旋转不变纹理分类：局部二进制计数新方法

142 浏览量更新于2024-08-26 收藏 388KB PDF 举报

本文主要探讨了旋转不变纹理分类中的一个重要进展，即"Completed Local Binary Count (CLBC)"。该研究论文由Yang Zhao, De-Shuang Huang 和 Wei Jia三位作者提出，发表于2012年10月的IEEE Transactions on Image Processing第21卷第10期。论文的核心创新是引入了一种新颖的局部描述符——局部二进制计数（LBC），用于提升纹理分类的旋转不变性。 LBC描述符的独特之处在于它专注于提取局部灰度级的二值差异信息，而不是传统的局部二进制模式（LBP）那样同时考虑结构信息。尽管LBC编码并不直接反映视觉微观结构，但通过统计分析，LBC特征能够有效地捕捉到局部纹理的特性。这使得LBC在处理旋转变化的纹理时，具有更高的鲁棒性和稳定性。为了进一步提高纹理分类的性能，论文提出了一个完整版本的LBC，即CLBC。相较于基础的LBC，CLBC可能包含了更多的信息或改进了特征处理方法，从而提高了分类精度。实验部分展示了CLBC在三个不同的纹理数据库上的表现，结果显示，它能够达到与CLBP相当甚至更优的准确分类率，这对于许多依赖于纹理分析的应用，如图像检索、物体识别等具有重要意义。这篇论文对旋转不变纹理分类领域做出了重要贡献，特别是在设计高效、旋转不变的局部特征表示方面。LBC和CLBC算法不仅简化了特征计算，而且提高了分类任务的性能，为后续的研究者提供了新的视角和方法。对于那些关注纹理特征提取和计算机视觉的读者来说，这篇文章提供了深入理解旋转不变纹理分类技术的关键洞察。

4492 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 21, NO. 10, OCTOBER 2012

Correspondence

Completed Local Binary Count for Rotation Invariant

Texture Classiﬁcation

Yang Zhao, De-Shuang Huang, Senior Member, IEEE,

and Wei Jia, Member, IEEE

Abstract—In this brief, a novel local descriptor, named local

binary count (LBC), is proposed for rotation invariant texture

classiﬁcation. The proposed LBC can extract the local binary

grayscale difference information, and totally abandon the local

binary structural information. Although the LBC codes do not

represent visual microstructure, the statistics of LBC features can

represent the local texture effectively. In addition, a completed

LBC (CLBC) is also proposed to enhance the performance of

texture classiﬁcation. Experimental results obtained from three

databases demonstrate that the proposed CLBC can achieve

comparable accurate classiﬁcation rates with completed local

binary pattern.

Index Terms— Local binary pattern (LBP), local binary count

(LBC), rotation invariance, texture classiﬁcation.

I. INTRODUCTION

Texture classiﬁcation is very useful in many applications, such as

remote sensing, biomedical image analysis, image recognition and

retrieval. Thus, it is an important issue in image processing and

computer vision. Generally, texture images captured in the real world

may have obvious orientation variations. Rotation invariant texture

analysis is therefore immensely needed from both the practical and

theoretical viewpoints.

So far, many approaches have been proposed to achieve rotation

invariance for texture classiﬁcation which can be broadly divided into

two categories, i.e., statistical methods and model based methods,

respectively. In statistical methods, texture is generally described by

the statistics of selected features, e.g., invariant histogram, texture

elements, and micro-structures. Davis et al [1] exploited polarograms

and generalized co-occurrence matrices to obtain rotation invariant

statistical features. Duvernoy et al [2] proposed Fourier descriptors

to extract the rotation invariant texture feature on the spectrum

domain. Goyal et al [3] proposed a method by using texel property

histogram. Eichmann et al [4] presented texture descriptors based

on line structures extracted by Hough transform. In model based

methods, texture is usually presented as a probability model or as

Manuscript received October 17, 2011; revised May 2, 2012; accepted

May 20, 2012. Date of publication June 12, 2012; date of current

version September 13, 2012. This work was supported by the National

Science Foundation of China, under Grant 61133010, Grant 31071168, Grant

61005010, Grant 60905023, and Grant 60975005, and the China Post-

Doctoral Science Foundation under Grant 20100480708. The associate editor

coordinating the review of this manuscript and approving it for publication was

Prof. Zhou Wang.

Y. Zhao is with the Department of Automation, University of Science and

Technology of China, Hefei 230027, China, and also with the Hefei Institute

of Intelligent Machines, Chinese Academy of Science, Hefei 230031, China

(e-mail: zyknight@mail.ustc.edu.cn).

D.-S. Huang is with the School of Electronics and Information Engineering,

Tongji University, Shanghai 201804, China (e-mail: dshuang@tongji.edu.cn).

W. Jia is with the Hefei Institute of Intelligent Machines, Chinese Academy

of Science, Hefei 230031, China (e-mail: icg.jiawei@gmail.com).

Digital Object Identiﬁer 10.1109/TIP.2012.2204271

a linear combination of a set of basis functions. Kashyap et al [5]

developed a circular simultaneous autoregressive (CSAR) model for

rotation invariant texture classiﬁcation. Cohen et al [6] characterized

texture as Gaussian Markov random ﬁelds and used the maximum

likelihood to estimate rotation angles. Chen and Kundu [7] addressed

rotation invariant by using multichannel sub-bands decomposition and

hidden Markov model (HMM). Porter et al [8] exploited the wavelet

transform for rotation invariant texture classiﬁcation by using the

Daubechies four-tap wavelet ﬁlter coefﬁcients.

Although these aforementioned methods are proven to be rotation

invariant, they are not very robust to variances of illumination. In [9],

Ojala et al proposed an efﬁcient method, namely Local Binary Pattern

(LBP), for rotation invariant texture classiﬁcation. As shown in Fig. 1,

the algorithm of LBP contains two main steps, i.e., thresholding step

and encoding step. In the thresholding step, the values of neighbor

pixels are turned to binary values (0 or 1) by comparing them with

the central pixel. Obviously, the local binary grayscale difference

information is extracted in the thresholding step. In the encoding step,

the binary numbers are encoded to characterize a structural pattern,

and then the code is transformed into decimal number. Aiming

at achieving rotation invariance, Ojala proposed rotation invariant

uniform LBP (LBP

riu2

), in which only rotation invariant uniform

local binary patterns were selected. It was believed that LBP is an

excellent measure of the spatial structure of local image texture since

it can effectively detect micro-structures (e.g., edges, lines, spots)

information.

Since Ojala’s work [9], a lot of variants of the LBP for rotation

invariant texture classiﬁcation have been proposed, some of which

were focused on how to extract more discriminative patterns.

For example, Heikkila et al [10] proposed center-symmetric

LBP (CS-LBP) by comparing center-symmetric pairs of pixels

instead of comparing neighbors with central pixels. Liao et al [11]

presented Dominant LBP (DLBP), in which dominant patterns were

experimentally chosen from all rotation invariant patterns. Others

tried to further explore the contrast information. For example, Tan

and Triggs [12] proposed the method of Local Ternary Pattern

(LTP), which extends original LBP to 3-valued codes. Recently,

Guo et al [13] proposed the completed LBP (CLBP) by combining

the conventional LBP with the measures of local intensity difference

and central gray level. Khellah [14] presented a new method for

texture classiﬁcation, which combines Dominant Neighborhood

Structure (DNS) and traditional LBP. It should be noticed that the

LBP encoding process is used in all of these variants mentioned

above because it is believed that structural patterns characterized by

the binary codes are more important for rotation invariant texture

recognition while local binary grayscale difference information is

considered to be merely a supplement of micro-structures. Recently,

there is also another way to look at LBP, e.g., LBP is regarded as

a special ﬁlter-based textureo

perator [19], [20].

Then, is micro-structure information really the main role in LBP

for rotation invariant texture recognition? It seems that nobody has

investigated this problem in depth so far. In this paper, we shall try to

address this question by proposing a new local operator that discards

the structural information from LBP operator, which is named as

Local Binary Count (LBC). Experimental results illustrate that the

most discriminative information of local texture for rotation invariant

texture classiﬁcation is not the ‘micro-structures’ information but

the local binary grayscale difference information. Motivated by the

下载后可阅读完整内容，剩余5页未读，立即下载

weixin_38704786

粉丝: 13
资源: 1001

旋转不变纹理分类：局部二进制计数新方法

论文研究-完备局部凸凹计数模式及其旋转不变纹理分类.pdf

局部二进制模式

用于纹理分类的中值鲁棒扩展局部二进制模式

采用完整局部二进制模式的伪装语音检测.docx

ps部分功能的matlab源代码-textureclassification_CLDP_ICIP2016:“用于旋转不变纹理分类的完整局部导数

局部二进制模式矩阵（LBP）：局部二进制模式（LBP）是用于描述图像纹理的方法之一。-matlab开发

局部二进制模式：计算图像 LBP（局部二进制模式）。-matlab开发

BRINT：二进制旋转不变性和耐噪纹理分类

旋转不变纹理分类的局部导数模式源代码

FPGA实现的局部二进制模式实时纹理分类硬件架构

最新资源