Employing Aesthetic Principles for Automatic PhotoBook Layout

Philipp Sandhaus1, Mohammad Rabbath1, and Susanne Boll2

1 OFFIS – Institute for Information Technology, Oldenburg, Germanyfirstname.lastname@offis.de

2 Carl-von-Ossietzky Universita Oldenburg, Oldenburg, Germanysusanne.boll@informatik.uni-oldenburg.de

Abstract. Photos are a common way of preserving our personal memories. Thevisual souvenir of a personal event is often composed into a photo collage orthe pages of a photo album. Today we find many tools to help users creatingsuch compositions by different tools for authoring photo compositions. Sometemplate-based approaches generate nice presentations, however, come mostlywith limited design variations. Creating complex and fancy designs for, e.g., apersonal photo book, still demands design and composition skills to achieve re-sults that are really pleasing to the eye – skills which many users simply lack.Professional designers instead would follow general design principles such asspatial layout rules, symmetry, balance among the element as well color schemesand harmony. In this paper, we propose an approach to deliver principles of designand composition to the end user by embedding it into an automatic compositionapplication. We identify and analyze common design and composition principlesand transfer these to the automatic creation of pleasant photo compositions byemploying genetic algorithms. In contrast to other approaches, we strictly baseour design system on common design principles, consider additional media typesbesides text in the photo book and specifically take the content of photos into ac-count. Our approach is both implemented in a web-based rich media applicationand a tool for the automatic transformation from blogs into photo books.

Keywords: photo composition, photo book, photo album, genetic algorithms, aesthet-ics, photo collage, multimedia presentation

1 Introduction

Photos are a popular means to preserve the memory to important events in one’s life.They document a baby’s first developments, a vacation to an exotic destination, or thewedding of a happy couple. To remember and share these experiences, people tend tocreate visual compositions such as collages, calendars, or photo books. By the layoutand style of the composition the users aim to create visually appealing presentationswhich should reflect his or her experience of the event captured in the photos. However,many users do not have the compositional and technical skills needed in a creativeauthoring process with today’s commercial editing programs. A common approach toaddress this issue is to provide professionally designed templates as in commerciallyavailable authoring tools34. Presentations created this way usually lead to a pleasingresult but often rather factual and uniform presentation.

3 http://www.smilebooks.com4 http://www.apple.com/ilife/iphoto/

Skilled users might exploit all the different options of professional graphics author-ing tools and typically spend much time to manually achieve an if at all satisfying result.Depending on the user’s abilities the results might be great but given the skills and thetime the author needs to create the composition this is addressing only a few expertusers. In a time in which the number of digital photos is exploding, companies in thefield of photo finishing and photo services are searching for new products and businessmodels to convert these photos into visually appealing digital presentations or physicalproducts. We argue that the driving factor for success will be to “bring design to thepeople”, i.e., to transfer knowledge from visual arts and design into a wizard-like au-thoring environment. We can literally hear the outcry by professional designers sayingthat layout and design will never be automatable and that the end result will never bemeeting the standards of a manually and professionally curated presentation. Well, let’sgive it a try and see how far we can get for the end user.

In this paper, we present an approach for the automatic generation of photo com-positions based on fundamental design and layout principles. We follow a generativeapproach in which the photo composition is created from a set of underlying rules. Eachcomposition is unique and arranged according on the individual photo set. Existing ap-proaches are either purely template-based or provide algorithms producing layouts thatare usually not following any sophisticated, aesthetic design principles. We take a dif-ferent approach and specifically take aesthetic rules as our starting point. Additionally,we explicitly consider not only photos but also other media types such as texts for theresulting layout and accommodate for the specific characteristics of these.

The visual layout of a photo photo only attributes to part of the overall quality and isembedded into processes which consists of other activities such as content selection andannotation[21]. We have addressed these activities in earlier works[4,5]. In this paperwe assume that the content for a photo album has already been determined. We focus onthe aspects of distributing these contents over the pages and generating pleasing pagelayouts.

We start by reviewing similar works in the field and comparing them with our ap-proach in Section 2. We identify relevant design and layout rules from the literature andexamples of professionally designed photo albums which are the basis for the develop-ment of our automatic layout system described in Section 3. These rules are the basisfor our system for automatic photo book layout presented in Section 4. We describe twoapplications for our system in Section 5 before closing with a conclusion and outlookto future work.

2 Related Work

Most of the approaches for automatic layout and design of multimedia presentationsaim at optimizing the page layout, e.g., based on minimal white space or maximiza-tion of the number of photos or regions of interest. The creation of photo collagespresented by Girgensohn et al. [11] analyses photos for the region of interest and con-structs stained-glass like photo collages from photos with faces. Other approaches aimto infer a tree-like structure on the photos and base their layout on these structure. Forexample, the approaches presented in [1,2,14] map the result of hierarchical clusteringof photos directly to the spatial layout of the page. Others employ optimization tech-niques to minimize or maximize parameters such as the whitespace on the page or theocclusion of salient regions [23].

We also find approaches that derive clusters and importance of photos by contentand context analysis and employ this information to select appropriate, pre-defined lay-

out templates and placing the photos based on these [6,24]. AutoCollage [19] assemblesa set of images on a canvas by alpha masks to hide joins between the different imagesand employs energy maps and region of interest detection for distributing the imagesover the page. In a template-based approach by Diakopoulos et al. [8], pre-designedtemplates are used which consist of cells for photos and annotations applied to thesecells. The layout is filled by matching the metadata of photos to the annotations in thecells using an optimization algorithm. One of the few approaches that integrate designprinciples into the automatic presentation generation is presented by Lok et al. [16]. Intheir system, the authors introduce the concept of a WeightMap to solve the problem ofvisual balance for presentations. A work with a motivation similar to ours for automaticphoto collage layout is the one proposed by Geigel et al. [10]. The authors try to mimicthe artistic nature of the album layout process by employing genetic algorithms. A morerecent work [9] presents an interesting authoring system for the selection, of photos forphoto books and theme-based grouping, automatic background selection and automaticcropping. Following a similar goal, this approach however is based on a limited setof basic templates and does not consider images as backgrounds which, however, arevery common in professionally designed photo books. Another recent work of the samegroup [3] presents an aesthetically-driven layout engine for the automatic generation ofpage-based presentations with pre-defined static content where pre-assigned areas candynamically be filled with content whilst following aesthetic principles.

In conclusion, the approaches we find in the field provide methods which are algo-rithmically elegant but often do not sufficiently address the aesthetics of the generatedresults. We advance the work in the field in a number of ways: (1) We do not only con-sider photos as input to the system but also headings and text blocks, and we reflect theirspecific characteristics in the layout, taking into account that in the digital age photosoften come with additional labels and descriptions. (2) We implement a page designthat is strongly based on aesthetic principles stemming from common design rules andon the analysis of existing, professional designs. (3) We aim at creating compositionsthat are unique and reflect the features of just the individual media set.

3 Aesthetic Principles for Photo Books

One does not need to be a skilled designer to distinguish between an appealing andan unaesthetic photo book presentation. Some photo books instantly catch the viewer’seye and others are not really a pleasure to look at. Looking more closely at such photobooks reveals certain patterns and applied rules regarding layout and design in the moreappealing books. Many of these rules have been known for a very long time and areemployed by skilled artists and designers. In this section, we review selected rules forvisual layout from the literature and discuss how they can be mapped to the automaticgeneration of photo books.

The rules and principles we consider for visual layout are mainly adopted fromthe the works presented in [12,15]. Lidwell [15] gives a good overview over universaldesign principles which are approved over many years by many designers Itten [12]provides a good overview over different models for color combinations. In the follow-ing we further highlight two aspects of visual layout which has driven the design ofour system. Besides these, we also reviewed a couple of professionally designed photobooks.

3.1 Spatial layout

Two of the central aspects of a layout are the size and position of the contained items.A well-known underlying principle creating the impression of balance and stability isthe golden ratio (divine proportion) which describes a rule for the relation between twolengths: If a line ac is divided by a point b into the segments ab and bc the resultingsections follow the golden ratio if bc/ab = ab/ac = (

√5 − 1)/2. The golden ratio

can be applied to many aspects in the photo composition such as the proportions of thesize of different sub-areas or the position and relation of width and height of items ona page. Layouts the golden ratio are usually perceived as more balanced and appealingthan layouts that do not. The golden ratio is well-known for the design of still images,e.g., by placing the horizon or the main object in a picture not in the middle of the imagebut at a point according to the golden ratio. Another principle for distributing objectsand partitioning areas are different forms of symmetry. According to [15], symmetrycreates an impression of health and balance. Lidwell distinguishes between differentkinds of symmetry: reflection, rotation, and translation symmetry. Reflection symmetryrefers to the mirroring of an equivalent element around a central axis or mirror line.Rotation symmetry is referred to the rotation of equivalent elements around a commoncenter. Translation symmetry refers to the location of equivalent elements with the sameorientation and size in different areas of space. Translation symmetry is often combinedwith additional restrictions such as the placement of several elements along a line. Thesesymmetry principles can, e.g., be seen in nature: A butterfly or a human’s face exhibitreflection symmetry, a sunflower exhibits rotation symmetry in its petals around thecenter of the blossom.

3.2 Color layout

When different colors are applied to the same design they usually lead to different per-ceived emotions [15]. An important guideline is to limit the number of colors to theamount which can be processed at one glance. [15] suggests about 5 colors for this.We can observe that some color layouts are perceived as more appealing than others.Looking more closely at such appealing color combinations one can observe a numberof patterns. [12] has structured such combinations of colors in different color schemes.One of the main attributes of a decent color layout is the limitation of colors. [12] sug-gests not to consider more than three and has structured such color combinations intosix schemes. We employ these color schemes to determine appropriate color combina-tions for our automatic page layout system and to rate the overall color layout of a photobook page.

4 Automatic Photo Album Layout

The aesthetic principles for photo album layout discussed in the previous section formthe basis for our proposed system for automatic layout of digital photo books. As inputto our system we assume an ordered set of photos and text elements. Usually the orderof the photos is derived from their time stamps but can also depend on the application.The text elements are assumed to be assigned to one photo or a series of photos. Anorigin of these kinds of media could be a travel blog with text and images from whichone would like to create a photo book. But also other sources like photo managementtools or social media platforms would come with photos and additional text in somekind of order. We provide descriptions of a few applications in Section 5. Based on

Text Text

Text

Page 1 Content

Page 2 Content

Preperation Content Distribution

Background + High Level Layout

Detailed Foreground Layout

Text

Text

Fig. 1. System overview for our automatic photo album layout

the input media the system creates an automatic layout in five consecutive steps whichare illustrated in Figure 1. The Preprocessing step analyzes and structures the inputelements into different groups that form the basis for the different pages. These groupsare then assigned to the different pages in the Content Distribution step. The creationof the layout begins with the determination of the Background Layout of the individualpages. The creation of the layout of the foreground is divided into two steps: For a High-Level Foreground Layout the pages are divided into separate areas according to certainlayout principles, one for each element group. These areas are then laid out detailed inthe Detailed Foreground Layout step.

4.1 PreprocessingTo be able to assign the different image and text items to the single pages, the contentdistribution step expects the elements to be structured in a certain way. This is accom-plished in the preprocessing step. Basically two operations are performed on the inputphotos and text elements:

– In case no pre-defined order and structure is present in the photo set, they are or-dered and clustered according to their time stamps. For this we employ the algo-rithm presented in [17]. By this we keep semantically close photos also togetherin the final photo book and thereby support the story-telling aspect of a photo al-bum. The parameters of the clustering algorithm are chosen in a way that at most 5photos form a group and the majority of groups only consists of one photo.

– Large text blocks are divided into several text blocks to also allow longer text pas-sages in the photo book while avoiding cluttering the pages with text and maintain-ing a minimum threshold for the font size.

4.2 Content DistributionThe purpose of the content distribution step is to distribute the preprocessed elementsover the photo book pages. For this we assume that the number of pages is predefined,

which is reasonable in practical applications: For commercial photo book printing ser-vices, due to production limitations, usually only a fixed number of page numbers (e.g.8, 16, 32, ...) and page sizes are offered. An optimal distribution of photos and textelements is defined by best meeting the following restrictions:

R1 The predefined order of items should also be kept in the layout of the photo book.R2 The pages should be visually balanced, this means that roughly all pages should

contain the same amount of items.R3 The text to photo ratio should be roughly the same on all pages. Therefore, the

system should avoid to have pages only containing text or only photo items.R4 The content distribution should not extend the defined maximum of pages but at

the same time avoid empty pages. This restriction stems from practical facts insome of our applications. Usually, when designing a printed photo book, one canonly choose to increase the number of pages in multitudes of pages of, e.g. eight.A person who wants to order such an album usually wants to avoid having emptypages in the photo book but also wants to limit the number of pages as the price ofthe printed photo book increases with the number of pages.

R5 Ensure that all page elements do really fit the page. Images should not be scaleddown too much to ensure their visibility on the page and the font size of a theelement should not be too tiny to be easily readable. This leads to minimum sizesboth text and images.

R6 The color layout of the page should be balanced, this means combinations of pho-tos corresponding to one of the color schemes presented in Section 3 should bepreferred.

The problem of content distribution can be seen as an optimization problem ofwhich an optimal solution is that solution which best meets these partially compet-ing conditions. We opted to solve this problem with the help of a genetic algorithm.The result of the content distribution is an ordered set of pages each having assignedone or more groups of content elements.

4.3 Background Layout

After distributing all elements over the pages, the single pages are laid out individually.First for every page a background is determined as follows: An important principleregarding background vs. foreground of presentations is that the background should notdistract from the foreground. We also have found out that the majority of photo books isdesigned by placing a suitable photo in the background. We therefore wanted to preferlayouts where a non-distracting photo is used as the background for a page and if nosuitable photo can be determined, a color complementing the colors of the photos onthe page should be taken.

In our system, distraction is modeled by two components: Color and Saliency. Weassume that a photo with a lot of different colors distract the viewers eye more thanphotos only consisting of a few colors. We also assume that photos having a lot ofsalient regions like e.g. a photo showing a rough sea scenery do distract the viewermore than e.g. a photo with a very calm sea scenery. Thus, we have modeled these twoaspects as follows:

– Color: To determine the degree of colorfulness of a photo we analyze the histogramof the photo in the HSV Color space. We only consider the histogram of the Hcomponent and count the bins which exceed a certain threshold value. The degreeof colorfulness is then determined by the number of bins exceeding this value.

– Saliency: For every image a saliency map is determined. This map indicates regionsin a photo which potentially catch the attention of the viewer. For this map we em-ploy an extended version of the algorithm presented in [13]. The saliency map fromthe original algorithm is overlaid with a gaussian filter, which amplifies regions nearthe middle of the photo. This stems from our observations, that the important partsof an image are usually placed in the middle of the photos of lay photographers.Additionally, we employ the face detection algorithm of [22] to amplify regionscontaining faces. A score for the overall saliency of the image is determined bysimply calculating the average saliency value of the resulting saliency map.

Photos exceeding a certain threshold value are automatically rejected as a back-ground photo candidate. For the remaining candidates additionally the number of facesin each photo is determined. Then the photo with the least number of faces is chosen asthe background photo. If there are more than one photo with the least number of facesthe one is chosen with the least distraction rating. If there are no remaining photos auniformly colored background is generated which best complements the main colorspresents in the photos.

4.4 High-Level Foreground LayoutThe design of the page’s foreground is divided into two phases, a rough and a detailedlayout process. In this first step the page area is divided into several rectangular areas,one for each group. The spatial layout of these areas follows several, partly competingrestrictions:

RL1 All elements of one group have to fit in the assigned area.RL2 The pre-defined order of connected groups (text groups) should be reflected in the

layout.RL3 Keep text items within their preferred spatial extension.RL4 Prefer overall layouts following principles of golden section and symmetry.RL5 Prefer aspect ratios following the golden ratio for text items.RL6 Avoid changing the aspect ratio of groups with one photo.RL7 Prefer keeping headings in the upper part of the page.RL8 Do not cover salient areas of the background.RL9 Items should be evenly distributed over the page and thus the visual weight should

be placed in the middle of the page.

To meet these rules we again formulated the automatic design realized a geneticalgorithm with a fitness function rating designs according to these rules. The generationof rough layouts is accomplished by recursively dividing the page into rectangular areasfollowing principles of reflection symmetry and the golden ratio.

For this, we model the high-level page layout as a hierarchical graph with eachinner leaf holding to values hOrV to decide if the remaining area is split horizontallyor vertically and splitAmount to decide to which degree. The values of splitAmountare chosen according to the golden ratio. The leaves of the graph represent a sub-areaon the page. Two examples of such layout graphs and the resulting layout is depicted inFigure 2.

The fitness function of the genetic algorithm evaluating the resulting layout consistsof two parts. The first part consists of functions testing for mandatory conditions tothe layout. If these conditions are not met, the corresponding chromosome is deletedfrom the population. These conditions consist of the above mentioned restrictions RL1-RL3. If these restrictions are met, the chromosome is rated according to the followingdefinition (we also indicate in brackets [] which restriction is addressed):

Fig. 2. Examples of resulting layouts according to recursive algorithm

– r1 = #bordrat

#bord : This determines the percentage of borders in the layout, whichfollow the golden section. Considering the nested nature of constructed layouts, itcan happen, that borders do not follow this rule despite they are split accordingto the pre-defined slice primitives. This can also happen when considering aspectratios of pages, which do not follow the golden section. [RL4]

– r2 = #textgs

#text : The percentage of text items on the page, which aspect ratios followthe golden section. [RL5]

– r3 =P

i∈images coveredAreai

#images : The percentage to which the area can be covered bythe image for single image groups. This favors layouts following the aspect ratioof the image and downgrades areas being degraded to long stripes or not followingthe image’s aspect ratio. [RL6]

– r4 =P

h∈headings distanceToBottomh

#headings : For every heading in the page it’s distance tobottom of the page is determined favoring headings being in the upper part of thepage. [RL7]

– r5 = salienceSum(freeArea): If a page consists of a background photo, oneempty group is added to the page to allow keeping parts of the page free showinga very salient background. This rating determines the average salience value of thepart of the background being covered with this empty group. If the page does notcontain a background image, this part of the rating is discarded. [RL8]

– r6 = 1−P

e∈elementspage

ex−widthpage/2widthpage

+ey−heightpage/2

heightpage

#elementspage: The visual balance of the

page. By ex and ey the center of an element group e is defined. The closer the aver-age of all of these is placed to the center of the page, the higher the correspondingrating for the visual balance is. [RL9]

The overall rating function is defined as rlayout =∑6

i=1 ri.

4.5 Detailed Foreground Layout

Giving a high level layout for a page, the different sub-areas are further laid out depend-ing on the kind of their content. Areas containing only a single photo are filled entirelywith this photo while keeping a small border. If the photo does not have the same as-pect ratio as the assigned sub-area, it is cropped accordingly. For this the saliency mapof the photo is determined and the photo is cropped in a way that the resulting photocorresponds to the aspect ratio of the sub-area while keeping the most salient regions ofthe photo.

Areas consisting of text items filled entirely with this text. Additionally we ensure,that the text is easily readable on the background. For this, as described in Section 4.3,we analyze the color histogram of the covered area of the background. If the backgroundconsists of only a few dominant colors (we chose 2), a color for the text is chosen whichbest complements these colors. If the background consists of too many dominant colors,the part of the background is overlaid with a translucent white area and the text color isset to black ensuring easy readability.

Areas consisting of groups of photos are laid out according to a set of pre-definedlayout primitives. These are not templates but rather a set of visual rules to define thespatial relationships of a small set of photos. Figure 3 illustrates some of the layoutprimitives for our application which follow the principles of symmetry and the goldenratio. The figure shows only examples for a specific aspect ratio, the primitives areadjusted accordingly for different aspect ratios.

Fig. 3. Examples of layout primitives used for positioning photos inside a sub-area of a page

5 Application

One feature of our system is its ability to consider texts along with photos as input. Ad-ditionally, an already existing structure on these image and text items is also reflectedin the resulting layout. This enables the system to be used in applications where thesestructures are present and textual content is associated together with the photos. Exam-ples for this are e.g. travel blogs or photos hosted on social network platforms. Userswho want to convert for example a photo into a photo book would appreciate if theirvaluable editing effort be maintained and reflected in the automatic layout. On the otherside, our system is flexible enough to also provide sufficient results when this informa-tion is not available and can e.g. also work on raw photo sets. We deployed our layoutsystem for the automatic generation of photo books from blogs [20] and social commu-nity platforms [18]. Figure 4 shows the result of such an automatic transformation of atravel blog documenting a trip through Canada and the USA into a digital photo book.Shown are four pages. On three pages a photo was chosen as the background and on thefourth page a matching, uniformly colored background was added. Headings from theoriginal blog have resulted in headings and the entries’ text and image contents haveresulted in text and images on the pages laid out by our system [20].

6 Conclusion

We presented an approach for automatic layout of photo compositions which incorpo-rates the knowledge about aesthetic design principles. We identified and analyzed rules

Fig. 4. Example of two double-pages generated from a blog documenting a journey throughCanada and the USA

and principles of design in literature and from existing usage in professional photo al-bum pages for their applicability in the domain of photo compositions. The discussedprinciples of layout and design were systematically integrated the design of our auto-matic layout system. We implemented our approach both in a web-based rich mediaapplication for the manual and automatic creation of photo compositions and a systemfor the automatic transformation of blogs and social media into photo books.

With our automatic design aid end users are now able to create layouts that areappealing with very low effort and limited design skills. The effectiveness of our sys-tem could be validated by an informal user study with 10 subjects which showed thatthe advantages of an automatic creation of appealing compositions from photos wereappreciated by the users. The study supports the proposed kind of design support andshows that the resulting presentation are to the users’ satisfaction. Although the sub-jects of our user study were very pleased with the results produced by our system wesee potential for enhancement. In addition to the layout process we aim to support theinitial selection of photos. Not only the design but also the selection of photos may fol-low aesthetics of photographs as proposed by Datta et al. [7] who aim to qualitativelydistinguish between pictures of high and low aesthetic value.

References1. Atkins, B.: Adaptive photo collection page layout. ICIP ’04 5, 2897–2900 Vol. 5 (Oct 2004)2. Atkins, B.: Blocked recursive image composition. In: ACM MM ’08. pp. 821–824 (2008)3. Balinsky, H.Y., Howes, J.R., Wiley, A.J.: Aesthetically-driven Layout Engine. In: Proceed-

ings of the DocEng ’09. pp. 119–122. ACM Press (2009)4. Boll, S., Sandhaus, P., Scherp, A., Thieme, S.: Metaxa—context- and content-driven meta-

data enhancement for personal photo books. In: Cham, T.J., Cai, J., Dorai, C., Rajan, D.,Chua, T.S., Chia, L.T. (eds.) Advances in Multimedia Modeling. LNCS, vol. 4351, pp. 332–343. Springer (2007)

5. Boll, S., Sandhaus, P., Scherp, A., Westermann, U.: Semantics, content, and structure ofmany for the creation of personal photo albums. In: ACM Multimedia. p. 641. ACM Press,Augsburg, Bavaria, Germany (September 2007)

6. Chen, J.C., Chu, W.T., Kuo, J.H., Weng, C.Y., Wu, J.L.: Tiling slideshow. In: ACM MM ’06.pp. 25–34. ACM Press (2006)

7. Datta, R., Li, J., Wang, J.Z.: Learning the Consensus on Visual Quality for Next-Generation Image Management. In: Proceedings of the ACM Multimedia Confer-ence (September 2007), http://infolab.stanford.edu/∼wangz/project/imsearch/ALIP/ACMMM07A/

8. Diakopoulos, N., Essa, I.: Mediating photo collage authoring. In: UIST ’05. pp. 183–186.ACM Press (2005)

9. Gao, Y.e.a.: MagicPhotobook: Designer inspired, User perfected Photo Albums. In: Proceed-ings of MM ’09. pp. 979–980. ACM Press (2009)

10. Geigel, J., Loui, A.: Using genetic algorithms for album page layouts. IEEE MultiMedia10(4) (2003)

11. Girgensohn, A., Chiu, P.: Stained glass photo collages. Proceedings of ACM Symposium onUser Interface Software and Technology pp. 13–14 (2004)

12. Itten, J.: The Art of Color: The Subjective Experience and Objective Rationale of Color.Wiley & Sons (1997)

13. Itti, L., Koch, C., Niebur, E.: A model of saliency-based visual attention for rapid sceneanalysis. IEEE Transactions on Pattern Analysis and Machine Intelligence 20(11), 1254–1259 (1998)

14. Kustanowitz, J., Shneiderman, B.: Meaningful presentations of photo libraries: rationale andapplications of bi-level radial quantum layouts. In: JCDL ’05. ACM Press (2005)

15. Lidwell, W., Holden, K., Butler, J.: Universal Principles of Design. Rockport Publishers(2003)

16. Lok, S., Feiner, S., Ngai, G.: Evaluation of visual balance for automated layout. In: IUI ’04.pp. 101–108. ACM Press (2004)

17. Platt, J.C., Czerwinski, M., Field, B.A.: PhotoTOC: Automatic Clustering for Browsing Per-sonal Photographs. Tech. rep., Microsoft Research (2002)

18. Rabbath, M., Sandhaus, P., Boll, S.: Automatic Creation of Photo Books from Stories in So-cial Media. In: 2nd SIGMM Workshop on Social Media (WSM2010). Florenze, Italy (2010)

19. Rother, C., Bordeaux, L., Hamadi, Y., Blake, A.: Autocollage. In: SIGGRAPH ’06. pp. 847–852. ACM Press (2006)

20. Sandhaus, P., Rabbath, M., Boll, S.: Blog2Book - Transforming Blogs into Photo BooksEmploying Aesthetic Principles. In: ACM Multimedia 2010 - Technical Demonstrations.Florenze, Italy (2010)

21. Sandhaus, P., Thieme, S., Boll, S.: Processes of photo book production. Special Issue ofMultimedia Systems Journal on Canonical Processes of Media Production (2008)

22. Viola, P., Jones, M.: Rapid Object Detection using a Boosted Cascade of Simple Features.In: Proceedings IEEE Conf. on Computer Vision and Pattern Recognition (2001)

23. Wang, J., Quan, L., Sun, J., Tang, X., Shum, H.Y.: Picture collage. Computer Vision andPattern Recognition, IEEE Computer Society Conference on 1, 347–354 (2006)

24. Xiao, J., Zhang, X., Cheatle, P., Gao, Y., Atkins, C.B.: Mixed-initiative photo collage author-ing. In: ACM MM ’08. pp. 509–518. ACM Press (2008)