Chapter 24 - Epilogue—Perspectives in vision

Abstract

This chapter is concerned with the methodology and limitations of computer vision. It starts by discussing the parameters of importance in vision, including “nasty realities” such as noise, occlusions, optical distortions, and the effects of stray light and shadows. It then discusses the limitations of vision algorithms and the specifications they need to have for successful implementation. It also develops the idea of tradeoffs between parameters and the need for optimization and compromise. It considers the gains resulting from the continuing effects of Moore’s law and the dangers of being locked into any particular type of hardware system. It notes that the way algorithms are built up is important, reflecting different representations of the processing tasks, warning of the dangers that some representations may in the end hold up progress. It also analyzes the place of deep learning networks in computer vision. Last, it considers what can be said reliably about the future of computer vision and its potential applications.