A computer vision inspection system has been developed that learns the difference between good and bad parts by being shown several identified good and bad parts. The model, formed during the training session, contains identifying points which are used for locating parts and inspection tests which apply only to pertinent regions of the part. Using the model, the system can distinguish between good parts and bad parts with an arbitrary number of defects. It can also learn to classify parts if it is shown the different parts during the training session. Examples of INSPECTOR inspecting industrial parts are shown.

A new method for segmenting images using abrupt changes in intensity (edge points) to separate regions of smoothly varying intensity is discussed. Region segmentation using edge points has not been very successful in the past because small gaps would allow merging of dissimilar regions. The present method uses an expansion-contraction technique in which the edge regions are first expanded to close gaps and then contracted after the separate uniform regions have been identified. In order to preserve small uniform regions, the process is performed iteratively from small to large expansions with no expansion for edge regions that separate different uniform regions. The final result is a set of uniform intensity regions (usually less than 100) and a set of edge boundary regions. The program has successfully segmented scenes with industrial parts, landscapes, and integrated circuit chips.

Control arm bushings, made with rubber of different hardness or resilience, are clas sified by a notch code. During the assembly process, the bushings are manually inspected to ensure that the correct type of bushing has been inserted into the control arm. A computer vision system to automatically classify bushings using the standard notch code has been developed. The program first determines the shape of the outer boundary of the bushing which is circular except for notch indenta tions. Next it determines the location of each notch and the type of notch. Finally, it compares the observed notch pattern with several models and thereby classifies the bushing.

A vision system has been developed which can determine the position and orientation of complex curved objects in gray-level noisy scenes. The system organizes and reduces the image data from a digitized picture to a compact representation having the appearance of a line drawing. This compact image representation can be used for forming a model under favorable viewing conditions or for locating a part under poor viewing conditions by a matching process that uses a previously formed model. Thus, models are formed automatically by having the program view the part under favorable lighting and background conditions. The compact image representation describes the boundaries of the part.