Pairwise Rotation Invariant Co-occurrence Local Binary Pattern

Xianbiao Qi, Rong Xiao, Chun-Guang Li, Yu Qiao, Jun Guo , Xiaoou Tang

Beijing University of Posts and Telecommunications, Microsoft Coroperation,

Shenzhen Institutes of Advanced Technology of Chinese Academy of Sciences, The Chinese University of Hong Kong.

Abstract

Designing effective feature is a fundamental problem in computer vision. However, it is usually difficult to achieve great trade-off between the discriminative power and transformation invariance. Spatial co-occurrence could boost the discriminative power of the features, but it always suffers from the geometric and photometric variations. In this work, we investigate rotation invariant property of co-occurrence feature, and introduce a novel pairwise rotation invariant co-occurrence local binary pattern (PRI-CoLBP) feature which incorporates two types of context, spatial co-occurrence and orientation co-occurrence. Different from traditional rotation invariant local features, pairwise rotation invariant co-occurrence features preserve the relative angles between the orientations of individual features. The relative angle depicts the local curvature information, which is discriminative.

The proposed PRI-CoLBP is computationally efficient and has been applied to six types of applications, including texture classification, material classification, flower recognition, leaf recognition, food recognition and scene recognition. Superior performances on such applications demonstrate the effectiveness of the proposed PRI-CoLBP.

 

Paper

Xianbiao Qi, Rong Xiao, Chun-Guang Li, Yu Qiao, Jun Guo, Xiaoou Tang. Pairwise Rotation Invariant Co-occurrence Local Binary Pattern. Published on IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2014. pdf .

 

Features

Discriminative power and transformation invariance are the two most important properties of local features. Spatial co-occurrence could greatly boost the discriminative power of the features, but its transformation invariant property is hard to obtain. To achieve transformation invariance for co-occurrence feature, we should promise the following two conditions:

  1. The correspondence of described points set should be promised under different image transformations.
  2. The descriptor for points set is transformation invariant.

In practice, we densely sample each points of the images. For the same point A in the same scene under different image transformation, we should promise that the same points set {Bi} could be uniquely determined.

When the correspondence of described points is promised, we should promise that the descriptor for the same points set is transformation invariant.

 

Properties

The proposed feature has the following properties:

 

Source Code

Source Code Download               Source Code Download (Access Link From China)

With the provided source code, you could easily run and get our results.

 

Applications

The proposed feature is extremely effective on a lot of applications, such as texture classification, material recognition, flower recognition, leaf recognition, food recognition and scene classification.

In our work, we test our feature on eight datasets, including Brodatz, CUReT, and KTH-TIPS, Flickr Material Database, Oxford Flower 102, Swedish Leaf Database, Food database and Scene-15 database.

The following table presents the detailed information for each database.
Databases

Detailed experimental results and analysis for all applications can be found at follows:

Texture Classification        Material Recognition               Flower Recognition

Leaf Recognition                Food Recognition                     Scene Classification

 

Feature Matrices

The feature matrices are available at This Link.

 

Relevant Toolboxs

VLFeat.

LibSVM.

 

Acknowledgments

This work was supported by National Natural Science Foundation of China(Grant No.61005004 and 61175011),the 111 project(Grant No.B08004), and the Fundamental Research Funds for the Central Universities(Grant No.2012RC0108).