Texture synthesis and unsupervised recognition with a nonparametric multiscale Markov random field model

Abstract

In this paper we present noncausal, nonparametric, multiscale, Markov Random Field (MRF) model for synthesising and recognising texture. The model has the ability to capture the characteristics of a wide variety of textures, varying from the structured to the stochastic. For texture synthesis, we use our own novel multiscale approach, incorporating local annealing, allowing us to use large neighbourhood systems to model some complex textures. We show how we are able to manipulate the statistical order of our high dimensional model without over compromising the integrity of the representation. Also by varying the statistical order of our model we are able to optimise it for the unsupervised recognition of textures with respect to textures that have not been modelled.

Hidden Markov Random Field, an Application to Railway Infrastructure Diagnosis

“…In our application, the Markov field is used to learn the repeated singular point patterns of the track. Our approach to learn the local conditional probability distributions of X is thus closer to those found in Texture Synthesis or Recognition (Paget and Longstaff 1984) and is nonparametric. Then, the vector of parameters (Ψ, Φ) describing the model has thus two different parts :…”

“…However, with nonparametric estimation, the local conditional probability distributions may not define a valid joint distribution. Nevertheless, in practice, such an estimation approach has already proved is efficiency in texture recognition (Paget and Longstaff 1984).…”

Investigating the effects of scale in MRF texture classification

“…Another possibility for future research is to attempt to incorporate texton positioning information into the texture representation scheme. Previous work by Paget and Longstaff [11] attempted to incorporate larger scale texture effects for the purposes of texture synthesis. This approach may be applicable to texture classification, and it would be interesting to see if the spacial positioning of the textons plays an important role in texture classification in a non parametric MRF framework.…”

“…Due to the very good rates of classification obtained by their method (abbreviated as the VZ classifier), it is used as a benchmark for comparison with the methods developed in this work. Texture Synthesis is also an area where MRF approaches have been proven to work well, as demonstrated by Efros and Leung [4], Paget and Longstaff [11] and Zalensy and Van Gool [17].…”

Robust tile-based texture synthesis using artificial immune system

“…On the other hand, global methods use some criteria to evaluate the similarity of the input, then the entire texture can be evolved as a whole. Most existing global approaches either model only pixel-to-pixel interactions which are insufficient to capture large-scale structures of the sample texture [5,6], or introduce too complex formulations to optimize [7,8]. Kwatra et al [9] defined a texture energy function to quantitatively measure the quality of the synthesized texture, unfortunately the synthesizing speed is still quite slow.…”