=Paper=
{{Paper
|id=Vol-3290/long_paper6539
|storemode=property
|title=The Computational Memorability of Iconic Images
|pdfUrl=https://ceur-ws.org/Vol-3290/long_paper6539.pdf
|volume=Vol-3290
|authors=Lisa Saleh,Nanne van Noord
|dblpUrl=https://dblp.org/rec/conf/chr/SalehN22
}}
==The Computational Memorability of Iconic Images==
https://ceur-ws.org/Vol-3290/long_paper6539.pdf
The Computational Memorability of Iconic Images
Lisa Saleh, Nanne van Noord∗
Multimedia Analytics Lab, University of Amsterdam
Abstract
The perception of historic events is frequently shaped by speci昀椀c images that have been ascribed an
iconic status. These images are widely reproduced and recognised and can therefore be considered
memorable. A question that arises given such images is whether the memorability of iconic images is
intrinsic or whether it is shaped. In this work we analyse the memorability of iconic images by means
of computational techniques that are speci昀椀cally designed to measure the intrinsic memorability of
images. To judge whether iconic images are inherently more memorable we establish two baselines
based on datasets of diverse imagery and of newspaper imagery. Our 昀椀ndings show that iconic images
are not more memorable than modern day newspaper imagery or when compared to a diverse set of
everyday images. In fact, by and large many of the iconic images analysed score on the low end of
the memorability spectrum. Additionally, we explore the variation in memorability of reproductions
of iconic images and 昀椀nd that certain images have been edited resulting in higher memorability scores,
but that the images by and large are reproduced with memorability close to the original.
Keywords
Memorability, Iconicity, Computer Vision
1. Introduction
The need to capture historic events in a visual frame has been around long enough to have
captured events from centuries ago [18]. From ancient greek wall paintings of the battle of
Marathon (490 BC) to the Black Death (14th century) in elaborate oil paintings [24, 6]. His-
toric events are o昀琀en remembered through the visual imagery it is captured in [18]. Due to
technological development in recent history, it has become conventional to capture events
with photography. A unique share of historic photographs displayed in media are considered
iconic. To illustrate, Tank Man, the image of a man in front of the tanks in a street of Tienanmen
Square in Beijing, is an image which many people immediately can recall. Iconic photographs
can be de昀椀ned as “photographs that are widely recalled and recognized by individuals across
social groups and generations, that they connect with speci昀椀c historical events, and that have
emotional signi昀椀cance for them and their national community” [7]. Iconic photographs are
thus inherently connected with historical events and may in昀氀uence the views of people on this
historic event. For example, the photograph of Alan Kurdi during the Syrian refugee crisis in
2015 caused such an emotional response with most individuals that would in昀氀uence the interna-
tional politics surrounding the Syrian refugee crisis [1]. Thus iconic images may even in昀氀uence
CHR 2022: Computational Humanities Research Conference, December 12 – 14, 2022, Antwerp, Belgium
∗
Corresponding author.
£ lisa.saleh@student.uva.nl (L. Saleh); n.j.e.vannoord@uva.nl (N. van Noord)
ȉ 0000-0002-5145-3603 (N. van Noord)
© 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
Workshop
Proceedings
http://ceur-ws.org
ISSN 1613-0073
CEUR Workshop Proceedings (CEUR-WS.org)
55
the course of history. Due to this in昀氀uence iconic images have been widely researched with
the aim to understand what makes them unique.
Di昀昀erent aspects of iconic images have been researched. One o昀琀en noted criteria for iconic-
ity is the importance of the captured historic event. Due to the democratisation of photography,
images of recent historic events are abundant. Thus an image that merely displays a signi昀椀cant
historic event is not su昀케cient to reach iconic status. For an image to become iconic it must be
widely spread in the media [11]. Therefore it is useful to know what images the media tends
to publish and which characteristics they carry. For example, one study on media portrayal
of photographs of hurricane Katrina uncovered that photographs with some visual themes are
more o昀琀en published than others [8]. Besides visual themes, there are plenty of other aspects
that can in昀氀uence if an image can be iconic. To recall the de昀椀nition of an iconic photograph,
the photograph must be widely recalled and recognized. An image that is widely recalled and
recognized needs either to be really memorable or to be displayed very o昀琀en. A more memo-
rable image is more likely to be recalled. Thus high memorability of an image may contribute
to its iconic status. As there has been no research on the intrinsic memorability of iconic im-
ages, this will be the focus of this paper. Speci昀椀cally, we investigate whether there are general
patterns concerning memorability of iconic images that are salient even when considering the
unique circumstances by which these images have become iconic.
Memorability of images is a complex characteristic to predict; Using only the surface features
of the image for predicting memorability is not comprehensive. Automating the prediction
of memorability with Convolutional Neural Networks (CNN) has been a successful solution
to this problem [16]. It is proven that features such as object size and brightness do a昀昀ect
image memorability [10]. However, a CNN like MemNet outperforms these surface features
in predicting memorability [16]. When considering image memorability what a method like
MemNet measures is the intrinsic memorability of an image, which can be interpreted as the
innate probability of an image to be remembered. It does not guarantee that an image will be
remembered, and it is certainly not the only factor, but all other factors being equal an image
with a higher intrinsic memorability is more likely to be remembered.
Besides the advantages of automatizing the process by using a CNN, the usage of a predic-
tion CNN has another advantage: As iconic photographs are images that are widely spread it
is likely that a great share of people has already seen said image. Therefore it would be di昀케-
cult to test the memorability of these images with a memory task. It would be a challenge to
guarantee that these people have never seen these images before. Thus the challenge of test-
ing the memorability of iconic photographs context-independent can be solved by using the
above-mentioned MemNet. Additionally, the iconicity of some images is determined by the
fame of the subject. MemNet is not trained to recognize celebrities or other symbols. In this
research, the focus is on what image aspects the memorability in昀氀uenced, without analysis of
the context of the images. This makes the usage of MemNet a great 昀椀t. In this paper, iconic
photographs are tested on their memorability by the use of the CNN MemNet to explore to
what extent the memorability of iconic photographs can be measured.
To investigate this we formulate the following two research questions:
RQ 1: How does the memorability of iconic images compare to a baseline?
RQ 2: To what extent does the memorability of iconic images di昀昀er across variations?
56
2. Related work
2.1. Iconic Images
There has been little to no computational research on iconic images, but in the following we
will give a brief overview of other research directions. As iconic images are not a bound set,
di昀昀erent de昀椀nitions and what criteria they should meet have been proposed [7, 17, 2, 11]. There
are di昀昀erences in the criteria brought up, but the following six proposed by Perlmutter [22] are
used most o昀琀en: ”(1) signi昀椀cance of the reported event; (2) capacity to represent the event
as a whole; (3) celebrity of the image promoted by the media; (4) prominence of display of
the image; (5) frequent repetition of the image across media outlets; (6) ability to generate a
primordial theme in society such as good versus evil.” These criteria are not met to the same
extent for every iconic image; it is not a blueprint for iconic images. For example, both the
image of the Falling Man of the Twin Towers and the hijacked plane crashing into the Twin
Towers represent the September 11 attacks. The image of the Falling Man has a less capacity
to represent the event as a whole than the image of the airplane, as the attack itself is displayed
more comprehensively in the latter image. But even though the Falling Man image does not
meet the second criteria to the full extent, it is still an image to be considered iconic.
Additionally, other studies have added criteria to be more comprehensive. For example, in
some studies more stress has been placed on the symbolic meaning that people have of the
image [7]. An image can become a symbol with a meaning beyond the historical image it is
attached to. The Guerrillero Heroico image of Che Guevara is used as a symbol for revolution
beyond in Cuba [20]. Even though this is not signi昀椀cant for every iconic image, it is still a
unique aspect for some. Furthermore, the criteria from Perlmutter [22] put the focus on the
image and less on the reception of the image. One aspect of the reception that is unmentioned
in these six criteria is that an iconic image also should be widely recalled and part of the col-
lective memory of a certain with a group’s identity [7]. This is another criterion o昀琀en added
to complement the criteria from Perlmutter. That an image must be widely recalled is impor-
tant to add because this can be a way of researching what image have an iconic status with
qualitative research as done by for example Hoeven [12].
But as other factors of the reception, like the symbolic meaning or being a part of the col-
lective memory, of an iconic image are not fully encoded in the image itself and are subjective,
it is yet challenging to study computational [19]. It is yet di昀케cult to study the semantics of
images computationally, as this o昀琀en is can not be conducted from the visual data or metadata.
Therefore the study of the perception of iconic images is o昀琀en limited to interviews or large-
scale surveys [19]. Hence the study of iconic images in computer vision is yet to be immersed,
and potentially the computational study of memorability can open this door.
2.2. Memorability
Due to technological advancements capturing and sharing have become more convenient.
Mass media and its consumption lead people to be exposed to countless images every day.
People are exposed to many images daily and some of those are better remembered than oth-
ers. Which image people remember and which they forget is related to the memorability of
the image. Di昀昀erent academic 昀椀elds have an interest in image memorability. For example,
57
research in cognitive sciences answered how certain activities in the brain correlate to the
memorability of images. Additionally, image memorability has been a subject of research in
the computer vision 昀椀eld in recent years [16, 10, 9, 14, 4]. Taking the next step from com-
puting how memorable images are with MemNet, to analyzing what image attributes leads to
memorable image and what connection memorability has with other qualities like the emotion
it portrays. In line with researching what attributes in昀氀uence memorability, GANalyze was
created [10]. With GANalyze images are tweaked to increase memorability. Some attributes
emerged to o昀琀en be increased when optimizing the memorability, such as redness, brightness
and object size [10]. Thus images that have higher redness, brightness or a bigger object size
are more likely to be memorable. Even though these attributes might be simple and intuitive
ways to predict memorability, the prediction of memorability is complex and can not solely be
explained by these attributes [16].
When diving further into the correlation between memorability and some image aspects,
collected across di昀昀erent studies, the following observations were made. In some studies, it
appeared that a number of object categories, like people, animals and vehicles were relatively
more memorable [21]. Images with humans are especially memorable when the face is visi-
ble and has eye contact with the camera [13]. In scenic images, indoor scenes appeared to be
relatively more memorable [13]. Additionally, it appeared that spatial layout is highly corre-
lated with memorability; Images where the object is bigger and centred in the image are more
likely to be memorable [21]. This is in line with research which showed that images with a
high aesthetic level are also more likely to be memorable [15]; the spatial layout of an image is
strongly connected to its aesthetics. Moreover, it also appeared that images in which humans
portray negative emotions like disgust and fear, tend to be recalled better [16]. Finally, only
the most memorable images have a correlation with popularity [16]. Even though mapping for
these above attributes and many more, there is still about 25% of the variance in memorability
that is unaccounted for [23]. Thus looking for the attributes in images that are correlated with
memorability will give some intuitive sense of memorability, but will not be comprehensive.
2.3. MemNet & LaMem
In this paper, memorability will be measured computationally through a score given by Mem-
Net. MemNet is a CNN trained on an annotated dataset called LaMem. LaMem is a large-scale
dataset constructed of multiple already existing datasets that were annotated in a perception
study with human subjects who performed a memorability task. The datasets used are highly
diverse, which leads to LaMem being a diverse dataset totalling 60.000 images [16]. It contains
images of humans, animals, landscapes, and even art, which also includes abstract images. This
diversity of image types is displayed in Figure 1.
For the LaMem annotation task a stream of images was shown to participants and a昀琀er a
varying number of distractors the participants had to indicate if they had seen the image before.
Thus when an image was remembered o昀琀en the image would be scored as more memorable.
A昀琀er performing this task on a group of participants a memorability score for each image
could be calculated. This score is the annotation for each image in the LaMem dataset. This
makes LaMem thereby a su昀케cient foundation for a CNN to compute a memorability score
(0 − 1 scalar) for di昀昀erent types of images [16]. LaMem was shown to have an inter-annotator
58
Figure 1: Sample images from LaMem [16] with diversity in image subject. The images are arranged
by their memorability score, decreasing from le昀琀 to right.
rank correlation of 0.68, with MemNet achieving a remarkably high rank correlation of 0.64
[16, p. 6]. MemNet has since been further validated in subsequent studies and is, therefore, a
reliable measure of intrinsic image memorability [23, 21, 14, 4].
3. Memorability of Iconic Images
The following step-by-step procedure, shown in Figure 2 represents our proposed method that
can be used for measuring a memorability across a set of images.
Figure 2: Flowchart overview of the memorability comparison procedure for two datasets.
The procedure we propose involves a comparative process between two datasets (A and B).
Here, dataset A represents the collection to be studied, whereas dataset B functions as a baseline.
The general idea of this approach is to not only measure the memorability scores but also
judge whether the scores di昀昀er from a meaningful baseline. Each dataset is pre-processed by
transforming the images to a suitable input resolution for the CNN (typically a low resolution
like 256 × 256) and normalizing them. Once the dataset are pre-processed the images will be
passed through the pre-trained CNN to obtain a per-image memorability score.
The method for comparing between the memorability scores of datasets A and B depends
on the sizes of the respective datasets. In the case of small datasets, the analysis will be mostly
done by qualitatively looking at the images to draw conclusions based on the speci昀椀c images.
When dealing with large-scale datasets, the comparison can be based on a statistical analysis.
Whilst our baseline datasets are large-scale, the dataset of iconic images used is too small scale
to reliably perform statistical analysis.
Further analysis beyond the comparative analysis will consist of sampling images which are
59
either random or statically interesting, according to their attribute score. Sampling random
images will give an indicative perspective on how the data is constructed. This method of
analysis is based on Distant Viewing [3], a method for studying large visual corpora, where the
main take way is to perform the computational analysis whilst also viewing the corpora. The
latter is import because the intuitive semantics that may get lost in a computational analysis
will thereby also be studied. As the semantics are important for iconic images, this method
suits this paper well. Conclusions drawn from these analyses can then be linked to prior work.
3.1. Comparing the Memorability of Iconic Images
For our dataset of iconic images we use a selection of 26 images that were initially in a large-
scale survey on iconic images [12]. Smits and Ros [25] further used this set and collected
variations of the images in the form of online circulations. These 26 images are a part of a
global visual memory: ’a limited set of images that people all over the world have seen and
remembered’ [12]. Each image and its scores are depicted in Figure 3. These images are in
order from highest to lowest memorability score. When comparing the scores with the images
we make the following observations. Firstly, it is notable is that the three images with the
highest memorability score are similar. The images of Che Guevara, Sharbat Gula, and the
Migrant Mother are all portraits where the face of the subject 昀椀lls much of the image. Their
expression is visible to the viewer as all three are looking in the (rough) direction of the camera.
That these portraits have a high memorability score is in line with the prior work; Images with
bigger object size, a face that has eye contact with the camera and humans as their main object
are more likely to have a higher memorability score. Which makes it in the lines of expectation
that images like the Holocaust survivors, Raising a 昀氀ag over the Reichstag and Tank Man have
a lower memorability score; These images have a small object size and either has no human
faces or a high amount of di昀昀erent faces. On the whole we see a diverse range of memorability
scores across the dataset.
3.2. Iconic Images compared to a Baseline
To place the memorability scores in context, we compare to two di昀昀erent baselines. The 昀椀rst
baseline used is the LaMem dataset; This dataset consists of images with di昀昀erent subjects
and themes, and functions as a diverse baseline representing a ‘general image collection’ [16].
This is an annotated dataset for memorability, but the annotations were not used; Images of
the LaMem dataset were inserted in MemNet to create a memorability score generated with
the same circumstances as the dataset it is compared with. The second baseline used was the
GoodNews dataset, which consists of all kinds of images used in the New York Times from 1818
until 2019 [5]. These images vary from news photography to sports photography, to images
from the cooking appendix and so on. The range of types of images is displayed in Figure 4.
These images are mainly images that were made by professional photographers and selected
by the editors of the paper. Thus these images o昀琀en have some journalistic value and meet
the aesthetic standards of the paper. Therefore the GoodNews dataset is used as a baseline to
represent images depicted in media. As iconic images are widely published in the media, the
GoodNews dataset will help us determine whether iconic images are particularly memorable
60
Sharbat Gula
Che Guevera Migrant Mother
0.9120
0.8894 0.8767
Survivor
of Hutu The Falling
Man on the Moon
death Man
0.8168
camp 0.8190
0.8290
Abu Hinden-
Ghraib burg Alan Kurdi
prisoner Disaster 0.7897
0.8098 0.7940
Mohandas
Mao Zedong Hijacked Airplane
Ghandi
0.7767 0.7697
0.7725
Viet Cong
Times Square Kiss Napalm Girl
Execution
0.7610 0.7588
0.7533
Burning Vultre and Situation
Monk the girl Room
0.7476 0.7462 0.7447
Raising the
Assassination of Kent State
flag on
Inejiro Anasuma Shootings
Iwo Jima
0.7248 0.7151
0.7173
Spanish Coup in
Tank Man
Soldier Chile
0.6939
0.6982 0.6981
Raising a
Flag over Holocaust
the survivors
Reichstag 0.6449
0.6796
Figure 3: Iconic images sorted on their memorability score. Each image is captioned with the name it
is known as and its memorability score. The photographer and year of origin can be found in Table 1.
61
when compared to other media images.
Figure 4: Samples from the GoodNews dataset, which contains diverse images from The New York
Times [5].
The distribution of the memorability scores of the LaMem and GoodNews dataset is visu-
alised in Figure 5. Additionally, the scores of the iconic images are plotted with a scatter plot
at the bottom of the Figure. From the distributions we can observe that the GoodNews images
are generally more memorable than the LaMem images. Additionally, the LaMem dataset has
a wider distribution than the GoodNews dataset. There are fewer images in the GoodNews
dataset with a very low memorability score, which might re昀氀ect that all images in GoodNews
have been approved by an editor of the paper. It is very likely an editor would select a pic-
ture with at least some traits that correlate with memorability. For example, images that are
selected by an editor o昀琀en need to have either a clear main object, thus big object size or some
aesthetic value. Taking into account scatter plot of the iconic images we observe that iconic
images are generally (slightly) less memorably than the images depicted in the media.
To further clarify how the iconic images compare to both datasets we plot the distribution
of the iconic images according to the quartiles of both LaMem and GoodNews in Figure 6. We
can observe that most iconic images fall into the lower quartiles for both datasets, scoring
slightly higher when compared to LaMem. As the LaMem dataset is representative of all kinds
of images, it can be concluded that these iconic images are slightly less memorable images.
When comparing the iconic images to the GoodNews dataset, it stands out that most of the
iconic images have a memorability score that overlaps with the 昀椀rst quantile of the GoodNews
dataset. The distribution of the iconic images in the quartiles of the GoodNews dataset is
signi昀椀cantly shi昀琀ed to the less memorable side. Thus iconic images are less memorable than
images depicted in the media.
4. Memorability across Variations
Until now we have analysed the most canonical versions of the 26 iconic images. However, it
is known that edits have been made in prominently published versions of iconic images. To
explore whether these edits have been made (implicitly or explicitly) to boost the memorability
of iconic images we further analyse the circulations collected by Smits and Ros [25]. For each
of the 26 iconic images they used the Google Cloud Vision API to retrieve online circulations,
thereby creating a dataset of 900k images. Per image the number of variations di昀昀ers from the
Che Guevara portrait having over 100k variations retrieved to the image of Mao Zedong and
62
Figure 5: The upper graph is the distribution plot of the LaMem and GoodNews dataset, where the
distribution is the density for the memorability scores. The graph below is a scatter plot of the iconic
images and how they compare the datasets.
Figure 6: Distribution of the iconic images across the LaMem and GoodNews datasets. The y-axis
represents how many iconic images have the memorability score for that quartile.
the founding of the PRC, having less than 3k variations. Figure 7 shows how many images
were retrieved for each of the images.
Due to how the variations were collected, in iterations based on previous retrieval results, the
63
Table 1
Reproduction of table from Smits and Ros [25] with the amount of circulations per image.
known as photographer year historical event circulations
Migrant mother Dorothea Lange 1936 Great Depression 41697
Falling Soldier Robert Capa 1936 Spanish Civil War 18194
The Hindenburg Disaster Sam Shere 1937 Zeppelin 36683
Times Square Kiss Alfred Eisenstaedt 1945 V-Day 65164
Raising the Flag on Iwo Jima Joe Rosenthal 1945 Pacific War 63249
Holocaust survivors Lee Miller 1945 Holocaust 18343
Raising a Flag over the Reichstag Yevgeny Khaldei 1945 World War II 90344
Gandhi and the Spinning Wheel Margaret Bourke-White 1946 Mohandas Gandhi 10893
The Founding of the PRC Hou Bo 1949 Mao Zedong 2865
Assassination of Inejiro Asanuma Yasushi Nagao 1960 post-war Japan 3921
Guerillero heroico Alberto Korda 1960 Che Guevara 108288
The Burning Monk Malcom Browne 1963 Vietnam War 18122
Saigon Execution Eddie Adams 1968 Vietnam War 18305
A Man on the Moon Neil Armstrong 1969 Space Race 186921
Kent State Shootings John Filo 1970 Kent State 7320
Accidental Napalm (Napalm girl) Nick Ut 1972 Vietnam War 38619
Allende’s Last Stand Luis Orlando 1973 South-American Coups 6997
Afghan Girl Steve McCurry 1984 Afghan War 47892
Tank Man Je昀昀 Widener 1989 Tiananmen Square Protest 63182
The vulture and the little girl Kevin Carter 1993 Sudan famine 30121
Survivor of Hutu death camp James Nachtwey 1994 Rwandan genocide 3395
The Falling Man Richard Drew 2001 9/11 11681
Hijacked airplane unknown 2001 9/11 6938
Abu Ghraib prisoner Sergeant Ivan Frederick 2003 Iraq War 3601
The Situation Room Pete Souza 2011 War on Terrorism 20102
Alan Kurdi Nilüfer Demir 2015 Refugee crisis 24432
total 947269
di昀昀erences to the original get progressively larger the deeper we get into the list of variations.
The variations of the iconic image vary in di昀昀erent crops, di昀昀erent shades of colors and so
on. There are also variations where other images and text are added to the iconic image, this
can di昀昀er from a logo of a broadcaster to a book cover where the original image also appears.
Furthermore, internet memes, collages and other types of photo-shopped images appear in the
dataset. Additionally, the API is not infallible, there are some images in the dataset where the
original image does not appear at all. Roughly speaking we observe that images which have
been circulated less there are also fewer edits. To control for some of this variation, and to limit
the scale of the comparison, we only use the 昀椀rst 10k variations for each iconic image.
The memorability scores for the variations of each iconic image are visualised in Figure 7.
Images to the right on the x-axis should generally have less resemblance with the original
image. When viewing the distributions of scores for the variations some patterns can be found.
We recognise four groups: (1) distributions with relatively little spread across the variations,
(2) distributions that fan outward towards the end of the plot, (3) distributions that have a large
spread from start to 昀椀nish, and (4) distributions where we can recognise clear clusters.
Examples of the 昀椀rst group, with little spread, at the Migrant Mother, the Situation Room, and
the assassination of Anasuma. Within this group we can recognise two subcategories, as they
64
Figure 7: Grid with a scatter plot per iconic image, where each scatter plot visualises the memorability
score across the variations in sequence. Higher x values are further from the original image.
are either images that have less than 10k variations or they are images where the canonical
form is most dominant. Because for the latter subcategory we have only looked at the 昀椀rst
10k images retrieved, it is still very likely that there are also many variations that di昀昀er more
strongly circulating on the internet. But these would only be retrieved further in the dataset
than the 昀椀rst 10k images selected. However, it is still noticeable that the 昀椀rst 10k variations
have more resemblance with the original iconic image than for other widely circulated iconic
images.
For the second group examples have most of their spread at the end of the graph, such as
the Burning Monk, the Falling man and the Tank Man images. In these images, it appears that
there is de昀椀nitely a big share of images that have a big resemblance with the original picture
but variations that di昀昀er more already appear within the 昀椀rst 10k variations.
In the third group there is a lot of spread from beginning to 昀椀nish, these are images like
Gandhi, Raising the 昀氀ag over Iwo Jima and the Image of the Spanish Soldier. These images
appear in a lot of di昀昀erent variations. This can be explained by it being popular photos for
web pages, book covers and other types of editing where the context of the image gets altered
heavily.
Lastly is the group where we can recognise di昀昀erent clusters. This group includes images
65
like Napalm Girl, the Hijacked Plane and the portrait of Sharbat Gula. All these images have an
alternative popular variation, which leads to a cluster forming in the 昀椀gure. In the following
section, the Napalm Girl image will be highlighted to give an example of what these variations
look like. Two variations of the Sharbat Gula image and the Hijacked plane are displayed
in Figure 8 with their memorability score in the caption of each image. The variations that
are highlighted in this 昀椀gure are average images from the main clusters. In the Sharbat Gula
variations graph, the bigger cluster on the top of the graph is the original image (Figure 8a) and
the cluster that lies under most of the images are depicted in Figure 8b. In the Hijacked Plane,
there is a similar pattern, the graph of the variations has two distinct clusters, one at the top
which consists of images like the original image (Figure 8c and the other cluster which has a
lower memorability score than the 昀椀rst cluster. This cluster consists of images like in Figure 8d,
which has a wider crop than the original image such that the building on the le昀琀 still remains
in the image.
(b) The variation of the original with a recent im-
(a) The original image with a memorability score age placed alongside has a memorability score
of 0.9120 of 0.8511
(c) The published image with a memorability (d) A variation of the published image with a
score of 0.7697 wider crop and a memorability score of 0.7095
Figure 8: Two variations of the Sharbat Gula and the Hijacked Plane image. Where the Figure 8a is
the original portrait of Sharbat Gula and Figure 8b is a variation where a more recent portrait is placed
alongside the original. The Figure 8c is the original published version of the Hijacked Plane and the
wider crop variation in Figure 8d
66
4.1. The Memorability of the Variations of the Napalm Girl Image
Figure 9: Scatter plot of the memorability scores on the y-axis and the sequence of variations of the
Napalm Girl image on the x-axis, where the red dotted line is a trend line.
A number of di昀昀erent crops of the Napalm girl image circulate online. The scatter plot which
displays the memorability scores and their variations is depicted in Figure 9. The outliers on the
upper side of this plot, images with a score above 0.85, are either images heavily photoshopped
or images that got in this dataset but in which the Napalm Girl image does not appear. One
of the aspects that stands out from this graph is the big cluster of images which are under the
trend line and appear from about 4.000 on the x-axis. When sampling these images it appears
that these are mainly the Napalm Girl image with a tighter crop than the original image. This
is displayed in Figure 10. Where the original image is displayed in Figure 10a and the tighter
crop in Figure 10b. The part of the image that is only visible in the wider crop consists of a big
part of the sky and a photographer on the right. The tight crop is more focused on the children,
which are the main subject of the picture, and thereby take up a larger portion of the image.
The tighter crop having a higher memorability score 昀椀ts the notion that images with a bigger
object size are more memorable and hints that memorability might be an (implicit) criteria for
edits done by photo-editors.
Another frequent variation was the image in Figure 10c. This image has a signi昀椀cantly
higher score than the original picture and was the only variation with a score in this range
that still fully depicted the original image. The original image was edited with a red-hued 昀椀lter
over the image. That this red-hued image has such a higher memorability score is expected as
the redness of an image has been demonstrated to positively a昀昀ect memorability.
5. Discussion
The 昀椀ndings in this paper are based on the result generated by MemNet. Even though MemNet
is a validated method of predicting memorability, the actual memorability and the computed
67
(a) The original full photograph with a memora- (b) Widely published tighter crop with a memo-
bility of 0.6767 rability of 0.7652
(c) A variation with a red 昀椀lter and a memorability of 0.8200
Figure 10: Three di昀昀erent variations of Napalm Girl found in online circulation.
memorability score could still di昀昀er. In conducted research, it became apparent that image
memorability and some qualities were correlated, like aesthetics and certain emotions. But
even though they are correlated, MemNet is not trained on these qualities; Thus, for example,
images that do not have the more common features that positively in昀氀uence the memorability,
like a big object size, but are very memorable because they portray a strong negative emotion
that also in昀氀uences memorability, possibly do not get a high memorability score from MemNet.
This highlights a mismatch between intrinsic memorability and actually being remembered. A
clear example of this mismatch is the Napalm Girl image, this image has a memorability score
below the average of LaMem. But Napalm Girl displays a scene of terror, that evokes strong
negative emotions like sadness and anger. Those strong emotions make this image more mem-
orable and highlight a limitation of (computational) intrinsic memorability methods. Moreover,
frequent exposure of less memorable images might also lead to increased remembrance, which
might also play a role for this image.
A possible limitation of this work is that the dataset was selected by Dutch researchers.
Despite being selected with the aim to represent international iconic images, the images are
predominantly known in the Western World. This could in昀氀uence the results, but this might
68
also interact with MemNet. As MemNet is trained on LaMem which consists mainly of images
from the Western World, this bias is is matched - whilst it should not in昀氀uence the analysis itself
it does limit to what extent we can generalize about the results. Additionally, the dataset of the
iconic images is mostly from the 20th century when color photography was not as common.
Most of the 26 images are in black and white. When experimenting with the di昀昀erences in
memorability score for the same image in greyscale as in color, it a昀昀ected the memorability
score; While the LaMem dataset averaged a memorability score of 0.7645 this score dropped
to 0.7456 when all images were converted to greyscale. From this we can observe that colour
plays a role, but not to the extent that is changes our conclusions. Even with taking these
points into account, the observations we made align with existing theories on memorability.
6. Conclusion
In answering the research question: How does the memorability of iconic images compare to
a baseline? It appeared that of the iconic images, the portraits where the face was a big part
of the image had a higher memorability score. This con昀椀rmed the research on memorability
where a correlation between higher memorability and big object size, and a face as an image
object was established. Additionally, the iconic images with small object sizes and small faces
had a lower memorability score. The memorability scores for these iconic images align with
theories on memorability. Furthermore, it appeared that images that are depicted in the media
are generally more memorable than all other images. Few images are depicted in media that
have a relatively low memorability score since all images depicted in media are selected by
editors. Iconic images are generally slightly less memorable than other images and are on the
lower side of the memorability of images in the media.
To answer the second research question: To what extent does the memorability iconic images
di昀昀er across variations? When comparing the spread of the memorability of the variations of
the iconic images, there were certain patterns to be found. Some had clear clusters, which were
other popular variations of the original image. Looking at examples of di昀昀erent variations and
memorability scores they show that altering an image can in昀氀uence the memorability score. In
these examples, it appeared that variations of the images with a red hue or tighter crop were
more memorable. This is in line with previous work.
On the whole we can conclude that computational measures of memorability do not fully
capture the memorability of iconic images, as many iconic images are remembered much better
than what their memorability score would imply. While the reasons for this may be manifold
we expect that frequent exposure and strong emotional content play an important role.
References
[1] R. Adler-Nissen, K. E. Andersen, and L. Hansen. “Images, emotions, and international
politics: the death of Alan Kurdi”. In: Review of International Studies 46.1 (2020), pp. 75–
95. doi: 10.1017/s0260210519000317.
69
[2] K. Andén-Papadopoulos. “The Abu Ghraib torture photographs: News frames, visual
culture, and the power of images”. In: Journalism 9.1 (2008), pp. 5–30. doi: 10.1177/1464
884907084337.
[3] T. Arnold and L. Tilton. “Distant viewing: analyzing large visual corpora”. In: Digital
Scholarship in the Humanities 34.Supplement_1 (2019), pp. i3–i16. doi: 10.1093/llc/fqz01
3.
[4] Y. Baveye, R. Cohendet, M. Perreira Da Silva, and P. Le Callet. “Deep Learning for Image
Memorability Prediction: The Emotional Bias”. In: Proceedings of the 24th ACM Interna-
tional Conference on Multimedia. Mm ’16. Amsterdam, The Netherlands: Association for
Computing Machinery, 2016, pp. 491–495. doi: 10.1145/2964284.2967269. url: https://d
oi.org/10.1145/2964284.2967269.
[5] A. F. Biten, L. Gómez, M. Rusiñol, and D. Karatzas. “Good News, Everyone! Context
driven entity-aware captioning for news images”. In: CoRR abs/1904.01475 (2019). arXiv:
1904.01475.
[6] P. Bruegel the Elder. Triumph of Death. 1563. url: https://www.museodelprado.es/en/th
e-collection/art-work/the-triumph-of-death/d3d82b0b-9bf2-4082-ab04-66ed53196ccc.
[7] A. A. Cohen, S. Boudana, and P. Frosh. “You Must Remember This: Iconic News Pho-
tographs and Collective Memory”. In: Journal of Communication 68.3 (2018), pp. 453–
479. doi: 10.1093/joc/jqy017.
[8] N. S. Dahmen and A. Miller. “Rede昀椀ning iconicity: A 昀椀ve-year study of visual themes of
Hurricane Katrina”. In: Visual Communication Quarterly 19.1 (2012), pp. 4–19.
[9] R. Dubey, J. Peterson, A. Khosla, M.-H. Yang, and B. Ghanem. “What Makes an Object
Memorable?” In: 2015 IEEE International Conference on Computer Vision (ICCV). 2015,
pp. 1089–1097. doi: 10.1109/iccv.2015.130.
[10] L. Goetschalckx, A. Andonian, A. Oliva, and P. Isola. “GANalyze: Toward Visual De昀椀ni-
tions of Cognitive Image Properties”. In: Proceedings of the IEEE/CVF International Con-
ference on Computer Vision (ICCV). 2019.
[11] R. Hariman and J. L. Lucaites. No caption needed: Iconic photographs, public culture, and
liberal democracy. University of Chicago Press, 2007.
[12] R. van der Hoeven. “The Global Visual Memory: A Study of the Recognition and Inter-
pretation of Iconic and Historical Photographs”. PhD thesis. Universiteit Utrecht, 2019.
[13] P. Isola, D. Parikh, A. Torralba, and A. Oliva. “Understanding the Intrinsic Memorability
of Images”. In: Advances in Neural Information Processing Systems. Ed. by J. Shawe-Taylor,
R. Zemel, P. Bartlett, F. Pereira, and K. Weinberger. Vol. 24. Curran Associates, Inc., 2011.
[14] P. Jing, Y. Su, L. Nie, and H. Gu. “Predicting Image Memorability Through Adaptive
Transfer Learning From External Sources”. In: IEEE Transactions on Multimedia 19.5
(2017), pp. 1050–1062. doi: 10.1109/tmm.2016.2644866.
[15] P. Jing, Y. Su, L. Nie, and H. Gu. “Predicting Image Memorability Through Adaptive
Transfer Learning From External Sources”. In: IEEE Transactions on Multimedia 19.5
(2017), pp. 1050–1062. doi: 10.1109/tmm.2016.2644866.
70
[16] A. Khosla, A. S. Raju, A. Torralba, and A. Oliva. “Understanding and Predicting Image
Memorability at a Large Scale”. In: 2015 IEEE International Conference on Computer Vision
(ICCV). 2015, pp. 2390–2398. doi: 10.1109/iccv.2015.275.
[17] M. Mortensen. “Constructing, con昀椀rming, and contesting icons: the Alan Kurdi imagery
appropriated by #humanitywashedashore, Ai Weiwei, and Charlie Hebdo”. In: Media,
Culture & Society 39.8 (2017), pp. 1142–1161. doi: 10.1177/0163443717725572.
[18] M. Moss. Toward the visualization of history: the past as image. Lexington Books, 2008.
[19] N. van Noord. “A survey of computational methods for iconic image analysis”. In: Digital
Scholarship in the Humanities (2022). doi: 10.1093/llc/fqac003.
[20] M. Parker. “The Retrospective Iconicity of ‘Guerrillero Heroico’”. In: Salford Postgraduate
Annual Research Conference. 2009, p. 292.
[21] S. Perera, A. Tal, and L. Zelnik-Manor. “Is Image Memorability Prediction Solved?”
In: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops
(CVPRW). Los Alamitos, CA, USA: IEEE Computer Society, 2019, pp. 800–808. doi: 10
.1109/cvprw.2019.00108.
[22] D. D. Perlmutter. Photojournalism and foreign policy : icons of outrage in international
crises. Praeger series in political communication. Westport, CO [etc: Praeger, 1998.
[23] N. C. Rust and V. Mehrpour. “Understanding Image Memorability”. In: Trends in Cognitive
Sciences 24.7 (2020), pp. 557–568. doi: https://doi.org/10.1016/j.tics.2020.04.001.
[24] Sarcophagus with the Battle of Marathon. 490 Bc. url: https://www.livius.org/pictures/it
aly/brescia-brixia/marathon-relief/.
[25] T. Smits and R. Ros. “Quantifying Iconicity in 940K Online Circulations of 26 Iconic
Photographs”. In: Computational Humanities Research. 2020.
71