative to the template pattern.  Correlation-based matching is preferred to shape–based matching (see below) for instances when objects may appear with slightly changing shapes, lots of texture, or may be out of focus (contours vanish in blurred images) in an image.  Correlation-based matching does not handle variations in scaling of the object in an image.

component based matchingComponent-based matching uses the edge contours of an object to locate its position and orientation in an image.  It is applied in cases when an object contains components between which distances or angles may change.  Component-based matching also tolerates variations in rotation, color, illumination, and cases where the object may be partially occluded and/or cluttered in an image.  It does not handle changes in scale.

 

 

Shape based matchingShape-based matching searches for objects based on edge contours.   Unlike gray-value and correlation-based matching, shape-based matching can determine the position and orientation of objects despite variations in scale (uniform or anisotropic), color, or if the object is occluded and/or cluttered.

 

 

Local deformable matching is a contour-based matching tool that can handle and return local deformations of an object.  It can also rectify the part of the image containing the deformed model. Local deformable matching works despite changes in illumination and an object’s color, and can handle occlusion and clutter.

Local deformable matchingPerspective deformable matching is a contour-based matching technique that can handle changes in the perspective view of the object in an image.  Perspective deformable matching can locate planar object contours in an image regardless of their 3D orientation and despite changes in illumination and an object’s color.  It can also handle occlusions and clutter.  Furthermore, perspective deformable matching can be used with 3D calibration, to determine the 3D pose of an object in an image.  This technique is best applied when a single planar part of the object is sufficient to distinguish it from other objects in the image.

 

 

Calibrated descriptor based matchingDescriptor-based matching uses a set of key feature points from a textured object to determine position and orientation. Like perspective deformable matching, descriptor-based matching can locate planar features on an object regardless of it’s 3D orientation.  If a 3D camera calibration has been performed, the 3D pose of the object can be derived.   Descriptor-based matching can accommodate changes in illumination, and can handle occlusion and clutter.    Descriptor-based matching is best applied for objects containing arbitrary but fixed texture in a single plane.

Note that while both perspective deformable matching and descriptor-based matching can be applied when objects are viewed orthogonally, they are better suited when imaging from perspective views since they are not as fast as the other approaches (which require orthogonal imaging).

Shape based 3D matchingShape-based 3D matching is a contour-based technique used when searching for objects that require more than one planar surface to uniquely distinguish the object from others in an image.  In this case a 2D template is not used, but rather a 3D model is created from a 3D CAD file of the object and used for the search. Shape-based 3D matching accommodates changes in perspective and illumination and can handle occlusions and clutter.  The object in the image must be the same size as the trained object model.

HDevelop Matching Assistant: To help quickly determine the best matching approach for an application, HALCON includes a powerful matching assistant tool in the HDevelop development environment.  The matching assistant allows users to easily switch between correlation-based, shape-based, perspective deformable, and descriptor-based matching techniques in order to quickly compare the results and determine the best approach for the application.  The assistant automatically generates the necessary code for use in the program once the best approach has been determined.

Download your trail copy of HALCON 11 here and try it out for yourself.

Need further advice?  Contact our sales team