SelectiveSearchforObjectRecognition资源-CSDN下载

需积分: 15 116 浏览量 2018-11-13 11:35:37 上传评论收藏 1.34MB PDF 举报

This paper addresses the problem of generating possible object locations for use in object recognition. We introduce selective search which combines the strength of both an exhaustive search and segmentation. Like segmentation, we use the image structure to guide our sampling process. Like exhaustive search, we aim to capture all possible object locations. Instead of a single technique to generate possible object locations, we diversify our search and use a variety of complementary image partitionings to deal with as many image conditions as possible. Our selective search results in a small set of data-driven, class-independent, high quality locations, yielding 99% recall and a Mean Average Best Overlap of 0.879 at 10,097 locations. The reduced number of locations compared to an exhaustive search enables the use of stronger machine learning techniques and stronger appearance models for object recognition. In this paper we show that our selective search enables the use of the powerful Bag-of-Words model for recognition. The selective search software is made publicly available (Software: http://disi.unitn.it/similar to uijlings/SelectiveSearch.html). ### Selective Search for Object Recognition #### 摘要与引言本文主要研究了在对象识别过程中生成可能的对象位置的问题。作者们提出了“选择性搜索”（Selective Search）这一概念，该方法结合了穷举搜索（Exhaustive Search）与分割（Segmentation）的优点。像分割一样，选择性搜索利用图像结构来指导抽样过程；像穷举搜索一样，其目标是捕捉所有可能的对象位置。选择性搜索并非采用单一技术生成可能的对象位置，而是多样化地进行搜索，并运用多种互补的图像分区方式来处理尽可能多的图像条件。结果是产生了一组数量较少、数据驱动、类独立且高质量的位置，能够达到99%的召回率以及10,097个位置上的平均最佳重叠均值（Mean Average Best Overlap）为0.879。相比传统的穷举搜索，这种减少的位置数量使得可以应用更强大的机器学习技术和更强的对象外观模型来进行对象识别。文章进一步展示了选择性搜索如何使得能够应用强大的“词袋模型”（Bag-of-Words Model）进行识别。 #### 方法论选择性搜索的核心在于它不仅考虑了图像中的单个像素，还考虑了图像的内在层次结构。例如，在一张包含沙拉碗、勺子和桌子的照片中，选择性搜索会识别出沙拉碗包含了沙拉和勺子，而沙拉碗本身又位于桌子上。通过这种方式，选择性搜索能够在不同层次上识别出对象。 #### 关键特点 1. **数据驱动**: 选择性搜索是一种基于数据的方法，它不依赖于预定义的规则或模板。 2. **类独立**: 这种搜索方法对于不同的对象类别具有普遍适用性，无需特定于某个类别的训练。 3. **高质量定位**: 产生的对象位置质量高，能够准确捕捉到目标物体的位置信息。 4. **高效性**: 相比于穷举搜索，选择性搜索极大地减少了需要检查的位置数量，从而提高了效率。 5. **适用性强**: 减少的位置数量使得可以应用更复杂的机器学习算法和模型。 #### 应用示例文章通过实验证明了选择性搜索方法的有效性。其中一个例子是将选择性搜索与词袋模型相结合用于对象识别。词袋模型是一种广泛应用于文本分类和检索的强大工具，它通过提取特征向量来表示图像内容。通过选择性搜索获得的高质量位置信息，可以进一步增强词袋模型的表现，提高对象识别的准确性。 #### 软件可用性为了便于研究者和开发者使用选择性搜索方法，作者们提供了公开可用的选择性搜索软件。该软件可以在以下网址获取：[http://disi.unitn.it/~uijlings/SelectiveSearch.html](http://disi.unitn.it/~uijlings/SelectiveSearch.html)。 #### 结论选择性搜索是一种创新的方法，它解决了对象识别中一个关键问题——如何有效地生成候选对象区域。通过结合分割和穷举搜索的优点，这种方法能够在保持高准确度的同时减少计算资源的需求。此外，选择性搜索还可以与其他高级技术如词袋模型结合使用，进一步提升识别效果。未来的研究方向可能包括探索更多样化的图像分区策略、改进搜索算法以及将其应用于更广泛的视觉任务中。

资源推荐

资源详情

资源评论

Int J Comput Vis (2013) 104:154–171

DOI 10.1007/s11263-013-0620-5

Selective Search for Object Recognition

J. R. R. Uijlings · K. E. A. van de Sande ·

T. Gevers · A. W. M. Smeulders

Received: 5 May 2012 / Accepted: 11 March 2013 / Published online: 2 April 2013

Abstract This paper addresses the problem of generating

possible object locations for use in object recognition. We

introduce selective search which combines the strength of

both an exhaustive search and segmentation. Like segmen-

tation, we use the image structure to guide our sampling

process. Like exhaustive search, we aim to capture all possi-

ble object locations. Instead of a single technique to generate

possible object locations, we diversify our search and use a

variety of complementary image partitionings to deal with

as many image conditions as possible. Our selective search

results in a small set of data-driven, class-independent, high

quality locations, yielding 99 % recall and a Mean Average

Best Overlap of 0.879 at 10,097 locations. The reduced num-

ber of locations compared to an exhaustive search enables

the use of stronger machine learning techniques and stronger

appearance models for object recognition. In this paper we

show that our selective search enables the use of the powerful

Bag-of-Words model for recognition. The selective search

software is made publicly available (Software: http://disi.

unitn.it/~uijlings/SelectiveSearch.html).

1 Introduction

For a long time, objects were sought to be delineated before

their identiﬁcation. This gave rise to segmentation, which

aims for a unique partitioning of the image through a generic

algorithm, where there is one part for all object silhouettesin

the image. Research on this topic has yielded tremendous

J. R. R. Uijlings (

)

University of Trento, Trento, Italy

e-mail: jrr[email protected]; [email protected]

K. E. A. van de Sande · T. Gevers · A. W. M. Smeulders

University of Amsterdam, Amsterdam, The Netherlands

progress over the past years ( Arbeláez et al. 2011; Comaniciu

and Meer 2002; Felzenszwalb and Huttenlocher 2004; Shi

and Malik 2000). But images are intrinsically hierarchical: In

Fig. 1a the salad and spoons are inside the salad bowl, which

in turn stands on the table. Furthermore, depending on the

context the term table in this picture can refer to only the wood

or include everything on the table. Therefore both t he nature

of images and the different uses of an object category are

hierarchical. This prohibits the unique partitioning of objects

for all but the most speciﬁc purposes. Hence for most tasks

multiple scales in a segmentation are a necessity. This is most

naturally addressed by using a hierarchical partitioning, as

done for example by Arbeláez et al. (2011).

Besides that a segmentation should be hierarchical, a

generic solution for segmentation using a single s trategy may

not exist at all. There are many conﬂicting reasons why a

region should be grouped together: In Fig. 1b the cats can

be separated using colour, but their texture is the same. Con-

versely, in Fig. 1c the chameleon is similar to its surrounding

leaves in terms of colour, yet its texture differs. Finally, in

Fig. 1d, the wheels are wildly different from the car in terms

of both colour and texture, yet are enclosed by the car. Indi-

vidual visual features therefore cannot resolve the ambiguity

of segmentation.

And, ﬁnally, t here i s a more fundamental problem.

Regions with very different characteristics, such as a face

over a sweater, can only be combined into one object after

it has been established that the object at hand is a human.

Hence without prior recognition it is hard to decide that a

face and a sweater are part of one object ( Tu et al. 2005).

This has led to the opposite of the traditional approach:

to do localisation through the identiﬁcation of an object.

This recent approach in object recognition has made enor-

mous progress in less than a decade ( Dalal and Triggs 2005;

Felzenszwalb et al. 2010; Harzallah et al. 2009; Viola and

123

Int J Comput Vis (2013) 104:154–171 155

Fig. 1 There is a high variety of reasons that an image region forms

an object. In (b) the cats can be distinguished by colour, not texture. In

texture, not colour. In (d) the wheels can be part of the car because they

are enclosed, not because they are similar in texture or colour. There-

fore, to ﬁnd objects in a structured way it is necessary to use a variety

of diverse strategies. Furthermore, an image is intrinsically hierarchical

as there is no single scale for which the complete table, salad bowl, and

salad spoon can be found in (a)

Jones 2001). With an appearance model learned from exam-

ples, an exhaustive search is performed where every location

within the image is examined as to not miss any potential

object location ( Dalal and Triggs 2005; Felzenszwalb et al.

2010; Harzallah et al. 2009; Viola and Jones 2001).

However, the exhaustive search itself has several draw-

backs. Searching every possible location is computationally

infeasible. The search space has to be reduced by using a reg-

ular grid, ﬁxed scales, and ﬁxed aspect ratios. In most cases

the number of locations to visit remains huge, so much that

alternative restrictions need to be imposed. The classiﬁer is

simpliﬁed and the appearance model needs to be fast. Fur-

thermore, a uniform sampling yields many boxes for which it

is immediately clear that they are not supportive of an object.

Rather then sampling locations blindly using an exhaustive

search, a key question is: Can we steer the sampling by a

data-driven analysis?

In this paper, we aim to combine the best of the intu-

itions of segmentation and exhaustive search and propose a

data-driven selective search. Inspired by bottom-up segmen-

tation, we aim to exploit the structure of the image to gener-

ate object locations. Inspired by exhaustive search, we aim

to capture all possible object locations. Therefore, instead of

using a single sampling technique, we aim to diversify the

sampling techniques to account for as many image condi-

tions as possible. Speciﬁcally, we use a data-driven grouping-

based strategy where we increase diversity by using a variety

of complementary grouping criteria and a variety of comple-

mentary colour spaces with different invariance properties.

The set of locations is obtained by combining the locations of

these complementary partitionings. Our goal is to generate a

class-independent, data-driven, selective search strategy that

generates a small set of high-quality object locations.

Our application domain of selective search is object recog-

nition. We therefore evaluate on the most commonly used

dataset for this purpose, the Pascal VOC detection challenge

which consists of 20 object classes. The size of this dataset

yields computational constraints for our selective search. Fur-

thermore, the use of this dataset means that the quality of

locations is mainly evaluated in terms of bounding boxes.

However, our selective search applies to regions as well and

is also applicable to concepts such as “grass”.

In this paper we propose selective search for object

recognition. Our main research questions are: (1) What are

good diversiﬁcation strategies for adapting segmentation as

a selective search strategy? (2) How effective is selective

search in creating a small set of high-quality locations within

an image? (3) Can we use selective search to employ more

powerful classiﬁers and appearance models for object recog-

nition?

2 Related Work

We conﬁne the related work to the domain of object recog-

nition and divide it into three categories: Exhaustive search,

segmentation, and other sampling strategies that do not fall

in either category.

2.1 Exhaustive Search

As an object can be located at any position and scale in the

image, it is natural to search everywhere ( Dalal and Triggs

2005; Harzallah et al. 2009; Viola and Jones 2004). How-

ever, the visual search space is huge, making an exhaustive

search computationally expensive. This imposes constraints

on the evaluation cost per location and/or the number of loca-

tions considered. Hence most of these sliding window tech-

niques use a coarse search grid and ﬁxed aspect ratios, using

weak classiﬁers and economic image features such as HOG

( Dalal and Triggs 2005; Harzallah et al. 2009; Viola and

Jones 2004). This method is often used as a preselection step

in a cascade of classiﬁers ( Harzallah et al. 2009; Viola and

Jones 2004).

Related to the sliding window technique is the highly

successful part-based object localisation method of Felzen-

szwalb et al. (2010). Their method also performs an exhaus-

tive search using a linear SVM and HOG features. However,

they search for objects and object parts, whose combination

results in an impressive object detection performance.

123

156 Int J Comput Vis (2013) 104:154–171

Lampert et al. (2009) proposed using the appearance

model to guide the search. This both alleviates the constraints

of using a regular grid, ﬁxed scales, and ﬁxed aspect ratio,

while at the same time reduces the number of locations vis-

ited. This is done by directly searching for the optimal win-

dow within the image using a branch and bound technique.

While they obtain impressive results for linear classiﬁers,

Alexe et al. (2010) found that for non-linear classiﬁers the

method in practice still visits over a 100,000 windows per

image.

Instead of a blind exhaustive search or a branch and bound

search, we propose selective search. We use the underly-

ing image structure to generate object locations. In contrast

to the discussed methods, this yields a completely class-

independent set of locations. Furthermore, because we do not

use a ﬁxed aspect ratio, our method is not limited to objects

but should be able to ﬁnd stuff like “grass” and “sand” as well

(this also holds for Lampert et al. (2009)). Finally, we hope

to generate fewer locations, which should make the prob-

lem easier as the variability of samples becomes lower. And

more importantly, it frees up computational power which can

be used for stronger machine learning techniques and more

powerful appearance models.

2.2 Segmentation

Both Carreira and Sminchisescu (2010) and Endres and

Hoiem (2010) propose to generate a set of class independent

object hypotheses using segmentation. Both methods gen-

erate multiple foreground/background segmentations, learn

to predict the likelihood that a foreground segment is a

complete object, and use this to rank the segments. Both

algorithms show a promising ability to accurately delineate

objects within images, conﬁrmed by Li et al. (2010)who

achieve state-of-the-art results on pixel-wise image classi-

ﬁcation using Carreira and Sminchisescu (2010). As com-

mon in segmentation, both methods rely on a single strong

algorithm for identifying good regions. They obtain a vari-

ety of locations by using many randomly initialised fore-

ground and background seeds. In contrast, we explicitly deal

with a variety of image conditions by using different group-

ing criteria and different representations. This means a lower

computational investment as we do not have to invest in the

single best segmentation strategy, such as using the excel-

lent yet expensive contour detector of Arbeláez et al. (2011).

Furthermore, as we deal with different image conditions sep-

arately, we expect our locations to have a more consistent

quality. Finally, our selective search paradigm dictates that

the most interesting question is not how our regions com-

pare to Carreira and Sminchisescu (2010), Endres and Hoiem

(2010), but rather how they can complement each other.

Gu et al. (2009) address the problem of carefully segment-

ing and recognizing objects based on their parts. They ﬁrst

generate a set of part hypotheses using a grouping method

based on Arbeláez et al. (2011). Each part hypothesis is

described by both appearance and shape features. Then, an

object is recognized and carefully delineated by using its

parts, achieving good results for shape recognition. In their

work, the segmentation is hierarchical and yields segments

at all scales. However, they use a single grouping strategy

whose power of discovering parts or objects is left unevalu-

ated. In this work, we use multiple complementary strategies

to deal with as many image conditions as possible. We include

the locations generated using Arbeláez et al. (2011) in our

evaluation.

2.3 Other Sampling Strategies

Alexe et al. (2012) address the problem of the large sam-

pling space of an exhaustive search by proposing to search

for any object, independent of its class. In their method they

train a classiﬁer on the object windows of those objects

which have a well-deﬁned shape (as opposed to stuff like

“grass” and “sand”). Then instead of a full exhaustive search

they randomly sample boxes to which they apply their

classiﬁer. The boxes with the highest “objectness” mea-

sure serve as a set of object hypotheses. This set is then

used to greatly reduce the number of windows evaluated by

class-speciﬁc object detectors. We compare our method with

their work.

Another strategy is to use visual words of the Bag-

of-Words model to predict the object location. Vedaldi (2009)

use jumping windows (Chum and ZissermanZ 2007), in

which the relation between individual visual words and the

object location is learned to predict the object location in

new images. Maji and Malik (2009) combine multiple of

these relations to predict the object location using a Hough-

transform, after which they randomly sample windows close

to the Hough maximum. In contrast to learning, we use the

image structure to sample a set of class-independent object

hypotheses.

To summarize, our novelty is as follows. Instead of an

exhaustive search ( Dalal and Triggs 2005; Felzenszwalb

et al. 2010; Harzallah et al. 2009; Viola and Jones 2004)

we use segmentation as selective search yielding a small

set of class independent object locations. In contrast to the

segmentation of Carreira and Sminchisescu (2010), Endres

and Hoiem (2010) instead of focusing on the best segmen-

tation algorithm ( Arbeláez et al. 2011), we use a variety of

strategies to deal with as many image conditions as possible,

thereby severely reducing computational costs while poten-

tially capturing more objects accurately. Instead of learning

an objectness measure on randomly sampled boxes ( Alexe

et al. 2012), we use a bottom-up grouping procedure to gen-

erate good object locations.

123

剩余17页未读，继续阅读

评论收藏

内容反馈

yifanwz

粉丝: 1

Selective Search for Object Recognition

Object Recognition

Selective Search for Object Recognition算法详解

Visual Object Recognition, Kristen Grauman

object_recognition:3D对象识别相关软件包

Selective Search for Object Recognition-附件资源

Indoor Scene Recognition by 3-D Object Search

selectiveSearchDraft.pdf

detection papers in deeplearning

R-CNN (R-CNN) 相关的三篇论文

机器视觉领域经典文章及其部分中文翻译

深度学习典型应用类论文汇总

计算机视觉算法实战书籍推荐-计算机视觉一些项目实战技术 计算机视觉.pdf

Android代码-对象检测识别

Objects_recognition

基于目标检测模型的人脸识别技术.pdf

SPP-Net演示文档

pdf转为扫描件，免费

DeepSeek从入门到精通-清华大学-202502.pdf

YOLOv8-deepsort 实现智能车辆目标检测+车辆跟踪+车辆计数

YOLOv8网络结构图，自制visio文件，yolov8.vsds，需要的自取，在原有的基础上直接改就行了

yolov8(2023年8月版本),已经下好yolov8s.pt和yolov8n.pt

DEEP SEEK 本地部署（Ollama + ChatBox）+ 私有知识库（cherry studio）教程

Transformer模型实现长期预测并可视化结果（附代码+数据集+原理介绍）

社交平台上经济类话题的文章热度信息，数据是真实的，但不是真实日期

行人跌倒数据集（VOC格式）

CIFAR10数据集免费下载

DeepSeek从入门到精通-清华大学

Unity：碰撞检测方法

量子计算与信息：从入门到精通

最新资源

计算机视觉算法实战书籍推荐-计算机视觉一些项目实战技术计算机视觉.pdf