Two channels synthesis by analysis of coloured textures

We present an unsupervised synthesis method of coloured textures resembling a given prototype. This method is aimed at reducing the number of parameters necessary to synthesise the three correlated components of a colour image. To this purpose we perform a Karhunen-Loeve (KL) analysis of the hue saturation brightness (HSB) components of the given prototype. The analysis and synthesis procedures is dramatically simplified exploiting the spatial cross correlation of the two principal components of the KL transform of the given texture only. The effects of the component reduction obtained using the KLT on a visual colour space representation of textures are discussed and synthetic examples are provided.