Three forces direct the evolution of machine vision: Enabling technology determines what is possible, algorithms determine what is practical, and applications determine what is desirable. Machine vision depends on two kinds of enabling technology, sensors and processors, but the field has not been economically powerful enough to significantly influence either. We use components designed primarily for other purposes and, from this, derive the enormous price/performance leverage generated by an economy much greater than our own. In both sensors and processors, we are witnessing a sea change brought on by the growth of consumer multimedia applications. In the near term, inexpensive complementary metal oxide semiconductor sensors will continue to appear in applications that are below the reach of more expensive charge-coupled devices. As CMOS volumes increase and the technology matures, the advantages of CCD image quality will disappear. In the midterm, the enormous gap between cheap, single-purpose photodetectors and, the low-end, general-purpose vision systems will be filled with application-specific sensors that are lower in cost than general-purpose systems.