知觉驱动的形态学搜索策略提升形状检索效率

20 浏览量更新于2024-08-26 收藏 859KB PDF 举报

"基于感知的形态学检索策略" 本文探讨了一种新的形状检索方法，即基于感知的形态学检索策略（PMMS），旨在提升传统形状匹配方法的检索效率。该方法受到人类感知习惯的启发，通过形态学操作来模拟这两个关键的感知规律，进而优化形状检索性能。首先，作者介绍了两个与形状检索密切相关的感知习惯。这些习惯反映了人类在识别和比较形状时的心理机制，是理解和改进形状匹配算法的重要依据。通过深入理解这些感知规律，可以更好地适应人类视觉系统对形状的识别方式，从而提高检索的准确性和效率。其次，文章提出将这两种感知习惯通过形态学操作进行建模。形态学操作是一种在图像处理和模式识别中广泛使用的工具，包括膨胀、腐蚀、开运算和闭运算等。这些操作可以有效地处理形状的边界、连接性和细节，使得形状特征更加突出，更符合人类的感知习惯。接着，PMMS被应用于改进内部距离形状上下文（IDSC）这一流行形状匹配方法的检索性能。IDSC是一种基于形状轮廓内点到轮廓边缘距离的描述符，它能够捕捉形状的内在结构。通过结合PMMS，IDSC的描述能力得到增强，提高了对形状相似性的判断力。此外，为了进一步提升检索性能，文章还引入了局部约束扩散过程（LCDP）方法。LCDP利用局部信息来传播和扩散形状特征，有助于消除噪声和增强形状的显著性，从而在复杂背景或变形形状的场景下提高检索准确性。实验结果在MPEG-7数据集上显示，采用PMMS和LCDP的组合方法达到了98.56%的检索率，这是一个显著的提升。同时，这种方法还在Swedish Leaf数据集、ETH-80数据集和Natural Silhouette数据集上进行了验证，实验结果均证明了该方法的有效性。这些数据集涵盖了各种形状和复杂度，验证了方法的普适性和鲁棒性。基于感知的形态学检索策略通过结合人类感知习惯、形态学操作和局部约束扩散过程，提供了一种有效的形状检索解决方案，尤其适用于提高形状匹配方法的性能。这种方法不仅在理论上具有创新性，而且在实际应用中展现出卓越的检索效果。

Perceptually motivated morphological strategies for shape retrieval

Rong-Xiang Hu

a,b

, Wei Jia

, Yang Zhao

a,b

, Jie Gui

Hefei Institutes of Physical Science, CAS, P.O. Box 1130, Hefei 230031, China

Department of Automation, University of Science and Technology of China, Hefei 230027, China

article info

Article history:

Received 22 June 2011

Received in revised form

21 February 2012

Accepted 23 February 2012

Available online 5 March 2012

Keywords:

Shape retrieval

Shape contexts

Morphological operation

Perceptual custom

abstract

In this paper, two perceptually motivated morphological strategies (PMMS) are proposed to enhance

the retrieval performance of common shape matching methods. Firstly, two human perception customs

are introduced, which have important relations to shape retrieval. Secondly, these two customs are

properly modeled by morphological operations. Finally, the proposed PMMS is applied to improv e the

retrieval performances of a popular shape matching method named Inner-Distance Shape Contexts

(IDSC), and then the Locally Constrained Diffusion Process (LCDP) method is exploited to further

enhance the retrieval performance. This combination achieves a retrieval rate of 98.56% on MPEG-7

dataset. We also conduct the experiments on Swedish Leaf dataset, the ETH-80 dataset and the Natural

Silhouette dataset . The experimental results obtained from four datasets demonstrate clearly the

effectiveness of the proposed method.

1. Introduction

Shape is one of the most important features of an object.

It plays a key role in human perception, and has been widely

exploited for many computer vision applications such as object

classiﬁcation, object recognition and object retrieval. Generally, in

a shape-based object retrieval application, given a query object,

the most similar objects will be retrieved from a dataset accord-

ing to certain similarity or distance measures, which are gener-

ated by shape matching algorithms. In the past decade, many

shape matching algorithms have been proposed [1–10]. Designing

a suitable shape matching method for robust shape retrieval,

however, is a very difﬁcult task. Some researchers therefore start

to explore new perspectives to enhance the retrieval performances.

Recently, many novel shape retrieval methods providing new

perspectives have been proposed [11–19], and have achieved

promising retrieval performances. Here, for convenience’s sake,

we call them as enhancing methods. Further, they could be divided

into three categories, i.e., context-based, knowledge-based and

fusion-based.

Context-based methods are the most widely studied enhan-

cing methods so far. Traditionally, in a common shape matching

method, matching a pair of shapes would generate a similarity or

distance measure, which has no relationship with other shapes.

Recently, it is believed that all shapes should be considered as a

group rather than pairs of shapes in shape retrieval. Precisely,

when the target similarity or distance measure is computed

between one pair of shapes, all similarity or distance measures

between other shapes within the group could provide useful and

complementary information to further correct the target measure.

In this way, more satisfactory retrieval results could be obtained.

Here, the similarity or distance measures between other shapes

within the group are regarded as the context of the target

measure. Thus, those methods based on this strategy are called

as context-based methods. Yang et al. [11,15] proposed a method

to improve the shape retrieval performance through graph

transduction by propagating the model through all existing

shapes. They took advantage of the manifold formed by the

existing shapes, and used an unsupervised graph transduction

approach to learn a better measure for retrieval. Kontschieder

et al. [12] proposed a modiﬁed mutual K nearest neighbor graph

to consider the underlying structure of the shape manifold.

The manifold is estimated from the shape similarity measures

between all the shapes within a dataset. Yang et al. [13] later

proposed the method of Locally Constrained Diffusion Process

(LCDP) to better enhance the retrieval performance, in which the

inﬂuence of other shapes is propagated as a diffusion process on a

graph formed by the given set of shapes. However, the classical

diffusion process is unstable for sparse space, so they add some

local constraints using K nearest neighbor graph to achieve a

more robust diffusion process. Egozi et al. [16] proposed a meta-

shape-similarity approach to characterize a given shape by its

similarity to its K nearest neighbors. This approach does not

propagate similarities, but aims to compare local graph structures

as intrinsic similarity measures. Recently, Yang et al. [19]

proposed a novel afﬁnity learning method using Tensor Product

Contents lists available at SciVerse ScienceDirect

journal homepage: www.elsevier.com/locate/pr

Pattern Recognition

doi:10.1016/j.patcog.2012.02.020

Corresponding author. Tel./fax: þ 86 5515591108.

E-mail address: icg.jiawei@gmail.com (W. Jia).

Pattern Recognition 45 (2012) 3222–3230

下载后可阅读完整内容，剩余8页未读，立即下载

weixin_38538381

粉丝: 6

知觉驱动的形态学搜索策略提升形状检索效率

ETH 行人视频数据数据集

ETH3D双目数据集

基于热映像的图像分割算法可让计算机迅速识别物体外形

基于颜色粗糙度与空间连通性的彩色图像智能分割算法

【信号处理艺术】：时域与频域信号增强策略

图像处理技术大揭秘：MATLAB图像增强与识别策略

【SDB访问控制机制】：精细化权限管理的实施策略

高维空间的几何艺术：超越三维的视觉与计算策略

HALCON图像分析进阶指南：边缘检测到特征提取的终极策略

TRDP数据采集与处理：10大策略助你实现高效实验设计

最新资源