=Paper=
{{Paper
|id=Vol-1391/30-CR
|storemode=property
|title=A Comparative Study of Fine-grained Classification Methods in the Context of the LifeCLEF Plant Identification Challenge 2015
|pdfUrl=https://ceur-ws.org/Vol-1391/30-CR.pdf
|volume=Vol-1391
|dblpUrl=https://dblp.org/rec/conf/clef/ChampLSJ15
}}
==A Comparative Study of Fine-grained Classification Methods in the Context of the LifeCLEF Plant Identification Challenge 2015==
https://ceur-ws.org/Vol-1391/30-CR.pdf
A comparative study of fine-grained
classification methods in the context of the
LifeCLEF plant identification challenge 2015
Julien Champ1,2 , Titouan Lorieul1,2 , Maximilien Servajean1,2 , and Alexis
Joly1,2
1
Inria ZENITH team, France, name.surname@inria.fr
2
LIRMM, Montpellier, France
Abstract. This paper describes the participation of Inria to the plant
identification task of the LifeCLEF 2015 challenge. The aim of the task
was to produce a list of relevant species for a large set of plant obser-
vations related to 1000 species of trees, herbs and ferns living in West-
ern Europe. Each plant observation contained several annotated pictures
with organ/view tags: Flower, Leaf, Fruit, Stem, Branch, Entire, Scan
(exclusively of leaf). To address this challenge, we experimented two
popular families of classification techniques, i.e. convolutional neural net-
works (CNN) on one side and fisher vectors-based discriminant models
on the other side. Our results show that the CNN approach achieves
much better performance than the fisher vectors. Beyond, we show that
the fusion of both techniques, based on a Bayesian inference using the
confusion matrix of each classifier, did not improve the results of the
CNN alone.
Keywords: LifeCLEF, plant, leaves, leaf, flower, fruit, bark, stem, branch,
species, retrieval, images, collection, species identification, citizen-science,
fine-grained classification, evaluation, benchmark
1 Introduction
Content-based image retrieval and computer vision approaches are considered
as one of the most promising solutions to help bridging the taxonomic gap, as
discussed in [5,1,36,34,17]. We therefore see an increasing interest in this trans-
disciplinary challenge in the multimedia community (e.g. in [26,10,2,25,20,12].
Beyond the raw identification performances achievable by state-of-the-art com-
puter vision algorithms, recent visual search paradigms actually offer much more
efficient and interactive ways of browsing large flora than standard field guides or
online web catalogs ([3]). Smartphone applications relying on such image-based
identification services are particularly promising for setting-up massive ecologi-
cal monitoring systems, involving thousands of contributors at a very low cost.
A first step in this way has been achieved by the US consortium behind LeafS-
nap3 , an i-phone application allowing the identification of 184 common american
3
http://leafsnap.com/
plant species based on pictures of cut leaves on an uniform background (see [23]
for more details). Then, the French consortium supporting Pl@ntNet ([17]) went
one step beyond by building an interactive image-based plant identification ap-
plication that is continuously enriched by the members of a social network spe-
cialized in botany. Inspired by the principles of citizen sciences and participatory
sensing, this project quickly met a large public with more than 300K downloads
of the mobile applications ([8,7]). A related initiative is the plant identification
evaluation task organized since 2011 in the context of the international evalu-
ation forum CLEF4 and that is based on the data collected within Pl@ntNet.
This paper presents the participation of Inria ZENITH team to the 2015-edition
of this challenge [9,19].
2 Related work
From a computer vision and technological perspective, our work is more gener-
ally related to image classification. Most popular methods for this problem are
typically based on the pooling of local visual features into global image repre-
sentations and the use of powerful classifiers in the resulting high-dimensional
embedded space such as linear support vector machines ([24,28]). The Bag-of-
word representation (BoW) notably remains a key concept although the raw
initial scheme of ([33]) is now outperformed by several alternative new schemes
([24,16,27,6,14]). Its principle is to first train a so called visual vocabulary thanks
to an unsupervised clustering algorithm computed on a given training set of lo-
cal features. The produced partition is then used to quantize the visual features
of a given new image into visual words that are aggregated within a single
high-dimensional histogram. Partial geometry can be embedded in the image
representation by using the Spatial Pyramid Matching scheme of ([24]). As it
relies on vector quantization, the BoW representation is however affected by
quantization errors. Very similar visual features might be split across distinct
clusters whereas more dissimilar ones might be affected to the same visual word.
This results in both mismatches and potentially irrelevant matches. To alleviate
this problem, several improvements have been proposed in the literature. The
first one consists in expanding the assignment of a given local feature to its near-
est visual words ([16,29,6,14]). This allows reducing the number of mismatches
without degrading much the encoding time. Other researchers have investigated
alternative ways to avoid the vector quantization step, using sparse coding ([38])
or locality-constrained linear coding ([37]). Such methods optimize the affecta-
tion of a given local feature to a small number of visual words thanks to sparsity
or locality constraints on the global representation. Another alternative is to
use aggregation-based models such as the improved Fisher Vector of [27] or the
VLAD encoding scheme ([14]). Such methods do not only encode the number
of occurrences of each visual word but also encode additional information about
4
http://www.clef-initiative.eu/
the distribution of the descriptors by aggregating the component-wise differ-
ences. When used with discriminative linear classifiers, such high-dimensional
representations benefit of both generative and discrimination approaches lead-
ing to state-of-the-art classification performances on fine-grained classification
benchmarks ([11]).
A radically different approach to image classification is the use of deep con-
volutional neural networks. Rather than extracting the features according to
hand-tuned or psycho-vision oriented filters, such methods directly work on the
image signal. The weights learned by the first convolutional layers allows to au-
tomatically build relevant image filters whereas the intermediate layers are in
charge of pooling these raw responses into high-level visual patterns. The last
fully connected layers work more traditionally as any discriminative classifier
on the image representation resulting from the previous layers. Deep convolu-
tional neural networks have been recently proved to achieve better results on
large-scale image classification datasets such as ImageNet ([22]) and do attract
more and more interest in the computer and multimedia vision communities. A
known drawback of Deep Convolutional Neural Networks is however that they
require a lot of training data mainly because of the huge number of parameters
to be learned. Their performances on fine-grained classification are consequently
more controversial and they are still often outperformed by local features based
approaches, as shown in our experiments. Besides, it is important to notice that
they inspire the investigation of new deep learning models making use of more
traditional visual features embedding methods (e.g. [31]).
3 Experimented fine-grained image classification systems
We did experiment two families of image classification techniques that are known
to provide state-of-the-art classification performances, in particular in fine-grained
recognition challenges ([11,18]).
3.1 Convolutional neural networks
Convolutional Neural Networks (CNN) have been mainly used since the 90’s for
their performances in digit classification. But since a few years, they appear to
have now surpassed all state of the art methods for large-scale image classifi-
cation [22]. In this experimentation, we have used Caffe [15], a Deep Learning
Framework, allowing us to use CNN architectures and models from the literature.
We have chosen in the Caffe model Zoo the ”GoogLeNet GPU implementation”
model, based on Google winning architecture in the ImageNet 2014 Challenge
[35], and we fine-tuned this model on the LifeCLEF datasets.
The GoogLeNet architecture consists of a 22 layers deep network with a
softmax loss as the top classifier. It is composed of three ”inception modules”
stacked on top of each other. Each intermediate inception module is connected
to an auxiliary classifier during training, so as to encourage discrimination in the
lower stages of the classifier, increase the gradient signal that gets propagated
back, and provide additional regularization. These auxiliary classifiers are only
used during the training part, and then discarded.
Experiments Setup The previously described GoogLeNet CNN uses square
images as input. For each image in the training and test sets, we therefore
cropped the largest square in the center, and re-sized it to 256x256 pixels. In-
stead of starting to train our CNN from scratch only on plant images, and as
it was authorized in this year’s challenge, we started with a CNN trained on
the popular generalist ImageNet dataset. We only removed its top layers (the
fully connected ones), changed the number of outputs, and trained this new
model using the desired dataset. As it was implemented within Caffe library, it
makes also use of a simple data augmentation technique, consisting in cropping
randomly a 224x224 pixels image, and eventually mirroring it horizontally.
During our preliminary experiments, we have tried several training strategies
that are presented are presented in Table 1.
Table 1. Various approaches using CNNs.
Name # CNNs Data Augmentation
CNN1 1 CNN with all images No
CNN2 1 CNN with all images Yes
CNN3 7 CNNs (1 for each view) No
We have tested all these configurations using the PlantCLEF 2014 data and
groundtruth (500 species, 47815 train images and 13146 test images). CNN1
configuration was the simplest and the first that we have tested, but finally also
the one providing the best results. The Data Augmentation method proposed
for CNN2 configuration increased significantly the number of train images as we
generated 8 new images by applying rotations, and a set of colorimetric transfor-
mations with randomized parameters, i.e. brightness & saturation modulation
in the HSL color space (multiplier factor randomized between 0.8 and 1.2), and
contrast modulation (multiplier factor randomized between 0.7 and 1.3). Even
with additional iterations to train the CNN, results remained nearly the same
than those for CNN1. The CNN3 configuration consisted in training several
CNNs, one for each view type (thanks to the tags provided in the meta-data).
On one hand, as some species haven’t images for all views, the number of output
for each CNN is lower than 1000 and that could help to obtain better results
because of the reduction of the confusion risk. On the other hand, some images
from a given view (Branch for example) can really help to identify some images
tagged with another view (Entire for example). Results were slightly lower for
the Branch, Entire, Leaf, Fruit, and Flower views than what was obtained with
the standalone CNN. This could be explained by a less important number of
images to train the network, and proves that images from a given view can help
when identifying an image tagged with another view. This conclusion is not true
for the Stem and LeafScan views. The reason is probably that the LeafScan view
is specific, very different from other views, and does not contain background in-
formation, and as the Stem tag identifies a closeup view of the plant which is
not really apparent on other images.
Training parameters As a reminder, here are the most important parameters
for Caffe to obtain our submitted run (CNN1). The base learning rate parameter
was set to 10−5 . The learning rate is divided by 10 every 60k iterations. After
150k iterations the training is over, and the batch size was fixed to 32. All other
parameters were unchanged.
3.2 Fisher vectors & Logistic Regression
Fisher vectors (FV) were first introduced in image classification by [27] and
proved to be very efficient in fine-grained classification tasks later on ([11]).
According to recent surveys such as [13], it is the best performing pooling strat-
egy currently available. We will only recall here the main steps used to extract
Fisher vectors, for detailed explanations of the theoretical derivation and for
performance analysis we redirect the readers to [30]. The pipeline for computing
the Fisher vector describing an image consists in:
1. Dense extraction of local features: descriptors, often SIFT descriptors, are
extracted on densely sampled overlapping patches at several scales.
2. PCA transformation: the descriptors are then de-correlated and compressed
using a Principal Component Analysis.
3. Feature space density estimation: the distribution of features is modeled
as a Gaussian Mixture Model (GMM) that is learned using the popular
Expectation-Maximisation (EM) algorithm. We thus obtain a probability
PK
distribution of the form of u(x) = k=1 wk uk (x) where uk follows a Gaus-
sian distribution of mean µk and covariance matrix Σk , uk ∼ N (µk , Σk ),
with Σk being diagonal because the features are decorrelated,
P and where wk
is the weight of the k-th Gaussian, these weights satisfy k wk = 1.
4. Encoding and pooling: the features are encoded and pooled using
N
1 X xi − µk
Gµk = √ γk (xi )
wk i=1 σk
N
1 X γk (xi ) xi − µk 2
Gσk = √ √ −1
wk i=1 2 σk
where all the divisions and squaring are element-wise operations and where
γk (x) = PK wk w
uk (x)
0 u 0 (x)
. Theses 2K vectors are concatenated to produce the
k0 =1 k k
final representation of dimension 2dK.
5. Post-processing: the vectors
p are L2-normalized and element-wise square-
rooted using x 7→ sign(x). |x|.
Usually, the classification of Fisher Vectors is performed using a linear clas-
sifier as it has been shown that using kernelization techniques on such high-
dimensional spaces does not improve significantly the performances. In our ex-
periments, we used the Logistic Regression classifier implemented within the
LibLinear library ([4]). This method was preferred over Support Vectors Ma-
chine because it directly outputs probabilities which then can be used for fusion
purposes.
Here, we used two types of Fisher Vectors with two different types of de-
scriptors. The first system was built with RootSIFT descriptors, l2-normalized
and square-rooted SIFT descriptors, of 128 dimensions which are then reduced
to 80 dimensions through PCA. The second one was based on some comple-
mentary descriptors used in the Pl@ntNet application [17]. It consists in the
concatenation of several basic descriptors such as Fourier histograms, Edge Ori-
entation Histograms (EOH), HSV histograms and Hough transform histograms.
This concatenation was then compressed and de-correlated using PCA. The as-
sociation of descriptors used depends on the organ, for Branch, Entire, Leaf,
LeafScan, Stem only Fourier, EOH and Hough histograms are used resulting
in 44-dimension final descriptors compressed to 14 dimensions after PCA while
Flower and Fruit add HSV histograms giving descriptors of dimension 74 re-
duced to 38 after compression. In both systems, the GMM used to estimate the
probability distribution of the features learns a codebook of 128 words.
4 Fusion methods
Combining multiple classifiers or even multiple results (i.e. several images of a
single observation) from a single classifier is a way to increase the classification
quality. This section presents three main approaches we used to merge the various
results from our classifiers.
4.1 Max and Borda
Maximum and Borda Count are two approaches used to merge top-k lists. While
the maximum relies on the score of each class with the lists, Borda Count uses
their rank.
More precisely, the maximum based approach associates to each class the
maximum score it reaches among the different lists. In the Borda Count ap-
proach, we have associated each class within a list to a score decreasing while
the rank increases. In more details, since we only retrieve the top-K most likely
classes, the score of a given species s is computed as follows:
X
score(s) = K − rc (s) (1)
c∈C
where rc (s) is the ranking of species s returned by the classifier c.
4.2 Bayesian inference
Framework presentation This fusion method is inspired by what is done
in crowdsourcing multi-labeled classification tasks [21,32]. For this purpose we
used the Bayesian inference framework described in Figure 1.
!(k) #
"(k) ti
ci(k)
k = 1, …, K
i = 1, …, N
Fig. 1. A Bayesian network to merge multiple classifiers identifications.
In such inference framework, we are given a set of classifiers k ∈ 1, ..., K and
a confusion matrix π (k) is assigned to each one of them. Such matrix enables to
(k)
evaluate the classification quality of each classifier. In a more precise way, πi,j
refers to the probability that the classifier k, given an image, will answer class
j while the right class is i. The set of all confusion matrices is noted Π. Notice
that, as presented in Figure 1, the confusion matrix π (k) is directly derived from
the parameters matrix α(k) . The set of all parameters matrices is noted A. In
parallel, each observation (i.e. set of images corresponding to a single plant) is
associated to a distribution probability, noted ti for the ith observation. This
probability depends on the proportion of each species in the database, and we
note κ the vector referring to this proportion. Finally, based on the probabilities
ti and on the confusion matrix of a given classifier k, we can infer the probability
(k)
of the classifier’s answer for the ith observation, noted ci .
Therefore, the joint probability of this Bayesian framework follows Equa-
tion 2.
N K
(k)
Y Y
p(Π, t, c|A, κ) = {κti π (k) }p(Π|A) (2)
ti ,ci
i=1 k=1
(k)
Once the classifiers answers (i.e. the set of answers ci for all k and i) are
known, the probabilities of A, Π, κ and t can be updated, thus inferring the
correct class of each observation (i.e. the one with the highest probability in
ti ). In the following, we suppose κ known thanks to the very large size of the
training set.
Addressing the large dimensionality Generally, in the state of the art so-
lutions, several approaches are proposed to compute the posterior probabilities
such as Gibbs sampling [21] or Variational Bayes [32]. In our experiments we had
to face the very large dimension of the problem: each confusion matrix being of
size 1000 × 1000. Classical method are therefore intractable in our context. To
address this challenge, we used a single-shot approach: only p(ti = j|rest) is
computed and used to update A and π – recall that κ is known and does not
need to be updated. Thus, the confusion matrix of each classifier evolves while
the number of identifications increases and the quality of inference is refined
more and more.
Experiments Setup In this subsection, we present three aspects of the setup:
parameters initialization, parameters refinement and classifier’s confusion refine-
ment.
An important part of the fusion is to learn the confusion matrix (and its
parameters). To do so, we have initialized each parameters matrix A with a
value of S in the diagonal and S/(dimension − 1) in the other cells, meaning
that there is a 50% probability that the classifier will be correct and that given
the correct class and a wrong one, it is more likely that the classifier will return
the correct one. In our experiments the value of S has been fixed to 5 (best
choice among several runs).
Then, we tried to enhance the confusion matrix quality based on the training
data. For each image of the set, we asked the classifiers to re-propose a top-30
classification, and, given the correct class i, we have added in each cell ai,j of
the matrices A a value inversely proportional to the species rank in the top-30:
1
rank .
Finally, to be as fine-grained as possible, each classifier was associated to
several confusion matrices corresponding to each plants organs. Thus, the system
knows the confusion of each classifier for all possible organs. In a way, we consider
each couple {organ, classif ier} as a single classifier.
5 Official Results
5.1 Runs details
3 runs were finally submitted to the LifeCLEF 2015 plant challenge:
– INRIA Zenith Run 1 is based on the results provided by the single Con-
volutionnal Neural Network finetuned using all provided data (CNN1), and
described in 3.1. Observations composed of several images, are combined
using a Max function to provide Observation Results.
– INRIA Zenith Run 2 is based on Fisher Vectors described in 3.2. To obtain
Observation Results we used the Borda Count Algorithm.
– INRIA Zenith Run 3 is the combination of the results obtained by previous
methods (CNN and Fisher Vectors) using the Bayesian inference method
described in 4.2.
5.2 Results
Table 2 summarizes the scores of the 11 best submitted runs out of a total of 18
runs. Figure 2 gives a complementary graphical overview of all results obtained
by the participants.
Table 2. PlantCLEF 2015 scores of the 11 best runs.
Name Score
SNUMED INFO run4 0.667
SNUMED INFO run3 0.663
QUT RV run2 0.633
QUT RV run3 0.624
SNUMED INFO run2 0.611
INRIA ZENITH run1 0.609
SNUMED INFO run1 0.604
INRIA ZENITH run3 0.592
QUT RV run1 0.563
ECOUAN run1 0.487
INRIA ZENITH run2 0.300
If we compare the best runs of each team, the INRIA Zenith Run 1, the one
using CNN, is ranked 3rd regarding to observation results. We can note that all
the 4 best teams used Deep Neural Networks. Our second run, INRIA Zenith
Run 2, the one using Fisher Vectors, is disappointingly distanced by the CNN
runs: its final score is two times lower (0.3 instead of 0.609 for INRIA Zenith Run
1 ). In LifeCLEF 2014, the best performances were obtained by Fisher Vectors,
but the use of external training data was not allowed which explains why CNN
were not performing better.
Our final run, INRIA Zenith Run 3, is the Bayesian inference fusion method
using previous runs. It was made in order to benefit from both technologies.
Unfortunately, the results obtained are a little bit lower than the standalone
CNN of INRIA Zenith Run 1 (0.592 instead of 0.609). Two main reasons can
be highlighted to explain this quality loss. First, the two classifiers are not nec-
essarily independent, thus, there combination does not enable to obtain quality
gain. Second, building a confusion matrix for such high dimension problems (i.e.
1000 × 1000) is very challenging and the size of the test set is not enough to
learn an accurate confusion.
6 Conclusion
Inria Zenith team submitted 3 runs, using different strategies. The first run was
based on the well-known GoogLeNet CNN architecture, finetuned over Imagenet
dataset, and using a max method to fuse image results to observation results.
Our second run did not used external data, and was based on fisher vectors which
Fig. 2. Official results
was last year winning technology. The conclusion is that Deep Neural Networks
outperforms fisher vectors for such classification tasks, particularly with an im-
portant number of classes, and when you have large training datasets. Our last
run consisted in trying a new fusion method, based on Bayesian inference, to
merge results of the two previous runs. However results were not as good as
expected, probably because the first run is already two times better than the
second one.
7 Appendix: Complementary Results
Table 3. Results for individual images
Methods Score
FV (color + basic texture) 0.184
FV (SIFT) 0.267
CNN 0.581
References
1. Cai, J., Ee, D., Pham, B., Roe, P., Zhang, J.: Sensor network for the monitoring of
ecosystem: Bird species recognition. In: Intelligent Sensors, Sensor Networks and
Information, 2007. ISSNIP 2007. 3rd International Conference on. pp. 293–298
(Dec 2007)
Table 4. Fusion results for observations
Descriptors Maximum Borda Bayesian
FV (color + basic texture) 0.196 0.197 0.186
FV (SIFT) 0.286 0.283 0.259
FV (color + basic texture + SIFT) 0.285 0.300 0.293
CNN 0.609 0.607 0.598
Global fusion 0.599 0.487 0.592
Fisher Vectors
CNN
0.0 0.2 0.4 0.6 0.8 1.0
Fig. 3. Distribution of the top 1 probabilities returned by the CNN and the Fisher
Vectors with Logistic Regression.
2. Cerutti, G., Tougne, L., Vacavant, A., Coquin, D.: A Parametric Active Polygon
for Leaf Segmentation and Shape Estimation. In: 7th International Symposium
on Visual Computing. p. 1. Las Vegas, United States (Sep 2011), https://hal.
archives-ouvertes.fr/hal-00622269
3. Ellison, A.M., Farnsworth, E.J., Chu, M., Kress, W.J., Neill, A.K., Best, J.H.,
Pickering, J., Stevenson, R.D., Courtney, G.W., VanDyk, J.K.: Next-generation
field guides (2013)
4. Fan, R.E., Chang, K.W., Hsieh, C.J., Wang, X.R., Lin, C.J.: Liblinear: A library for
large linear classification. The Journal of Machine Learning Research 9, 1871–1874
(2008)
5. Gaston, K.J., O’Neill, M.A.: Automated species identification: why not? Philosoph-
ical Transactions of the Royal Society of London B: Biological Sciences 359(1444),
655–667 (2004)
6. van Gemert, J.C., Veenman, C.J., Smeulders, A.W., Geusebroek, J.M.: Visual
word ambiguity. Pattern Analysis and Machine Intelligence, IEEE Transactions on
32(7), 1271–1283 (2010)
7. Goëau, H., Bonnet, P., Joly, A., Affouard, A., Bakic, V., Barbe, J., Dufour, S.,
Selmi, S., Yahiaoui, I., Vignau, C., et al.: Pl@ ntnet mobile 2014: Android port and
new features. In: Proceedings of International Conference on Multimedia Retrieval.
p. 527. ACM (2014)
8. Goëau, H., Bonnet, P., Joly, A., Bakic, V., Barbe, J., Yahiaoui, I., Selmi, S., Carré,
J., Barthélémy, D., Boujemaa, N., et al.: Plantnet mobile app. In: Proceedings of
the 21st ACM international conference on Multimedia. pp. 423–424. ACM (2013)
9. Goëau, H., Joly, A., Bonnet, P.: Lifeclef plant identification task 2015. In: CLEF
working notes 2015 (2015)
10. Goëau, H., Joly, A., Selmi, S., Bonnet, P., Mouysset, E., Joyeux, L.: Visual-based
plant species identification from crowdsourced data. In: MM’11 - ACM Multimedia
2011. pp. 0–0. ACM, Scottsdale, United States (Nov 2011), https://hal.inria.
fr/hal-00642236
11. Gosselin, P.H., Murray, N., Jégou, H., Perronnin, F.: Revisiting the fisher vector
for fine-grained classification. Pattern Recognition Letters 49, 92–98 (2014)
12. Hsu, T.H., Lee, C.H., Chen, L.H.: An interactive flower image recognition system.
Multimedia Tools Appl. 53(1), 53–73 (May 2011), http://dx.doi.org/10.1007/
s11042-010-0490-6
13. Huang, Y., Wu, Z., Wang, L., Tan, T.: Feature coding in image classification: A
comprehensive study. Pattern Analysis and Machine Intelligence, IEEE Transac-
tions on 36(3), 493–506 (2014)
14. Jégou, H., Perronnin, F., Douze, M., Sánchez, J., Pérez, P., Schmid, C.: Aggre-
gating local image descriptors into compact codes. Pattern Analysis and Machine
Intelligence, IEEE Transactions on 34(9), 1704–1716 (2012)
15. Jia, Y., Shelhamer, E., Donahue, J., Karayev, S., Long, J., Girshick, R., Guadar-
rama, S., Darrell, T.: Caffe: Convolutional architecture for fast feature embedding.
arXiv preprint arXiv:1408.5093 (2014)
16. Jiang, Y.G., Ngo, C.W., Yang, J.: Towards optimal bag-of-features for object cat-
egorization and semantic video retrieval. In: Proceedings of the 6th ACM interna-
tional conference on Image and video retrieval. pp. 494–501. ACM (2007)
17. Joly, A., Goëau, H., Bonnet, P., Bakić, V., Barbe, J., Selmi, S., Yahiaoui, I., Carré,
J., Mouysset, E., Molino, J.F., et al.: Interactive plant identification based on social
image data. Ecological Informatics 23, 22–34 (2014)
18. Joly, A., Goëau, H., Glotin, H., Spampinato, C., Bonnet, P., Vellinga, W.P.,
Planque, R., Rauber, A., Fisher, R., Müller, H.: Lifeclef 2014: multimedia life
species identification challenges. In: Information Access Evaluation. Multilingual-
ity, Multimodality, and Interaction, pp. 229–249. Springer (2014)
19. Joly, A., Müller, H., Goëau, H., Glotin, H., Spampinato, C., Rauber, A., Bonnet,
P., Vellinga, W.P., Fisher, B.: Lifeclef 2015: multimedia life species identification
challenges
20. Kebapci, H., Yanikoglu, B., Unal, G.: Plant image retrieval using color, shape and
texture features. Comput. J. 54(9), 1475–1490 (Sep 2011), http://dx.doi.org/
10.1093/comjnl/bxq037
21. Kim, H.C., Ghahramani, Z.: Bayesian classifier combination. In: International con-
ference on artificial intelligence and statistics. pp. 619–627 (2012)
22. Krizhevsky, A., Sutskever, I., Hinton, G.E.: Imagenet classification with deep con-
volutional neural networks. In: Advances in neural information processing systems.
pp. 1097–1105 (2012)
23. Kumar, N., Belhumeur, P.N., Biswas, A., Jacobs, D.W., Kress, W.J., Lopez, I.C.,
Soares, J.V.: Leafsnap: A computer vision system for automatic plant species iden-
tification. In: Computer Vision–ECCV 2012, pp. 502–516. Springer (2012)
24. Lazebnik, S., Schmid, C., Ponce, J.: Beyond bags of features: Spatial pyramid
matching for recognizing natural scene categories. In: Computer Vision and Pattern
Recognition, 2006 IEEE Computer Society Conference on. vol. 2, pp. 2169–2178.
IEEE (2006)
25. Mouine, S., Yahiaoui, I., Verroust-Blondet, A.: Advanced shape context for plant
species identification using leaf image retrieval. In: Ip, H.H.S., Rui, Y. (eds.) ICMR
’12 - 2nd ACM International Conference on Multimedia Retrieval. ACM, Hong
Kong, China (Jun 2012), https://hal.inria.fr/hal-00726785
26. Nilsback, M.E., Zisserman, A.: Automated flower classification over a large number
of classes. In: Computer Vision, Graphics Image Processing, 2008. ICVGIP ’08.
Sixth Indian Conference on. pp. 722–729 (Dec 2008)
27. Perronnin, F., Dance, C.: Fisher kernels on visual vocabularies for image cate-
gorization. In: Computer Vision and Pattern Recognition, 2007. CVPR’07. IEEE
Conference on. pp. 1–8. IEEE (2007)
28. Perronnin, F., Sánchez, J., Mensink, T.: Improving the fisher kernel for large-
scale image classification. In: Computer Vision–ECCV 2010, pp. 143–156. Springer
(2010)
29. Philbin, J., Chum, O., Isard, M., Sivic, J., Zisserman, A.: Lost in quantization:
Improving particular object retrieval in large scale image databases. In: Computer
Vision and Pattern Recognition, 2008. CVPR 2008. IEEE Conference on. pp. 1–8.
IEEE (2008)
30. Sánchez, J., Perronnin, F., Mensink, T., Verbeek, J.: Image classification with the
fisher vector: Theory and practice. International journal of computer vision 105(3),
222–245 (2013)
31. Simonyan, K., Vedaldi, A., Zisserman, A.: Deep Fisher networks for large-scale
image classification. In: Advances in Neural Information Processing Systems (2013)
32. Simpson, E., Roberts, S., Psorakis, I., Smith, A.: Dynamic Bayesian Combina-
tion of Multiple Imperfect Classiers. In: Decision Making with Imperfect Decision
Makers Springer (2012)
33. Sivic, J., Zisserman, A.: Video google: A text retrieval approach to object match-
ing in videos. In: Computer Vision, 2003. Proceedings. Ninth IEEE International
Conference on. pp. 1470–1477. IEEE (2003)
34. Spampinato, C., Mezaris, V., van Ossenbruggen, J.: Multimedia analysis for ecolog-
ical data. In: Proceedings of the 20th ACM international conference on Multimedia.
pp. 1507–1508. ACM (2012)
35. Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Van-
houcke, V., Rabinovich, A.: Going deeper with convolutions. CoRR abs/1409.4842
(2014), http://arxiv.org/abs/1409.4842
36. Trifa, V.M., Kirschel, A.N.G., Taylor, C.E., Vallejo, E.E.: Automated species recog-
nition of antbirds in a Mexican rainforest using hidden Markov models. Journal of
The Acoustical Society of America 123 (2008)
37. Wang, J., Yang, J., Yu, K., Lv, F., Huang, T., Gong, Y.: Locality-constrained
linear coding for image classification. In: Computer Vision and Pattern Recognition
(CVPR), 2010 IEEE Conference on. pp. 3360–3367. IEEE (2010)
38. Yang, J., Yu, K., Gong, Y., Huang, T.: Linear spatial pyramid matching using
sparse coding for image classification. In: Computer Vision and Pattern Recogni-
tion, 2009. CVPR 2009. IEEE Conference on. pp. 1794–1801. IEEE (2009)