Parametric and Non-parametric Models of Linear Prediction Error for Color Texture Segmentation

This paper presents a comparison of parametric and non-parametric models of multichannel linear prediction error for supervised color texture segmentation. Information of both luminance and chrominance spatial variation feature cues are used to characterize color textures. The method presented consists of two steps. In the first step, we estimate the linear prediction errors of color textures computed on small training sub images. Multichannel complex versions of linear prediction models are used as image observation models in RGB, IHLS and L*a*b* color spaces. In the second step, overall color distribution of the image is estimated from the multichannel prediction error sequences. Both parametric and non-parametric approaches are used for this purpose. A multivariate Gaussian probability approximation is used as the parametric law defining this color distribution. For non-parametric approximation, we have used a multivariate version of k-nearest neighbor algorithm. Error rate, based on well classified pixels, for different linear prediction models using different color spaces are compared and discussed.