Texture classification: are filter banks necessary?

We question the role that large scale filter banks have traditionally played in texture classification. It is demonstrated that textures can be classified using the joint distribution of intensity values over extremely compact neighborhoods (starting from as small as 3 /spl times/ 3 pixels square), and that this outperforms classification using filter banks with large support. We develop a novel texton based representation, which is suited to modeling this joint neighborhood distribution for MRFs. The representation is learnt from training images, and then used to classify novel images (with unknown viewpoint and lighting) into texture classes. The power of the method is demonstrated by classifying over 2800 images of all 61 textures present in the Columbia-Utrecht database. The classification performance surpasses that of recent state-of-the-art filter bank based classifiers such as Leung & Malik, Cula & Dana, and Varma & Zisserman.

1 Introduction

Texture research is generally divided into four canonical problem areas [7]: (1) synthesis; (2) classification; (3) segmentation; and (4) shape from texture. Significant progress was made during the 1990s on the first three areas (with shape from texture receiving comparatively less attention). The success in these areas was largely due to learning a fuller statistical representation of filter bank responses [1], [2], [10], [11], [13], [17]. It was fuller in three respects: firstly, the filter response distribution was learnt (as opposed to recording just the low order moments of the distribution); secondly, the joint distribution, or co-occurrence, of filter responses was learnt (as opposed to independent distributions for each filter); and thirdly, simply more filters were used than before - typically between ten and fifty filters or wavelets - to measure texture features at a set of scales and orientations.