Computer Vision
An introduction to nonlinear dimensionality reduction by maximum variance unfolding
AAAI'06 proceedings of the 21st national conference on Artificial intelligence - Volume 2
| Hi-index | 0.00 |
This paper addresses the topic of how architectural visual experience can be represented and utilised by a software system. The long-term aim is to equip an artificial agent with the ability to make sensible decisions about aesthetics and proportions. The focus of the investigation is on the feature of line distributions extracted from digital images of house façades. It is demonstrated how the dimensionality reduction method isomap can be applied to calculate non-linear "street-manifolds" where each point on the manifold corresponds to a house façade. Through interpolation between manifold points and the application of an inverse Hough transform, basic structure plans for new house façades are obtained. If the interpolated points are close to the manifold it can be argued that the new plans reflect the character of the surrounding streetscape. The method is also demonstrated using basic examples which can be represented by circles.