![](bg1.png)
LARGE-SCALE-INVARIANT TEXTURE RECOGNITION
Muhammad Rushdi and Jeffrey Ho
Computer and Information Science and Engineering, University of Florida, Gainesville, U.S.A.
Keywords:
Texture classification, Scale-invariance, Gray-level co-occurrence matrices.
Abstract:
This paper addresses the problem of texture recognition across large scale variations. Most of the exist-
ing methods for texture recognition handle only small-scale variations in test images. We propose using
microscopic-scale textures to classify texture images at any coarser scale without prior knowledge of the rel-
ative scale. In particular, given a test camera image, we compute the average error of approximating the test
texture with patches of the microscopic texture for certain category and scaling factor. Recognition is made
by selecting the category with the minimum average error over all categories and scaling factors. Experiments
on camera and low-magnification microscopic images show the validity of the proposed method.
1 INTRODUCTION
This paper explores the problem of classifying tex-
ture across large scale variations. In particular, using
high-magnification microscopic textures, we aim to
classify textures at any coarser scale. The difficulty
of the problem stems from several facts. First, im-
ages of the same material with large variations of the
imaging scale may appear so different even for a hu-
man observer (Figure 1). Second, accurate and fast
techniques need to be developed to relate the mate-
rial appearances at different scales. Although a lot
of work has been done in the area of texture recogni-
tion (Davies, 2008), (Varma and Zisserman, 2009),
little attention has been made to the effect of large
scale variations. The CUReT database (Dana and
Koenderink, 1999) captures texture images for 61 cat-
egories where each category is represented by 205 im-
ages of different viewing and illumination conditions.
However, this database lacks examples of scale vari-
ation except for 4 materials that have slightly scaled
images. Varma and Zisserman (Varma and Zisser-
man, 2009) claim that their MRF texture model is not
adversely affected by scale changes. However, their
experiments were done on the aforementioned scaled
CUReT images which have only a small scale factor
of 2. Kang (Kang and Nagahashi, 2005) developed
a framework for scale-invariant texture analysis using
multi-scale local autocorrelation features. Neverthe-
less, the experiments were limited to small changes in
scale ranging from 0.7 to 1.3. Leung and Peterson
(Leung and Peterson, 1992) used moment-invariant
and log-polar features to classify texture. However,
scale variations in their experiments were limited to
0.5, 0.67, and 1.0.
Our contribution in this paper is threefold. Firstly,
we introduce a new approach for classifying texture
across large scale variations. In particular, we show
how an approximation of a test image using micro-
scopic textures can be used to recognize textures at
any scale. Secondly, we employ our approach to es-
timate the relative scale of a test image with respect
to microscopic texture. Thirdly, we provide a dataset
of multi-scale textures that can be used to assess the
robustness of texture classifiers to scale changes.
2 COLLECTING MULTISCALE
TEXTURES
Many texture databases are freely available including
the CUReT database (Dana and Koenderink, 1999)
and the UIUC database (Lazebnik and Ponce, 2005).
While these databases sample reasonably the varia-
tions in illumination and viewing points, none of them
properly captures scale variations of the textured ma-
terials. To fill this gap, we started collecting multi-
scale texture data. In this paper, we show experiments
on five categories of materials that have challenging
textural patterns: cloth, loofa, marble, sponge, and
granite plaster (Figure 1). For every category, we
captured images using two imaging devices. Firstly,
camera texture images were collected using a high-
resolution 8-MB digital camera. Twenty images were
taken at different distances, angles and illumination
442
Rushdi M. and Ho J..
LARGE-SCALE-INVARIANT TEXTURE RECOGNITION.
DOI: 10.5220/0003398904420445
In Proceedings of the International Conference on Computer Vision Theory and Applications (VISAPP-2011), pages 442-445
ISBN: 978-989-8425-47-8
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)