基于凸包的前景感知显著性检测算法

136 浏览量更新于2024-08-26 收藏 1.28MB PDF 举报

"前景对象感知以进行显着性检测" 这篇研究论文主要探讨了一种新的显着性检测算法，称为“前景对象感知”。该算法旨在改进现有的显着性检测技术，特别是那些依赖边界先验的算法。传统的算法通常假设图像边缘周围区域为背景，但这种方法可能无法准确地捕捉到图像中的显著对象。作者包括来自上海交通大学计算机科学与工程系的Hengliang Zhu、Bin Sheng、Lizhuang Ma和University of Shanghai for Science and Technology的Xiao Lin、Yangyang Hao。他们提出了一种快速且有效的显着性检测方法，该方法首先通过Harris角点检测来近似定位前景对象，并利用这些角点形成凸包。这种方法的优势在于，它可以更精确地识别出图像中的前景元素。在定位了前景对象后，算法将不同区域的显著性值分为两部分，分别生成前景和背景的提示图。这些提示图随后被合并成一个凸包优先图，这有助于区分图像的显著部分和非显著部分。接着，研究者引入了一个基于距离到凸包中心的新先验，用以替换传统的中心先验。这是因为靠近凸包中心的像素往往更有可能属于显著对象。最后，结合凸包优先图和凸包距离先验，算法能够生成一个综合的显着性地图，从而更准确地标识出图像中的突出对象。这种方法不仅考虑了对象的边界，还考虑了对象的几何形状和位置信息，提高了显着性检测的准确性。这篇论文为显着性检测提供了一个创新的视角，通过更好地理解和利用前景对象，提高了算法在复杂场景下的性能。这对于视频分析、图像分割、目标检测等领域的应用具有重要意义。

Foreground Object Sensing for Saliency Detection

Hengliang Zhu

, Bin Sheng

, Xiao Lin

, Yangyang Hao

, Lizhuang Ma

Dept. of Computer Science and Engineering, Shanghai Jiao Tong University, China

School of Optical-electr ical and Computer Engineering, University of Shanghai for science and technology, China

{hengliang_zhu, haoyangyang2014}@sjtu.edu.cn, lin6008@126.com, {shengbin,

ma-lz}@cs.sjtu.edu.cn

ABSTRACT

Many state-of-the-art saliency detection algorithms rely on the bound-

ary prior, but these algorithms simply suppose the boundaries around

an image as background regions. Here we propose a fast and effec-

tive algorithm for salient object detection. First, a novel method is

proposed to approximately locate the foreground object by using

the convex hull from Harris corner. On this basis, we divide the

saliency values of different regions into two parts and generate the

corresponding cue maps (foreground and background), which are

combined into a convex hull prior map. Then a new prior based

on distance to the convex hull center is proposed to replace the

center prior. Finally, the convex hull prior map and the convex hul-

l center-biased map are combined to be the saliency map, which

is then optimized to get the ﬁnal result. Compared with eighteen

existing algorithms and tested on several datasets, the present algo-

rithm performs well in terms of precision and recall.

CCS Concepts

•Computing methodologies → Interest point and salient region

detections;

Keywords

Foreground object; Harris corner; Convex hull; Saliency map

1. INTRODUCTION

Following the study on the information processing mechanism

of the biological (human) vision system, the task of visual saliency

detection has rapidly progressed in recent years and is successfully

used in numerous computer vision applications, such as object seg-

mentation [34], image compression [13, 23], image resizing [29],

image retrieval [8,35], and image quality assessment [46]. The aim

of “visual saliency” is to ﬁnd the abundant information regions of

an image, where the salient regions are perceptually distinguishable

from non-salient regions.

From the psychophysical perspective [22], saliency detection can

be classiﬁed into bottom-up (data-driven) methods [7,12,15,32,39,

45,48] and top-down (goal-driven) methods [20,27,47]. In contrast

Permission to make digital or hard copies of all or part of this work for personal or

classroom use is granted without fee provided that copies are not made or distributed

for proﬁt or commercial advantage and that copies bear this notice and the full cita-

tion on the ﬁrst page. Copyrights for components of this work owned by others than

ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or re-

publish, to post on servers or to redistribute to lists, requires prior speciﬁc permission

and/or a fee. Request permissions from permissions@acm.org.

ICMR’16, June 06-09, 2016, New York, NY, USA

 2016 ACM. ISBN 978-1-4503-4359-6/16/06. . . $15.00

DOI: http://dx.doi.org/10.1145/2911996.2912008

to the latter taking advantage of high-level prior knowledge of an

image, the former makes use of low-level features of the image (e.g.

color, location, and texture) to detect saliency regions. With these

primitive features, in the previous works [1, 18, 28, 37, 48], prior

assumptions, such as the center prior, contrast prior, and boundary

prior, were raised to improve the accuracy of salient object detec-

tion. In this paper, we mainly focus on the bottom-up saliency

detection.

In recent years, the image boundary prior is adopted for saliency

detection, assuming that a narrow border of the image is the non-

salient regions. In other words, the image boundary tends to be

background. There are many state-of-the-art algorithms rely on the

boundary prior [19,41,45,48], which achieves better performance.

However, one limitation of these works is that the salient object

may slightly touch the image edge. In other words, the detected

salient object is incomplete at the image boundary regions.

To address this problem, we propose a simple and effective algo-

rithm for saliency region detection (see the workﬂow inFigure 1).

First, based on the corner detection algorithm (Harris corner [40]),

a new foreground object location algorithm is proposed, which us-

es two convex hull intersections to approximately locate the fore-

ground object. The intersecting region excludes more scattered

background and concentrates on the surrounding of a salient ob-

ject. Second, by incorporating global contrast cues, we introduce a

new region (image patch) saliency computation step (elaborated in

Section 3). We divide the saliency value of a region into two parts:

one part inside of convex hull (salient regions, which may contain

some background) and the other part outside of convex hull (back-

ground regions). They are used to generate two cue maps (fore-

ground and background) that are combined into a convex hull prior

map. Finally, the saliency map is optimized.

The remainder of the article is organized as follows. Section 2

reviews the related work. The details of the proposed saliency com-

putation method is presented in Section 3 and Section 4. Section 5

presents the experimental results and discussions. The conclusions

are given in Section 6.

2. RELATED WORKS

A detailed summary of saliency detection is provided in [4, 5].

The ﬁxation prediction models [3, 11, 14, 16], which track the eye

movements, are based on the exploitation of physical devices. In-

stead, this work is based on the study of image processing. In this

section, we brieﬂy review the major salient detection algorithms

and their differences.

The pioneering research about saliency detection was conducted

by Itti et al. [17], who deﬁned the saliency value of a pixel as the

local center-surround difference. The detection is based on a multi-

scale of image features, including color, intensity, and orientation.

111

下载后可阅读完整内容，剩余7页未读，立即下载

weixin_38656364

粉丝: 8
资源: 897

基于凸包的前景感知显著性检测算法

色量与色差结合的显着区域检测方法

深度感知的双目视觉显著区域检测方法

利用弱标签数据提升显著性对象定位

基于对象轮廓提取的显着性检测和区域自动增长

基于超像素的显着性检测

无监督联合显着区域检测和对象分割

基于双目视觉的显着区域检测

RGBD显着物体检测：基准和算法

用于显着目标检测的非本地深度特征

通过学习准确的背景模板来检测显着区域

最新资源