使用选择性搜索进行目标识别
需积分: 15 22 浏览量
更新于2024-07-18
收藏 1.34MB PDF 举报
"Selective Search for Object Recognition"
这篇论文《Selective Search for Object Recognition》探讨了在对象识别中生成可能的目标位置的问题。作者提出了一种称为“选择性搜索”的方法，它结合了详尽搜索和分割技术的优点。与图像分割类似，该方法利用图像结构来指导采样过程；同时，如同详尽搜索，目标是捕获所有可能的对象位置。通过采用多种互补的图像分割策略，选择性搜索旨在应对各种图像条件。
选择性搜索的结果是生成了一小批数据驱动、类别独立的高质量位置，召回率高达99%，平均最佳重叠度达到0.879，在10,097个位置上达到了这一指标。与详尽搜索相比，减少的位置数量使得可以使用更强大的机器学习技术和更强的外观模型来进行对象识别。论文中展示了选择性搜索使得强大的“词袋”模型能够应用于识别任务。
论文作者包括J.R.R. Uijlings、K.E.A. van de Sande、T. Gevers和A.W.M. Smeulders，该研究于2012年5月提交，2013年3月接受，并于同年4月在线发表。这项工作受到了Springer Science+Business Media New York的版权保护。
选择性搜索的软件已公开提供（软件链接：http://disi.unitn.it/similar to uijlings/SelectiveSearch.html），这为研究者和开发者提供了实现和应用该方法的便利。
在对象识别领域，选择性搜索是一个重要的步骤，它可以有效地减少候选区域的数量，从而提高后续处理的效率和准确性。通过减少计算负担，这种方法使得复杂的视觉模型如“词袋”模型能在实际应用中得以使用，提高了整体的识别性能。在计算机视觉和图像处理领域，这种技术对于自动目标检测和识别系统的设计具有重大意义。
Int J Comput Vis (2013) 104:154–171 157
Fig. 2 Two examples of our selective search showing the necessity of different scales. On the left we find many objects at different scales. On the
right we necessarily find the objects at different scales as the girl is contained by the tv
3 Selective Search
In this section we detail our selective search algorithm for
object recognition and present a variety of diversification
strategies to deal with as many image conditions as possible.
A selective search algorithm is s ubject to the following design
considerations:
Capture All Scales. Objects can occur at any scale within
the image. Furthermore, some objects have less clear
boundaries then other objects. Therefore, in selective
search all object scales have to be taken into account,
as illustrated in Fig. 2. This is most naturally achieved by
using an hierarchical algorithm.
Diversification. There is no single optimal strategy to
group regions together. As observed earlier in Fig. 1,
regions may f orm an object because of only colour, only
texture, or because parts are enclosed. Furthermore, light-
ing conditions such as shading and the colour of the light
may influence how regions form an object. Therefore
instead of a single strategy which works well in most
cases, we want to have a diverse set of strategies to deal
with all cases.
Fast to Compute. The goal of selective search is to yield
a set of possible object locations for use in a practical
object recognition framework. The creation of this set
should not become a computational bottleneck, hence
our algorithm should be reasonably fast.
3.1 Selective Search by Hierarchical Grouping
We take a hierarchical grouping algorithm to form the basis
of our selective search. Bottom-up grouping is a popu-
lar approach to segmentation (Comaniciu and Meer 2002;
Felzenszwalb and Huttenlocher 2004), hence we adapt it for
selective search. Because the process of grouping itself is
hierarchical, we can naturally generate locations at all scales
by continuing the grouping process until the whole image
becomes a single region. This satisfies the condition of cap-
turing all scales.
As regions can yield richer information than pixels, we
want to use region-based features whenever possible. To get
a set of small starting regions which ideally do not span
multiple objects, we use the fast method of Felzenszwalb
and Huttenlocher (2004), which Arbeláez et al. (2011) found
well-suited for such purpose.
Our grouping procedure now works as follows. We first
use (Felzenszwalb and Huttenlocher 2004) to create initial
regions. Then we use a greedy algorithm to iteratively group
regions together: First the similarities between all neighbour-
ing regions are calculated. The two most similar regions are
grouped together, and new similarities are calculated between
the resulting region and its neighbours. The process of group-
ing the most similar r egions is repeated until the whole image
becomes a single region. The general method is detailed in
Algorithm 1.
For the similarity s(r
i
, r
j
) between region r
i
and r
j
we
want a variety of complementary measures under the con-
straint that they are fast to compute. In effect, this means that
the similarities should be based on features that can be prop-
agated through the hierarchy, i.e. when merging region r
i
and
r
j
into r
t
, the features of region r
t
need to be calculated from
the features of r
i
and r
j
without accessing the image pixels.
3.2 Diversification Strategies
The second design criterion for selective search is to diver-
sify the sampling and create a set of complementary strategies
whose locations are combined afterwards. We diversify our
selective search (1) by using a variety of colour spaces with
123
剩余17页未读，继续阅读
156 浏览量
185 浏览量
142 浏览量
2024-10-10 上传
2023-05-31 上传
101 浏览量
125 浏览量
