KBVision System - Group Module

KBVision Grouper Module


A common difficulty with the extraction of features from images is they appear continuous and coherent on a perceptual level, but are often fragmented and discontinuous at a pixel level. The problem is feature representations of a single object of interest extracted by a segmentation process may have disconnected edges, edgels, lines, or regions.

The Grouper Module gives the users the ability to associate Features into larger groups to either fill in for missing components, or to create higher levels of abstractions, such as creating chains from edgels. Automatic grouping Tasks which use local evidence (features), such as co-linearity, adjacency, proximity, contrast, color, texture, shape, multispectral density, etc., are part of the Grouper Module.

Grouping between two or more classes of features allows the user to create hierarchical relationships, and allows the explicit specification and representation of complex structures. The Grouper Module will greatly facilitate the creation of applications incorporating model- and atlas-based recognition schemes in medical and industrial fields.

Grouper Tasks include: EdglTo Chn1 / PtNbrs / LinkFeats / MakeLinks / NeighFeats / TksOverlap / SpltCCnst / SpltPVCnst / MergeCnst / StatsGrp.



The KBVision Image Examiner shown above illustrates a Grouping example. The image is an aerial photograph of roads in a mountainous area. The roads are not straight, are at different altitudes, and follow the contours of the land. The objective is to identify the roads for mapping purposes. From the pixel image it is easy to see that the road is not continuous. Perhaps trees have obscured parts of it. The KBVision Segmenter algorithm applied is called: RidgeEdgl. This Task produces Ridge and Valley Edgels. The segmentation result is displayed in blue, next to the Image. The small white box in the lower-right of the edgel display is the zoom box for the display directly to the right. It is obvious that the edgels are not connected and many do not represent the object Road.

The lower three panes of the Image Examiner show the grouped edgels. The KBVision Grouper Task EdglToChn1 connected and filtered the edgels based on parameters such as: Minimum Seed Magnitude, Minimum Chain Points, Maximum Link Length, Maximum Edgel to Link Angle, Link Length Weight, Edgel to Link Angle Weight. The left pane shows the connected Roads with the white box representing the zoom area for the adjacent pane. The middle pane shows the edgels as points with connecting lines, with the white box representing the zoom area for the adjacent pane. The right pane shows that the edgels are not evenly spaced, but were grouped based on the parameter setting for maximun link length.


Grouper Module Page 1 of 1

Go to KBVision Modules


webmaster@aai.com (Last Update 10/17/95)
KBVision and KBView are Trademarks of Amerinex Artificial Intelligence, Inc.