Adaptive Image Analysis for
Abstract:
robustness and predictability are crucial because the fate of a battle and those who are fighting it is at stake. Computer-vision systems must have robust, predictable performance before they will be fully embraced in military applications and other safetycritical applications such as medical imaging and intelligent transportation systems. We try to deal with these requirements by keeping systems simple. Experience has shown that the systems that work best are those whose complexity has been minimized and that have been manually tweaked until they perform reasonably well as long as nothing changes too much. Unfortunately, things do change. Sometimes, what changes is the environment, such as the lighting, altitude, or weather. Other times, it is the technology, such as a new camera or the replacement of an algorithm. Even small changes often cause systems to break unpredictably. We can address such fragility in several ways. One approach is to build in more and more application-specific knowledge. However, this can lead to unmanageable complexity and performance characteristics that are not understandable. The dominant approach has been to develop mathematical analysis of imagery and applications. This has led to some highly
Citations
| 147 | Reflection and Semantics in a Procedural Language – Smith - 1982 |
| 79 | The early processing of visual information – Marr - 1976 |
| 9 | Texture segmentation using local energy in wavelet scale space – Xie, Brady - 1996 |
