Summary: | 碩士 === 中正理工學院 === 電機工程研究所 === 86 === Segmenting an image into uniformly textured
regions is a prerequisite for many computer vision
and image understanding tasks. Texture segmentation
can be achieved either by extracting uniform texture
regions or detected textural discontinuities ( texture
edge detection ) between image regions.
In this thesis an unsupervised texture analysis
model is proposed. In this model, the implicit self-
similarity property of texture is analyzed by employing
the fractal dimension. Using the variagram plot versus
a set of pixel distance pair, the scale of the texture
element can be obtained. Following the calculation of
fractal dimension of each pixel block by block, the
texture boundary can be detected. Various experiments
are conducted to test the proposed model.
|