PatternBook

--A Website that can recognize and cluster geometrical patterns using crowd-sourcing data and machine learning algorithms

4.570 Digital Heritage, Class Project, 2015

Instructor: Takehiko Nagakura

A geometrical pattern is a discernible regularity in nature, or in artificial objects. We can observe/feel pattern from any of the five senses. They are everywhere, and we recognize and remember them in our daily life, consciously or unconsciously.

It would be interesting if we can compare how we perceive patterns with other people, and it would be useful if we can have these patterns analyzed and clustered. However, there is no existing system or platform that can help human to share, record, analyze and cluster geometric patterns. This is because it is very difficult for a computer to recognize patterns. Using computer graphics techniques we can recognize the objects or detect the edges in an image, but we cannot recognize specific geometric structure. Also, if we have geometries, it is still difficult to recognize the repetitive units. Even if we can recognize the repetitive units, they have to be further analyzed by their symmetric properties. The ornaments and redundancy in the image make the computer confused. Moreover, some shapes in a image sometimes are ambiguous, people can make multiple interpretations on them.

In this project, I built a prototype website and proposed two approaches to classify geometric pattern images: topology analysis and clustering, and human perception crowd-sourcing clustering. Users can play the crowd-sourcing game, compare the result between human decision and topology classification, or upload a new image to see which cluster it belongs to.