Research Seminar - August 2003

Texture Synthesis by Example

Abstract:

My proposed research is focussed on texture synthesis by example. The key to texture synthesis is to identify what features or statistics must be shared between two images to be perceived as the `same' texture. This can be achieved by imitating discriminating texture properties such as randomness and regularity. To emulate the generative process of textures, current methods match selected image characteristics between the output and sample image. E.g. multiresolution histograms, patch edges, pixel neighbourhoods, complex wavelet statistics, and multi-scale filter responses. However, these methods have problems such as speed, output quality, required memory, robustness to different texture types, and user intervention. Hopefully, my proposed research will alleviate these shortcomings where possible, as well as extend texture synthesis to arbitrary surfaces.