Automated Assistance in Evaluating the Layout of On-screen ...harbusch/ICEIS-2016-SEAPTool.pdf · Automated Assistance in Evaluating the Layout of On-screen Presentations Karin Harbusch

Automated Assistance in Evaluating the Layout of On-screenPresentations

Karin Harbusch1, Denise Dunnebier1, Denis Krusko1

1Computer Science Department, University of Koblenz-Landau, Gernany{harbusch|dduennebier|kruskod}@uni-koblenz.de

Keywords: Human-Computer Interaction, Interface Design, Personalized Feedback, Presentation Design, PresentationLayout, Evaluation Assistant System

Abstract: Oral presentations can profit decisively from high-quality layout of the accompanying on-screen presentation.Many oral talks fail to reach their audience due to overloaded slides, drawings with insufficient contrast, andother layout issues. In the area of web design, assistant systems are available nowadays which automaticallycheck layout and style of web pages. In this paper, we introduce a tool whose application can help non-expertsas well as presentation professionals to automatically evaluate important aspects of the layout and design ofon-screen presentations. The system informs the user about layout-rule violation in a self-explanatory manner,if needed with supplementary visualizations. The paper describes a prototype that checks important generalguidelines and standards for effective presentations. We believe that the system exemplifies a high-potentialnew application area for human-computer interaction and expert-assistance systems.

1 INTRODUCTION

At the onset of their oral presentations, speakers oftenapologize for the potentially suboptimal quality of theaccompanying visual slides1. They wonder whetherthe audience can see presented curves at all becausethe contrast between foreground and background ispoor such as yellow on white background. They askwhether the audience in the back can read 12pt fontswell enough and similar questions meant to be rhetor-ical — the audience often perceives them as cynical.

Why can’t assistant systems inspect slides duringwriting? In many areas of human-computer interac-tion, such as web-site design, assistant systems areavailable nowadays but, to our knowledge, not in thearea of audiovisual presentations. The present paperdescribes a prototype that automatically checks vari-ous general guidelines and standards for effective au-diovisual presentations.

In our system, short traffic-light inspired bars in-form the user about the evaluation result — on de-mand supplemented by a more elaborate explanation.In the list of preferences, the user can deselect fea-tures s/he is not interested in along with personal-ized values overwriting the system’s defaults. For in-

1Although true slide projection is hardly in use anymore,the term slide is still very common for the virtual counter-part of the once physical exemplar.

stance, the slides might become more filled in a lec-ture than in a business talk. In this paper we focuson the feedback visualization of noticed violations ofpresentation rules. We illustrate precautions of thesystem so that novices as well as experts can easilyuse our system. The implementation of algorithmssuch as calculating the fullness of a slide or how todetect low contrast is not discussed here.

The paper is organized as follows. In the next sec-tion, we sketch the state of the art in assistant systems.In Section 3, we specify important to-be-evaluatedcriteria in the area of (audio)visual presentation de-sign. The current prototype is discussed inn Section4. In the final section, we draw some conclusions andaddress future work.

2 STATE OF THE ART INASSISTANT SYSTEMS

Automated assistance in user-interface design is a rel-atively young but dynamic field. The trend in this di-rection tries to countervail the huge amount of poorlydesigned interfaces — a phenomenon supported byeasy to use tools for implementing a dialogue system.An early seminal attempt constitutes the frameworkDON (Kim and Foley, 1993), which uses rules from a

knowledge base to provide expert assistance in user-dialogue design. It can generate layout variants in aconsistent manner. Subsequent development of theassistant systems proceeded in two main directions —graphic art (printing) and web — , and has alreadygiven rise to expert-assistant systems with commer-cial applications.

In the graphic-arts industry, quality control beforeprinting plays a crucial role by reducing the costs ofreprinting. The process has been dubbed ”pre-flight”.In general, the term designates the process of prepar-ing a digital document for final output as print, plateor for export to other digital document formats. Thefirst commercial application was ”FlightCheck”2 de-scribed in a paper entitled ”Device and method forexamining, verifying, correcting and approving elec-tronic documents prior to printing, transmission orrecording” Crandall and Marchese (1999). Recentproducts in the area provide an integrated preflightfunctionality (see, e.g., Adobe InDesign3 and AdobeAcrobat4). The main objective of these instrumentis to reveal possible technical problems of the doc-ument. Accordingly, they work with the followingprimary checklist: (1) Fonts are accessible, compat-ible and intact; (2) Media formats and resolution areconforming; (3) Inspection of colors (detection of in-correct/spot colors, transparent areas); (4) Page infor-mation, margins and document size.

According to Montero, Vanderdonckt & Lozano(2005), the abundance of web pages with poor usabil-ity is largely due to shortage of technical experts inthe field of web design. In web design, Ivory, Man-coff & Le (2003) present an overview of systems thatare capable of analyzing various aspects of the webpages. Historically different browsers have differentviews on the implementation of web-standards (see,e.g., (Windrum, 2004)) with as a consequence thatthe same web page may look differently in differentweb browsers. The above criteria have led to the situ-ation that tools for web-page analysis focus primarilyon technical and marketing aspects of the web pages.Current web analysis tools primarily check:

• W3C5 DOM, HTML and CSS standards;

• Search engine optimization (SEO) aspects;

• Web page performance and rendering speed;

2FlightCheck (Preflight for Print), http://markzware.com/products/flightcheck (Nov. 11, 2015).

3Adobe InDesign CC, http://www.adobe.com/

products/indesign (Nov. 11, 2015).4Adobe Acrobat, https://acrobat.adobe.com (Nov.

11, 2015).5World Wide Web Consortium (W3C), http://www.

w3.org (Nov. 11, 2015).

• Content, media and script sizes;

• Accessibility of various devices.

Despite the emphasis on purely technical aspects,several publications report on systems assisting userson other aspects of web design (e.g., (Tobar et al.,2008)). Some state-of-the-art systems (see, e.g.,(Nagy, 2013)) advise on visible content prioritizing,check the size of control elements (e.g., some con-trols may be too small for using on mobile devices),and distances between the visible elements of a webpage.

An essential question concerns whether or not as-sistant systems should react directly, in a daemon-likefashion, to any undesirable user action (maybe evenforbidding and overruling user actions), or should be-come active only on demand. The majority of systemsmentioned above prefer the on-demand dialogue. Ba-sically, the decision depends on the aspect evaluated.For instance, if the system cannot react to a user ac-tion such as saving a file in the current format, theconsequence should be brought to the user’s attention.In case of less disastrous effects, the system can reactin two manners. According to the first alternative, noill-formed result can be produced at all (e.g., auto-matic word correction for typos in SMS typing avoid-ing unknown words). However, in this mode, the usermight feel too much patronized. As a consequence,users tend to switch off such components. Moreover,the second alternative of giving advice on demand of-fers more freedom to the user (e.g., new words canbe typed). In design, violation of rules is used as astylistic matter (cf. provocative design).

3 PRESENTATION RULES

Here we summarize well-known standards for user-interface design in general that also apply to presen-tation design. Additionally, we list rules of thumbspecifically for presentation design in particular. Dueto space limitations, we cannot give a comprehensiveoverview of such rules and standards, and instead fo-cus on the type of rules that our system checks auto-matically.

Many user-interface design rules (cf. the norm ENISO 9241) can be applied for a slide presentation aswell: use only few different colors; avoid high colorsaturation levels; give sufficient contrast to the usedcolors; group related elements together potentiallywith a frame around them, and/or make sure thereis sufficient spacing between non-related items (cf.Gestalt theory; see, e.g. the reprint of original workin (Wertheimer, 2012)); do not make the interface too

crowded; distribute objects such that the virtually as-sumed grid lines are minimized (i.e. make the inter-face — in our case, the slides — look balanced andsophisticated; cf. (Galitz, 2007)). The recommenda-tion not to overtax the short-term memory of the userin interface design also holds for a slide: it restrictsthe number of presented items to 7+/- 2 per slide (cf.Miller’s rule, (Miller, 1956)). In total, no more than30% to 40% of a slide’s surface should be occupied.

For consistency reasons (cf. (Shneiderman andPlaisant, 2004)), font, size, position and color of theslides should remain the same in publishing media.This holds in particular for the title. Moreover, thelatter’s position should remain the same on each page.Often a predefined frame is assumed for a user inter-face (cf. the slide master in PowerPoint6 for the adap-tation to visual presentations).

Furthermore, a wide variety of books focuseson specific rules in visual-presentation design. Thebooks target at different user needs such as presen-tation for beginners or for professional presentersin business. For instance, for non-designers, RobinWilliams (2015) cites four principles of visual presen-tation design: Contrast, Repetition, Alignment andProximity.

We focus on the following rules of thumb that, weassume, hold for business presentations. They repre-sent the defaults of our prototype:1. Do not use more than two font types in a presen-

tation7;2. Do not use fonts smaller than 18pt;3. Do not use more than three colors;4. Avoid saturated colors (threshold 30%);5. Provide sufficient contrast for chosen colors/gray

values (threshold 10%);6. Provide sufficient distance between unrelated ob-

jects (as opposed to related objects which shouldbe closer together due to Gestalt theory effects;horizontal = 0.8cm, vertical = 0.8cm8);

7. Provide a balanced distribution of elements (max-imum number of grid lines = 20 with unified dis-tance of .3cm);

6See PowerPoint, http://products.office.com/

powerpoint (Nov. 12, 2015).7This task can be extended with a check for whether dis-

preferred fonts are being used (e.g. Antiqua; for pros andcons of various fonts, see, e.g., (Williams, 2015)). Our de-fault list is based on (Schildt and Kursteiner, 2006). Theuser can edit this list (as s/he can any default parameter ofthe system).

8Notice that these values can also be automatically cal-culated using the font size used in the currently consideredbox (cf. (Galitz, 1991)).

8. Slides should not be too full (threshold 30%).

For convenience of the audience, provide automaticprint versions without images and/or inversion of adark background to white with automatic inversion ofthe foreground colors to black or a user-defined value.Notice that this mode is not discussed in the followingfor reasons of space.

As will be outlined in the next section, the abovementioned features are first checked per slide accord-ing to the default or user-defined parameter settings.The per-slide evaluation reports are subsequently in-spected for overall consistency of the entire presenta-tion.

4 SEAP TOOL: A PRESENTATIONASSISTANT SYSTEM

The name SEAP stands for Software-ErgonomicAnalysis of Presentations. First, we describe SEAPtool’s system design, e.g., its input and output struc-ture. Then, we focus on the inspection per slide. InSection 4.3, we elaborate on the preferences the usercan express for any feature in any particular slide.Section 4.4 indicates how the content of the per-slideevaluation report is used for checking the overall con-sistency of the presentation.

4.1 System Design

Our prototype is implemented in Java 89. As themain input format, we use Portable Document Format(PDF) ), being the de facto standard for fixed-formatelectronic documents (cf. ISO 32000-1:200810). Thismeans that any presentation can be analyzed that canbe exported as PDF, irrespective of the slide presenta-tion program or the operating system with which thepresentation was created.

The PDF format also provides access to the pre-sentation’s internal content stored as text, as rasteror vector graphics, or as multimedia objects. Ifavailable, we use this information for the subsequentslide analyses. However, an analyzed slide mayconsist of only a picture, without any text informa-tion (e.g., when the entire slide is the snapshot ofa screen). In this case, or when graphical elementson the slide display text, we use the computer vi-

9Java Software, https://www.oracle.com/java/

index.html (Nov. 12, 2015).10See http://www.iso.org/iso/iso_catalogue/

catalogue_tc/catalogue_detail.htm?csnumber=

51502 (Nov. 12, 2015).

sion library OpenCV11 to identify the objects. Ob-viously, this variant is computationally more com-plex and more time consuming which is reflected inlower processing speed especially when producinga report for a larger input file. However, the sys-tem gains independence from the actual representa-tion format of the content of the slide. In the follow-ing, we do not elaborate on implementation detailsof the two different methods to obtain an evaluationresult. (See (Dunnebier, 2015). This paper also dis-cusses the estimated quality of the evaluation algo-rithms applied in SEAP.) For the levels of detail wediscuss here, it is sufficient to agree that any evalua-tion result we refer to in the following can automati-cally be calculated.

Given the decision to inspect a PDF file of the pre-sentation, the way SEAP tool provides the output isalso determined. As mentioned in Section 2, an assis-tant system can evaluate online during the design pro-cess, or produce a review on demand. The latter (alsoSEAP tool’s) has the following advantage of avoidingdisturbing the user, especially during stages where thefocus is on content rather than form. However, thisdecision has a drawback: information that would beimmediately at hand online (e.g.: Which areas belongto the master slide? Which text box is meant to be thetitle?) has to be recomputed.

We target different user groups: not only novicesbut also presentation professionals. Basically, the re-port aims at easy understandable comments (e.g., interms of visualizations rather than technical terms incase of novice users). Professionals receive shorttraffic-light-style comments only.

Moreover, the personal settings for all parametersof the individual evaluation algorithms allow differ-ent levels of detail. Inexperienced users see intuitivelabels. Professional users can operate an “Advanced”button to enter exact values. (e.g., see Figure 6 in Sec-tion 4.3 for the interface enabling personalization ofthe grid inspection parameters).

In the next paragraph, we delineate the evaluationof an individual page varying according to dedicateduser-preferences.

4.2 Report Generation per Slide

In the summary of any specific feature in an evalu-ation report, the green vs. red background indicatescompliance or failure of the rules. This traffic-light-style information helps professional users to speed upreading — on the assumption they search for red bars

11OpenCV (Open Source Computer Vision), http://

opencv.org (Nov. 12, 2015).

only (cf. Figure 1). It also supports users who are un-familiar with presentation rules. They can read thetraffic light colors as hints whether they are on theright track or not. Moreover, we present informativevisualizations whenever possible. If desired, the re-port can become personalized in two respects:

1. The user has the option to define personal pref-erences overruling the default settings used in thealgorithms check.

2. Additionally, the system offers the choice be-tween short or elaborate report.

Figure 1: Concise analysis report. The user has asked thesystem to check font size and crowdedness only: Positivefeedback for used fonts is displayed against a green back-ground, negative feedback on crowdedness against a redbackground.

In the following, we focus on the elaborate re-porting mode. On each slide, SEAP tool counts thenumber of different fonts and compares it against thethreshold (whose default value is two). It also checksthe occurrence of user-defined but generally dispre-ferred fonts. Figure 2 illustrates the most elaborateversion of a font warning generated by SEAP tool.Color saturation warnings and warning for too manydifferent colors on the same page look similar. Forreasons of space, we skip details here.

Figure 2: Elaborate font information based on the yellowrule of thumb in the right panel.

Whenever possible, visualizations are used to in-form the user in a self-explanatory manner so thatnon-professionals can use the system as well. For in-stance, the system exemplifies whether closely neigh-boring objects are presumably perceived as belong-ing together according to the Gestalt laws. The sys-

tem transforms such objects into one abstract box12

in line with the default or user-defined threshold (cf.Figure 313 corresponding to the slide depicted in Fig-ure 1). Notice that, here, the system does not attemptto warn against errors but merely visualizes the mostlikely grouping perceived by the audience. Therefore,only the user — not the system — can adapt the slideto the intended content. The image also illustratesthe difference between PDF-based and image-basedinspections. In the PDF file, the two text items areshown in one box (cf. green boxes in the grid repre-sentation in Figure 4 according to the predefined set-tings to highlight text compared to images outlinedin Figure 6 in the next subsection). However, giventhe current threshold setting, an image analysis of theslide would interpret the text items as two indepen-dent boxes. Consequently, the user might feel in-cluded to improve the slide by positioning the twotext items closer together. In SEAP tool, we currentlytake the PDF information about text to determine textboxes. Thus, no conflict needs to be resolved.

Figure 3: Recognized objects belonging together for athreshold bigger than the distance between the text box withthe two items and the two images but smaller than the dis-tance between the two images.

In a similar manner, the information is visual-ized whether a balanced distribution of objects pre-vails giving the impression that the user has immersedin the presentation design. A visualization depictsthe virtual grid according to a threshold determiningwhich distance is assumed to be one unit. For in-

12In this figure, we use black as the color denotingsuch boxes because this yields better interpretability of thescaled-down image. In SEAP tool, the user can select anycolor and any level of transparency.

13The obvious grid violation of an exact vertical align-ment of the two images is intended here. We use the sameimage for illustrating the virtual grid calculation later on inthe section. At the moment, one can see that the default pa-rameter for grid inspection can be considerably high. Obvi-ously, the original slide as presented in Figure 1 looks bal-anced.

stance, on the slide in Figure 1, the two images arenot fully vertically aligned (cf. Footnote 13). A veryexact threshold (e.g., .1cm) would indicate two verti-cal grid lines to the left and two vertical grid lines tothe right of the images. If the threshold would be setto a more lenient value, there is only one grid line cal-culated. Figure 4 delineates the result for the case ofan exact threshold in order to illustrate the power ofthe automatic calculation. As holds for all preferencesof SEAP tool, the color of boxes and lines displayedin order to highlight the meta information on a slidecan be determined by the user to clearly separate theprevailing colors on the slide from the evaluation in-formation added by SEAP tool.

Title

• Lorem ipsum dolor sit amet, consectetueradipiscing elit. Aenean commodo ligula egetdolor. Aenean massa. Cum sociis natoquepenatibus et magnis dis parturient montes, nascetur ridiculus mus.

• Donec quam felis, ultricies nec, pellentesqueeu, pretium quis, sem. Nulla consequat massaquis enim.

Figure 4: Visualization of the grid lines of objects illustrat-ing whether the order of objects is reduced or not.

As for contrast evaluation against a given thresh-old, in the current SEAP tool version, each slideis translated into a grayscale version by applyinga blackandwhite filter, e.g., a dithering algorithm14.The concise report can issue warnings that informa-tion has disappeared. Too close color similarities canalso be noticed if the threshold is refined. In the elab-orate version, slide areas with missing information arehighlighted, so that the user does not overlook easilymissed details. Currently, we run experiments withthe determination of contrast indicated directly on theoriginal page without applying a black-and-white fil-ter. Additionally, a new evaluation rule should beadded to the list of evaluated features which applies analgorithm that can recognize colors invisible to color-blind users.

In the next paragraph, we discuss how user-defined preferences are entered. Here, it is importantto use terminology that any kind of user understands— not only experts.

14For an easily understandable and nicely visually sup-ported description see http://www.tannerhelland.com/4660/dithering-eleven-algorithms-source-code/

(Nov. 12, 2015).

4.3 User-specific Preference-Dialoguesfor Individual Slide Inspection

In this section, we introduce parameter settings foran individualized slide inspection. Notice that theuser can tailor slide-specific defaults as well aspresentation-general ones. The latter ones are dis-cussed in the separate Section 4.4 because checksof overall consistency deploy the per-slide reports.Moreover, the preference menu offers a separate sub-menu for the overall presentation parameters. Thismenu also allows skipping the final overall evaluationif the user is not interested in this inspection, or whens/he is finalizing the presentation.

Invariably, the user can select which features tobe evaluated per slide. This can speed up the processconsiderably15. Moreover, the user might be inter-ested in specific feedback only. Thus, s/he gets a listproviding the options (1)-(8) presented in Section 3to select or deselect from. If s/he deselects an item, itbecomes gray and moves to the end of the list. Thisbehavior should elicit another user-option available inthis window. The user can order the sections of theevaluation report. In the top of the window, the useris informed that the list can be re-ordered if desired.In Figure 5, the window is depicted in the originalorder. However, the Figure depicts a state where theuser has deselected the last five items (cf. gray color).Of course, any choice and order can be revised beforeapplying. Pushing the ”Abort” button remains withthe previous settings. Pushing the preselected ”Ap-ply” button tailors the report according to the user’spreferences.

After the user has left the window — irrespec-tive of pushing the ”Apply” or ”Abort” button, the re-maining items provide the choice for a short or a longreport variant per feature in the subsequent window.This window provides a flipping choice button wherethe default is on long as we assume in the beginningthat the user, either a novice or even a professional,might like to become used to SEAP tool’s feedbackbehavior. For reasons of space, we have skipped thiswindow here.

Beside the dialogue about the overall order anddetailedness of the report, the user can overwrite anydefault-parameter setting of any feature chosen to bechecked in the report. Menu items referring to dese-lected features for evaluation remain inactive — de-picted in gray. We display items always in the same

15Notice that there exists an invisible part of the reportrequired for the overall presentation checking (see next sec-tion). If the user wants many general consistency features tobe checked according to the personal preferences, the sys-tem can obviously not speed-up.

Figure 5: Personalization of features to be evaluated alongwith the option to personalize the order of result presentedin the report.

order irrespective to the report order chosen by theuser as it is presumably faster to search in a fixed-order menu. Figure 6 shows an example that avoidsexact numbers to be changed which is assumed to bethe desired mode for novices. The example illustrateshow not professional presenters can intuitively workwith the SEAP tool. More abstract terms instead ofexact values are provided to allow the user to makea meaningful choice. Experts probably prefer a win-dow where they can change the default value directly.The current prototype is not fully capable of differentmenus for all features. Accordingly, we currently re-vise and extend these dialogues considerably for thenext version.

Figure 6: Upper half of the dialogue window: Setting of thegrid evaluation parameter in the preferences list in a mannerthat non-professional users can make a meaningful choice;in the lower panel, the button named ”Advanced” locatedabove the final choice ”Abort/Apply” provides a windowwith detailed setting options by numbers, preferably usedby the experts.

All defaults as outlined in Section 3 can be over-written. Furthermore, the list of non-accepted fontscan be modified. For reasons of space, we do not elab-orate on the fact that there are predefined forbiddenvalues (e.g., zero fonts to be used at all). Of course,the algorithms activated during the evaluation processfirst check explicitly whether the ranges set by theuser are acceptable. Otherwise, the system would un-

expectedly crash.Based on all these settings, the user gets a review

per page in his/her personal style (e.g., a brief traffic-light coding for some selected features only). In itsmost elaborate mode, the report sums up positive andnegative evaluation results for all inspected problems.Additionally, it can provide hints to why/how the slideshould be changed.

In the next section, we describe how the evaluationreport itself is deployed to detect overall consistencyviolations.

4.4 Evaluation of Overall Consistency

In this section we discuss features that can be checkedconsistently in all slides such as whether or not thesame font has been used throughout the presentation.The user can switch this evaluation on/off in the samemanner as for the features to be checked per slide.

Presuming that the user wants a consistencycheck, in case of a violation, the user gets a resumewhere to find these cases. The visible and invisi-ble information of the evaluation report per page, en-ables the system to produce such a report automati-cally. However, the system needs additional informa-tion about the presentation to do a more advanced job.

SEAP tool should know about facts like assumedtitle position. Basically this is known at design timeof the slides but not noticeable in the PDF file SEAPtool uses as input. As for a hypothetical minimalslide master, the system assumes as default a marginarea around the presentation of 1cm. This assump-tion fulfills a general rule of thumb that one shouldleave some margins all around a slide. A title areais not preset by the system because warnings aboutany violation would irritate the user who has no ideawhere the system assumes the title to be — there areno user expectations the system can take for granted.These (minimal) default settings avoid an obligatorydialogue with the user before running the system.

As for any preference in SEAP tool, the user canmake changes for these defaults. For better results,the preference menu provides windows that allow tai-loring these settings. Here, the user can define areasto remain as uninspected vs. inspected for identity16.Figure 7 illustrates the determination of a slide master— an empty frame is assumed on the example slide

16The current version of SEAP tool applies an exactmatch algorithm. We are aware of the fact, that the matchshould be less exact in order to license page numbers in thisarea as well as slight color/size variation as one often seemsto highlight the currently active section in the content listprovided in a frame or similar tiny differences that shouldnot account for non-identity.

depicted in Figure 1 — to be identical/ignored overall slides. The individual margin areas can be variedas indicated by the red arrows depicted in the mid-dle of each default margin of 1cm from each side ofthe slide. The inspection method deploys as defaultan identity check. We omit the dialogue to select be-tween the options of ignoring the area or to match itthroughout the presentation, respectively. This choicewindow pops up when activating the red arrow ofchanging an area or when double-clicking on the area.As a consequence, the color of the region changes.Blue means check exactly whereas red means ignorethe content completely. First experiments show hand-iness of the concept in the right vertical periphery —as is depicted as desired personalized setting in Fig-ure 7. This setting allows the user to violate the rightmargin supposed to be identical with the master slideintentionally due to longer lines.

Figure 7: User interface determining the master-slide areain the presentation in a dialogue with the user varying thedefault area to be matched exactly on all slides or to be ig-nored on all slides. The areas in blue reflect the wish for anexact match and the ones in red, for ignoring any differencein the chosen rectangular.

In a similar interactive window, the area for the ti-tle can be specified. Of course, for titles the check isnot meant to be an exact match. The user can deter-mine which features should be checked (defaults arefont type, font size and color). The same dialogue-window type opens if the user wants to keep more se-lected areas checked for consistency (e.g., page num-bers). In these windows the title of the blue area hasto give the region a unique name before applying. Theevaluation report refers to warnings for these areas byusing the user-defined field names. The same param-eterized procedure inspects the title area as well asthe user-defined areas (parameters: name, coordinatesand features to be checked throughout the presenta-tion). The source of inspection is the fully elaborateform of the evaluation report per slide that SEAP toolyields internally.

In the end of the report, the consistency-checksummary is provided to the user. It results from a

final inspection of all internal entries in the evaluationreport per page for each feature the user wants to beglobally checked. For instance, the system can gener-ate a warning in the final summary such as ’Attention,on slide 4, the font of the title is inconsistent. Pleasechange from Times to Arial.’.

5 CONCLUSIONS

In this paper, we have delineated a prototypical assis-tant system for the evaluation of visual presentations.We have illustrated the diversity of topics automati-cally checked by our system. So far, no such tool isavailable on the market. Given the often poor qual-ity of presentations in science and business, such asystem for automatic layout and design evaluation ofslides is highly desirable.

Our system, can automatically evaluate visual pre-sentations according to well-known rules for design-ing any kind of user interface as well as some specificpresentation rules. Our system reads-in a PDF fileof the presentation so that the system is independentfrom the way the user obtained the presentation. Thesystem performs different inspections on the PDF fileas well as some analyses of an image representation ofeach slide. Based on these results, the SEAP tool de-livers an evaluation report per slide in a personalizedmanner. Furthermore, the user can determine whichfeatures are evaluated at all and in which order theresults are presented. In addition, individual parame-ters for the evaluation calculation can be personalizedalong with the level of detail of the report. If the userwants, general consistency throughout the slides canbe evaluated as well at the end of the report.

As for future work, we plan to translate new rulesof presentation design and layout into automatic eval-uation procedures. For instance, as outlined in Sec-tion 4.2, color-blind proof-reading of slides shouldbe available. Furthermore, a deeper image analysisshould be imposed on the system. As mentioned inthe previous section, many existing analysis compo-nents of SEAP tool can be improved. Moreover, newchecks can be added based on the existing compo-nents. Additionally, user studies should be conductedto test the user interface of the system with specifictarget groups like novices and professional users, re-spectively. We paid attention to the fact that the dia-logues become clear even to novices. In this regard,we supported text with intuitive visualizations. How-ever, only a study can provide clear insights into anoptimal user interface. Moreover, we would like toask the users what is their ranked list of features theywould desire most.

ACKNOWLEDGEMENTS

We grateful to Gerard Kempen for his valuablecomments on earlier versions of the paper.

REFERENCESCrandall, R. and Marchese, P. G. (1999). Device and

method for examining, verifying, correcting and ap-proving electronic documents prior to printing, trans-mission or recording. US Patent 5,963,641.

Dunnebier, D. (2015). Software-gestutzte Generierungvon ergonomischen Verbesserungsvorschlagen zurDarstellung von Prasentationen. Bachelor Thesis,University of Koblenz–Landau.

Galitz, W. O. (2007). The Essential Guide to User Inter-face Design: An Introduction to GUI Design Princi-ples and Techniques. John Wiley & Sons, 3rd edition.

Ivory, M. Y., Mankoff, J., and Le, A. (2003). Using auto-mated tools to improve web site usage by users withdiverse abilities. Human-Computer Interaction Insti-tute, page 117.

Kim, W. C. and Foley, J. D. (1993). Providing high-levelcontrol and expert assistance in the user interface pre-sentation design. In Proceedings of the INTERACT’93and CHI’93 Conference on Human Factors in Com-puting Systems, pages 430–437. ACM.

Miller, G. A. (1956). The magical number seven, plusor minus two: some limits on our capacity for pro-cessing information. Psychological review, 63(2):81–97. Reprinted in Psychological review (1994),101(2):343.

Montero, F., Vanderdonckt, J., and Lozano, M. (2005).Quality models for automated evaluation of web sitesusability and accessibility. In International COST294workshop on User Interface Quality Models (UIQM2005) in Conjunction with INTERACT.

Nagy, Z. (2013). Improved speed on intelligent web sites.Recent Advances in Computer Science, Rhodes Island,Greece, pages 215–220.

Schildt, T. and Kursteiner, P. (2006). 100 Tipps und Tricksfur Overhead- und Beamerprasentationen. Beltz Ver-lag, 2. uberarbeitete und erweiterte Aufl. edition.

Shneiderman, B. and Plaisant, C. (2004). Designingthe User Interface: Strategies for Effective Human-Computer Interaction. Addison Wesley, 4th edition.

Tobar, L. M., Andres, P. M. L., and Lapena, E. L. (2008).Weba: A tool for the assistance in design and evalua-tion of websites. J. UCS, 14(9):1496–1512.

Wertheimer, M. (2012). On Perceived Motion and FiguralOrganization. The MIT Press.

Williams, R. (2015). The Non-Designer’s Design Book.Peachpit Press, 4th edition.

Windrum, P. (2004). Leveraging technological externali-ties in complex technologies: Microsoft’s exploitationof standards in the browser wars. Research Policy,33(3):385–394.