Texturesynthesisusingstasticalspatialmodels / Síntese de textura utilizando modelos estatísticos espaciais
AUTOR(ES)
Eduardo Quesado Filgueiras
DATA DE PUBLICAÇÃO
1998
RESUMO
The scope of this work is the use of methods capable of generating natural looking synthetic textures reproducing previously selected source patterns. Spatial statistic models are the basis for the used methods, specifically autoregressive moving average modeling for parameter estimation and texture pattern generation. Two methods are used for parameter estimation; both are based on autoregressive processes. The first method is based on using the linear autocorrelation function and on concatenating rows and columns of the analyzed texture. The second method uses a concatenated two dimensional autocorrelation function, and is able to avoid cumulative errors caused by undesired correlations generated by the row and column concatenation process using linear autocorrelation. The synthesis procedure is based on generating a zero mean two dimensional random white noise field which has the same variance as the residuals obtained from the modeling process. This white noise is the driver for a twodimensional autoregressive process, which results in a synthetic texture. For synthetic aperture radar images, a pre-processing technique using cubic root transformation on the image samples is used, as a way to obtain a near-Gaussian distribution. Images obtained by Landsat-5 and JERS-1 optical sensors are also tested, and so do natural textures such as marmour and granite. A graphic multiplatform computer-based program was developed using IDL language, containing all necessary functionalities for autoregressive modeling and texture synthesis.
ASSUNTO(S)
computer graphics modelos estatÍsticos statistical analysis modelagem computaÇÃo aplicada processos autorregressivos autoregressive processes sÍntese de textura computaÇÃo grÁfica
