General
1. There are more to machine vision than meets the eye.
----A machine vision system does not see things as the human eye does.
2. An eye is not a camera, A brain is not a computer.
----Machine vision systems should not necessarily be modelled on, or intended to emulate human vision.
3. No vision system should be required to answer the general question "What is this?"
----It is better for vision systems to answer more specific questions, such as "Is this widget well made? Verification (i.e. checking that widget is well made) is better than recognition, where few or no a priori assumptions are made.
4. Optimal solutions do not always exist.
----If they do exist, optimal solutions may be too complex, or impossible to find. We should therefore be prepared to search for and accept satisfactory solutions, rather than optimal ones.
5. Use a standard solution to a vision problem but only if it is sensible to do so.
----Wherever possible we should provide standard solutions to industral problems, since this help to broarden the application base.
6. Avoid the application of machine vision techniques for there own sake.
----It is vanity on the part of vision engineer to do so. There are plenty of other methods of solution available. Most of them are cheaper than vision.
7. Defect prevention is better than cure.
----We should consider using vision in closed loop feedback control of the manufacturing process.
8. Do not rely on second-hand information about the manufacturing process and environment.
----The vision engineer should always see the manufacuring process for himself. If the customer is unwilling to let the vision engineer into the factory, it may be necessary to abandon the application.
9. Vision systems need not be fully automatic.
----While it is more usual to use a fully automatic vision system, it can be used instead to enhance images for subsequent human analysis.
Lighting and Optics
1. Many hands make light work.
----But not very well. However, some people do apply proper engineering principles to the design of the optical sub-system and inevitably obtain better results.
2. The lighting is not constant.
----Lighting is never constant in either time or in space.
3. Never use software to compensate for a poor lighting system.
----It is not cost effective and will result in a poor system design.
4. It is cheaper to add a light-proof shroud to keep sun-light away from the object under inspection than to modify the software.
----Another universal truth which is often forgotten.
5. Nothing exceeds the speed of light.
----Any processing that can be done optically will save a lot of computer processing later.
6. It is all done by mirrors.
-----Wishful thinking, in view of the previous remark.
B.G. Batchelor and P.F. Whelan (1994), "Machine vision systems: Proverbs, principles, prejudices and priorities", Proceedings of the SPIE - The International Society for Optical Engineering, Vol.2347 - Machine Vision Applications, Achitectures, and Systems Integration III, Boston (USA), pp 374-383.