Department of Pattern Recognition

The bidirectional texture function (BTF) describes texture appearance variations due to varying illumination and viewing conditions. This function is acquired by large number of measurements for all possible combinations of illumination and viewing positions hence some compressed representation of these huge BTF texture data spaces is obviously inevitable. In this paper we present a novel efficient probabilistic model-based method for multispectral BTF texture compression which simultaneously allows its efficient modelling. This representation model is capable of seamless BTF space enlargement and direct implementation inside the graphical card processing unit. The analytical step of the algorithm starts with BTF texture surface estimation followed by the spatial factorization of an input multispectral texture image. Single band-limited factors are independently modelled by their dedicated 3D causal autoregressive models (CAR). We estimate an optimal contextual neighbourhood and parameters for each CAR. Finally the synthesized multiresolution multispectral texture pyramid is collapsed into the required size fine resolution synthetic smooth texture. Resulting BTF is combined in a displacement map filter of the rendering hardware using both multispectral and range information, respectively. The presented model offers immense BTF texture compression ratio which cannot be achieved by any other sampling-based BTF texture synthesis method.

This work presents a fast model-based algorithm for realistic multispectral BTF texture model capable of direct implementation inside the graphical card processing unit. The algorithm starts with surface range-map estimation from one texture image based on shape from shading technique [Frankot,Chellappa 88]. The estimated range-map is finally combined with probabilistic smooth synthetic texture. Synthetic BTF image is rendered according surface range-map for required view and illumination angle.

This work presents a new type of scratch removal algorithm based on a causal adaptive multidimensional multitemporal prediction. The predictor use available information from the neighbourhood of a missing multispectral pixel due to spectral, temporal and spatial correlation of video data but not any information from the failed pixel itself. The model assumes white Gaussian noise in each spectral layer, but layers can be mutually correlated. A significant improvement of the 3D model performance is obtained if the temporal information is included, i.e., using the 3.5D causal AR model. Such information is natural to obtain from previous or/and following frame(s) for which we know all necessary data, due to high between-frame temporal correlation. Thanks to this we can treat data from different frames (specified by the contextual neighbourhood) in the same way, so we attach to each data information about its shift according to predicted pixel placement. The contextual neighbourhood has to be causal (in the reconstructed frame lattice subspace) . It means that the predictor can use only data from the model history. Then if we assume normal-Wishart parameter prior the predictor have analytical (not iterative) solution.

This texture modelling aaproach is based on discrete distribution mixtures. Unlike some alternative approaches the statistical properties of textures are modelled by a discrete distribution mixture of product components. The univariate distributions in the products are represented in full generality by vectors of probabilities without any constraints. The texture analysis is made in the original quantized spectral level coding. An efficient texture synthesis is based on easy computation of arbitrary conditional distributions from the model. Several successful colour texture applications of the method demonstrate the advantages and but also weak points of the presented approach.

This fast multigrid colour texture synthesis algorithm starts with spectral factorization of an input colour texture image using the Karhunen-Loeve expansion. Single orthogonal monospectral components are further decomposed into a multi-resolution grid and each resolution factors are independently modeled by their dedicated Markov random field model. Finally single synthesized monospectral single-resolution texture factorss are collapsed into the fine resolution images and using the inverse Karhunen-Loeve transformation we obtain the required colour texture.