One class of problems that poses significant challenges to computer scientists is image processing and enhancement, particularly for subsequent use by machine learning algorithms. Detecting edges and paths reliably, despite relatively poor-quality source images taken at high framerates by cameras onboard robots, for example, is crucial if any sort of pathfinding algorithm is to be based on the resulting data. Applying filters to images in the hopes of getting something back that looks “good” is also difficult, as programming a degree of subjectivity into algorithms becomes quite taxing in terms of computational power required to examine the characteristics upon which that subjectivity is based increases exponentially as the image gets larger.

Improvements in particle swarm optimization algorithms, recently introduced by Alaa Sheta and Malik Braik at the Department of Information Technology at Al-Balga Applied University, provides intriguing advances in image recognition and processing. By introducing an objective fitness criterion that assigns higher weights to enhanced edge pixels, letter- or number-like patterns, greater relative differences between adjacent pixels (that is, a measure of contrast), and the like, a framework for efficient, effective image filters can be developed. Enhancing the quality of low-resolution camera phone images, sharpening CCTV images to make license plates and other identifying marks more readable, and reliably and quickly detecting edges in noisy images for later use by AI and pathfinding algorithms, are all intriguing potential applications of this technology. This technology stands to find broad applicability within the domain of typical consumers, in the form of image-processing filters for digital pictures and movies, enhancing camera phone pictures prior to publishing them online or sharing them with friends and family, locating objects (such as prescription medication bottles or glasses) in the homes of elderly people who misplace them with a simple camera monitoring each room, and the like.

Looking forward, algorithms like this particle swarm optimization one that achieve a desired outcome by modeling guided interactions between semi-randomly influenced separate entities look promising for applications to improved AI and simulations of real-world phenomena. Modeling the behavior of large, complex systems on the basis of its numerous tiny, independent constituents begins to approach the way many real-world systems (such as transportation networks, supply chains for manufacturing corporations, social groups of animals, and the like) are organized, and thus stands to offer a higher degree of relevancy and accuracy in the resulting models.

References

http://en.wikipedia.org/wiki/Particle_swarm_optimization
http://blogs.zdnet.com/emergingtech/?p=823
http://www.swarmintelligence.org/index.php
http://www.sciencedaily.com/releases/2008/02/080201093341.htm
http://www.vnunet.com/vnunet/news/2208761/boffins-swarm-enhance-digital