Texture Analysis with Shape

Co-occurrence Patterns

Gang Liu, Gui-Song Xia*, Wen Yang, Liangpei Zhang

[pdf] [code coming soon]

Abstract

This paper presents a flexible shape-based texture method by investigating the co-occurrence patterns of shapes. More precisely, a texture image is represented by a tree of shapes, each of which is associated with several attributes. The modeling of texture is thus converted to characterize the tree of shapes. To this aim, we first learn a set of co-occurrence patterns of shapes from texture images, then establish a bag-of-words model on the learned shape co-occurrence patterns (SCOPs), and finally use the resulted SCOPs distributions as features for texture analysis. In contrast with existing work, the proposed method not only inherits the strong ability to depict geometrical aspects of textures and the high robustness to variations of imaging conditions from the shape-based texture method, but also provides a more flexible way to consider shape relationships and high-order statics on the tree. To our knowledge, this is the first time to use co-occurrence patterns of explicit shapes as a tool for texture analysis. Experiments of texture retrieval and classification on various databases report state-of-the-art results and demonstrate the efficiency of the proposed method.

Methodology

Following the work of SITA [4](Refer to Related works), this paper contributes a more flexible shape-based texture analysis framework by investigating the co-occurrence patterns of shapes. The flowchart is illustrated in Fig.1. More precisely, given a texture, we first decompose it into a tree of shapes relying on a fast level set transformation [12], where each shape is associated with some attributes. We then learn a set of co-occurrence patterns of shapes from texture images, e.g. by K-means algorithm. Taking the learnt shape co-occurrence patterns as visual words, a bagof- words model is finally established to describe the texture. In contrast with SITA [4], the proposed method provides a more flexible way to consider complex shape relationships and highorder statics on the tree. Moreover, as we shall see, SITA can be regarded as a special case of the proposed one, where only marginal distributions and simple statistics of pair of shapes were taken into account. To our knowledge, the work in this paper is the first time to use explicit shape co-occurrence patterns for the analysis of textures. Several experiments on texture recognition demonstrate the efficiency of the proposed analysis method on various databases.

Figure 1: The proposed texture analysis method with shape co-occurrence patterns (SCOPs). First images are represented by tree of shapes via Fast Level Set Transform (FLST) . The branches of the trees are then collected and partitioned into different clusters, called SCOPs. An image is finally encoded by the learned SCOPs and the corresponding histograms are used as the texture model for analysis.

Retrieval Experiments

A. Retrieval Curve

The retrieval experiment consists in using one sample of the database as a query to retrieve the Nr most similar samples in the data set. For evaluation, the average number of correctly retrieved samples (generally called recall) when the query spans the whole database is drawn as a function of Nr.
We evaluates our method on three popular texture datasets, i.e., UIUC Texture database, UMD texture database, Brodatz texture database, and compare our results with the popular methods. Fig. 2 illustrates one of the retrieval result, more results can be accessed in UIUCTex Demoand UMDTex Demo.

uiuc umd brodatz
Fig. 2 The retrieval result. From left to right: UIUC database, UMD database, Brodatz database

B. Retrieval Examples:

UIUC T25_Plaid

UMD T9_Apple

Brodatz D100

Classification Results

In classification experiments, we evaluates our method on three popular texture datasets, i.e., UIUC Texture database, UMD texture database, Brodatz texture database, and compare our results with the popular methods (see Tab. 1).


Table.1: Classification results with standard deviations on UIUC, UMD and Brodatz databases.

Related Work

* Gui-Song Xia is the corresponding author.