Texture recognition from sparsely and irregularly sampled data

In this paper, we present methodology for recognising textures from irregularly sampled data. We use features constructed from the trace transform, which represents images with functional values along tracing lines rather than brightness values at sampling points. Once texture classification may be performed using line, as opposed to point representations, there is no problem about using irregularly sampled data. The analysis is performed using tracing lines identified by the Hough transform. The results obtained are compared with the results obtained by performing texture classification using samples on the conventional regular grid.