A method for using gray-scale statistics for the inspection of assemblies is described. A test image of an assembly under inspection is registered with a model image of a nondefective assembly and the two images are compared on the basis of two statistical tests: a Χ2 test of the two marginal gray-level distributions and the correlation coefficient of the joint distribution. These tests are made in local subareas that correspond to important structure, such as parts and subassemblies. The subareas are compiled in an off-line training phase. The Χ2 measure is most sensitive to missing or damaged parts, whereas the correlation coefficient is most sensitive to mispositioned parts. It is also possible to detect overall lighting changes and misregistration with these measures. Two examples are presented that show how the tests detect two types of defects.
[1]
W. A. Perkins,et al.
A Model-Based Vision System for Industrial Parts
,
1978,
IEEE Transactions on Computers.
[2]
Michihiro Mese,et al.
A process for detecting defects in complicated patterns
,
1973,
Comput. Graph. Image Process..
[3]
Robert J. Woodham,et al.
Reflectance map techniques for analyzing surface defects in metal castings
,
1978
.
[4]
Stephen Thomas Barnard.
The image correspondence problem
,
1979
.
[5]
W.C. Lin,et al.
Feasibility study of automatic assembly and inspection of light bulb filaments
,
1975,
Proceedings of the IEEE.