Department of Pattern Recognition

Content-based image retrieval (CBIR) systems, target database images using feature similarities with respect to the query. We introduce fast and robust image retrieval measures that utilise novel illumination invariant features extracted from three different Markov random field (MRF) based texture representations. These measures allow retrieving images with similar scenes comprising colour-textured objects viewed with different illumination brightness or spectrum. The proposed illumination insensitive measures are compared favourably with the most frequently used features like the Local Binary Patterns, steerable pyramid and Gabor textural features, respectively. The superiority of these new illumination invariant measures and their robustness to added noise are empirically demonstrated in the illumination invariant recognition of textures from the Outex database.

Statistically based preprocessing of screening mammograms is proposed with the aim to in-crease the diagnostic conspicuity of mammographic lesions. We estimate first the local statis-tical texture model of a single mammogram as a joint probability density of grey levels in a suitably chosen search window. The probability density in the form of multivariate Gaussian mixture is estimated from data obtained by pixel-wise scanning the mammogram with the search window. In the second phase we evaluate the estimated density at each position of the window and display the corresponding log-likelihood value as grey level at window center. Light grey levels correspond to the typical parts of the image and the dark values reflect unusual places. The resulting log-likelihood image closely correlates with fine structural details of the original mammogram and facilitates diagnostic interpretation of suspect abnormalities.

The 3D model of a dragon measured using laser range scanner.

Textural appearance of many real word materials is not static but shows progress in time. If such a progress is spatially and temporally homogeneous these materials can be represented by means of dynamic texture (DT). DT modelling is a challenging problem which can add new quality into computer graphics applications. We propose a novel hybrid method for colour DTs modelling. The method is based on eigen-analysis of DT images and subsequent preprocessing and modelling of temporal interpolation eigen-coefficients using a causal auto-regressive model. The proposed method shows good performance for most of tested DTs, which depends mainly on properties of the original sequence. Moreover, this method compresses significantly the original data and enables extremely fast synthesis of artificial sequence, which can be easily performed by the means of contemporary graphics hardware.

We present a framework for sampling-based texture synthesis where the learning phase is clearly separated and the synthesis phase is very simple and fast. The approach exploits the potential of tile-based texturing and produces good and realistic results for a wide range of textures. The synthesis phase is realizable directly in graphical hardware. As a part of the framework we also propose a fast and adjustable sub-optimal path search algorithm for finding minimum error boundaries between overlapping images. The algorithm may serve as an efficient alternative of traditional slow path search algorithms like the dynamical programming etc.