Papers
`
`CHI 99 15-20
`-
`
`MAY
`
`1999
`
`FofoFile: A Consumer Multimedia Organization and
`Retrieval System
`
`Allan Kudhinsky, Celine Pering, Michael L. Creech, Dennis Freeze, Bill Serra, Jacek Gvvizdka*
`Hewlett Packard Laboratories
`150 1 Page Mill Road
`Palo Alto, CA 94304 USA
`+1650857
`1501
`{ kuchinsk, celine, dff, creech, bills} @ hpl.hp.com
`(* Current address: jacek@ie.utoronto.ca
`, Interactive Media Laboratory: Department of Mechanical and Industrial
`Engineering, University of Toronto, 4 Taddle Creek Rd, Toronto, Ontario, Canada M5S lA4 )
`
`ABSTRACT
`for multimedia
`system
`is an experimental
`FotoFile
`organization and retrieval, based upon the design goal of
`making multimedia content accessible to non-expert users.
`Search and retrieval are done in terms that are natural to
`the
`task. The system blends human and automatic
`annotation methods.
`It extends textual search, browsing,
`and retrieval
`technologies
`to support multimedia data
`types.
`
`Keywords
`organization,
`information
`computing,
`Multimedia
`retrieval, browsing, visualization, content-based indexing
`and retrieval, digital photography, digital video, metadata,
`media objects
`
`INTRODUCTION
`Technologies and applications for consumer digital media
`are evolving rapidly. Examples of these technologies are
`digital still and video cameras, multimedia personal
`computers,
`broadband multimedia
`networks,
`and
`recordable CD/DVD.
`These
`technologies
`enable
`consumers to create and access ever-increasing amounts
`of content, from a wide variety of sources [l] and formats.
`As a result,
`there are significant
`challenges
`to be
`overcome to effectively organize and access this media
`information.
`Consumer research conducted by Hewlett Packard has
`found that organization and retrieval of digital images is a
`source of great frustration to customers. Consumers were
`found
`to be particularly
`resistant
`to
`the notion of
`organizing and managing home media, seeing
`these
`
`Permission to make digital or hard copies of all or part of this work for
`personal or classroom use is granted without
`fee provided
`that copies
`arc not made or distributed
`for profit or commcreial advantage and that
`copies bear this notice and the full citation on the first page. ‘l‘o copy
`otherwise,
`to republish, to post on servers or to redistribute
`to lists.
`requires prior specific permission and!or a fiie.
`CHI
`‘99 Pittsburgh PA USA
`Copyright ACM 1999 0-201-48559-1/99/05...$5.00
`
`496
`
`activities as tedious and error prone. They described
`photos thrown in shoeboxes and home videos sitting on
`shelves unviewed.
`We derived our approach to making multime,dia content
`accessible to non-experts by
`.
`analyzing
`the strengths and weaknesse.s of current
`commercial products and experimental systems, and
`conducting
`user
`research
`to
`understand
`the
`consumer’s perspective on the problem and to gauge
`customers’ reactions to the different approaches.
`
`.
`
`CURRENTAPPROACHES
`Technologies
`for multimedia organization and retrieval
`have been applied with some success to prolblems in the
`business/professional domain. It is not clear, h.owever, that
`these approaches and technologies are well-suited
`for
`consumer-oriented applications. Consumers, in general,
`have less time, patience, and motivation
`to learn new
`technologies.
`Traditional keyword-based search technologies are very
`powerful and flexible. There are a number of (commercial
`image management products that enable a user to search
`and retrieve visual information based upon ind:ices formed
`from the user’s annotations.
`Image database products
`from Extensis (Fetch) [2], Imscape (Kudd Image Browser)
`[3], Canto (Cumulus) [4], and Digital Now (Showcase) [5]
`allow a user to browse
`through
`files as galleries of
`thumbnails or as textual lists. The user can typically sort
`media objects by name, file type, folder, or volume.
`that
`The strength of
`the keyword-based approach
`is
`information about media objects can be expressed in
`terms that are personally meaningful
`to the user (i.e., in
`terms of attributes like creation date, location, subject, and
`identities of people). Such semantic information about
`media objects,
`frequently
`referred
`to as metudutu,
`provides a rich structure
`for effective searching. The
`
`Meta Platforms, Inc.
`Exhibit 1011
`Page 001
`
`

`

`CHI
`
`1999
`99 15-20
`MAY
`that making such metadata available
`disadvantage
`is
`usually means that keywords and textual annotations must
`be entered manually. This works for business applications,
`where there is an economic incentive
`for time and effort
`being devoted to indexing activities. Lacking these same
`economic incentives, consumers are more resistant to the
`task of data entry.
`indexing and
`An alternative approach, content-based
`retrieval, provides some degree of automation
`for this
`process by automatically extracting features, such as color
`or texture, directly
`from visual data [6]. Products from
`Virage [7] and IBM
`(QBIC)
`[8] implement mechanisms
`for content-based
`retrieval of
`images. By using
`the
`intrinsic
`visual attributes of
`images, such as color,
`structure, texture, and composition,
`to perform queries;
`users can search collections by instructing
`the system to
`retrieve
`images that are visually similar
`to the sample
`image.
`Images returned by the query are ordered by the
`degree of similarity
`to the base image.
`The content-based indexing and retrieval approach frees
`the user from the task of data entry, and it utilizes people’s
`perceptual abilities.
`These technologies work well
`in
`situations where a user wants to locate a visual image that
`is similar
`to a sample image. The disadvantage is that
`these systems only extract
`low-level syntactic
`features
`(measures of color and
`texture), which are not as
`personally meaningful
`to consumers as keyword-based
`attributes.
`An additional concern we had with current technological
`approaches was whether they correctly map to consumers’
`likely information-seeking behaviors. Much attention has
`been paid to the task of direct search, in which a user
`knows the target. Relatively
`little attention has been paid
`to
`the activities of browsing
`through collections of
`materials, where the user doesn’t have a very specific goal
`in mind, and serendipitous discovery is important [9]. It is
`likely
`that browsing will be a preferred
`information
`seeking behavior
`for
`consumers, and
`it
`should,
`accordingly,
`receive more systematic support
`from
`search/discovery technologies.
`
`USERRESEARCH
`the
`the consumer’s perspective on
`To understand
`multimedia organization and retrieval problem and to
`gauge customers’ reactions to the different approaches, we
`conducted a set of focus group sessions in the Denver and
`San Francisco areas. We were looking
`to more fully
`understand how people
`inherently
`organize
`visual
`materials and, in particular,
`to gather information on the
`perceived tradeoffs between
`. manual vs. automated annotation, and
`direct search vs. browsing.
`
`l
`
`Papers
`
`In order to understand the differences between business
`and consumer usage, we held different
`focus groups for
`business and home participants, respectively.
`the
`The sessions began with a discussion of how
`participants currently organize,
`find, and share photos.
`This was followed by a group exercise in organizing a set
`of travel photos. We then presented participants with
`mockups of concepts for keyword-based
`indexing and
`search, visually-based search (content-based indexing and
`retrieval), and visual overview (browsing).
`Our key findings from these sessions were that:
`.
`
`Keyword-based search was the easiest concept for
`home participants
`to grasp. However,
`they saw
`drawbacks, both
`in
`the
`time-intensive nature of
`entering keywords for photos and in the possibilities
`for many false “hits” while searching.
`Participants readily grasped the benefits of automated
`indexing. However,
`the home participants
`thought
`that
`they would use keyword-based search more
`frequently.
`favorably
`reacted very
`Home participants
`notion of browsing, much more favorably
`business participants.
`first, that
`We drew two conclusions from these findings;
`consumers would desire the benefits of both keyword-
`based search and automated indexing; second, that there
`may be a considerable role for browsing
`techniques in
`supporting consumers’ multimedia
`information seeking
`activities.
`
`.
`
`.
`
`the
`to
`than did
`
`THE FOTOFILE SYSTEM
`Based upon our analysis of current approaches and our
`findings
`from user research, we developed a hybrid
`approach
`to address
`the problems of multimedia
`organization and retrieval for consumers. We prototyped a
`number of techniques which make it easier for consumers
`to manually annotate content and to fit the annotation task
`more naturally
`into the flow of activities
`that consumers
`find enjoyable. We also utilized a number of automated
`content-based
`indexing
`techniques
`in order
`to both
`substitute for manual annotation where appropriate and to
`provide novel capabilities
`for content creation and
`organization. Finally, we augmented direct search tools
`with techniques for browsing and visualization of large
`digital media collections.
`for
`FotoFile,
`shown
`in Figure 1, is an application
`organizing and managing consumer digital media, such as
`It
`illustrates a
`photos and audio/video
`recordings.
`number of aspects of our hybrid approach.
`FotoFile displays multimedia
`in a photo-centric way by
`displaying media objects that consist of a photo with
`related sound and video attached. For video content,
`
`497
`
`Meta Platforms, Inc.
`Exhibit 1011
`Page 002
`
`

`

`Papers
`
`CHI 99 15-20
`
`MAY
`
`1999
`
`FotoFile generates photos by extracting keyframes from
`the video. The leftmost pane is a Content Index, which
`enables the user to annotate and search for materials. An a
`priori set of pre-defined metadata attributes is used to
`represent common properties of media objects, such as
`creation date,
`location,
`subject, people,
`title, and
`description. Users can assign arbitrary values within
`the
`defined metadata types, e.g. annotating the location of a
`photo as “Grand Canyon”. Another pre-defined metadata
`attribute, called favorite, can be used to tag certain images
`as the “best”
`images in a collection, e.g. my favorite
`photos from the Grand Canyon vacation.
`The central pane is an Image Palette, which provides
`functionality analogous to a light
`table. The user can
`arrange, delete, and display media objects at different
`resolutions in the Image Palette. The palette is also used
`to display search results and newly
`imported materials,
`and it also serves as a temporary storage area for creating
`albums.
`The rightmost pane is an Album Editor that provides tools
`for composition of digital albums, which can then be
`“played back” or sent electronically
`to others.
`In order to match the user’s expectations for how pictures
`are arranged, FotoFile uses a photo album as the primary
`organizational metaphor. A photo album is a metaphor
`with which people can quickly relate when thinking about
`organizing photos, and therefore the mental model relies
`on user
`intuition
`rather
`than explicit
`instruction.
`In
`FotoFile, an Album
`is a persistent collection of media
`objects, which are arranged on “pages”. Each image is
`also accompanied by annotations, which can be in the
`form of text, audio, or video. Furthermore,
`in order to
`simplify
`album
`retrieval,
`the user can assign a
`representative
`image
`for
`the album cover
`to aid
`in
`selection
`from a list. Having a cover
`image that
`is
`representative of the album in the user’s mind enables fast
`visual
`recognition,
`rather
`than relying on information
`recall.
`to Ease the Task of Manual Annotation
`Techniques
`Bulk Annotation
`for bulk annotation, which
`We provide mechanisms
`enable the user to quickly annotate large numbers of items
`For example, the
`with a minimal number of gestures.
`user can select multiple media objects in
`the Image
`Palette, select several values within
`the Content Index,
`and then press the Annotate button. This results in the
`assignment of all selected values to all of the selected
`media objects.
`
`Symmetry between Annotation and Search
`Since we were designing FotoFile
`for home usage, we
`designed the annotation and search interfaces to use the
`same basic mechanism. There is a visual and gestural
`symmetry between the actions for annotation and search.
`Users only need to learn one tool for both activities.
`To annotate content, the user selects one or more metadata
`attribute/value pairs, and presses the Annotate button. At
`that time, the selected attributes are applied to :a11 selected
`media objects. To retrieve content, the user again selects
`one or more attribute/value pairs, and presses the Search
`button. At
`that time, all media objects that have the
`selected attributes are immediately displayed in the Image
`Palette. There are several search modes,
`including
`Boolean operations and a similarity-based search built
`upon automated feature extraction [ 181.
`Since there is no default mode, the user is free ‘to intermix
`the annotation and search activities, which we believe will
`result in a better-annotated corpus of material than would
`occur if the user only had a dedicated authoring mode
`available.
`
`to Help Organize Content
`Use of Narrative Structure
`Annotating content manually
`is time consuming, and it
`transforms the process of creating photo albums from an
`enjoyable activity
`into a very tedious one. On the other
`hand, people like to tell stories with photos [IO] and the
`organization of photos into stories can provide us with a
`significant amount of
`information
`that can serve as
`is, we can use narrative
`structure
`metadata That
`underlying
`the events captured in photos as a source of
`their organization and annotation. This effectively
`turns
`the organization process into a storytelling activity, an
`activity
`that
`is more enjoyable
`than
`the
`task of
`organization, which carries with
`it
`the connotation of
`“work’.
`is
`Whereas with conventional photography, storytelling
`typically done using prepared albums and collages, whose
`structures are fixed, digital photography allows the user to
`collections
`of
`-photos
`in
`employ more dynamic
`storytelling. The user can arrange small groups of photos
`into segments that correspond to single narrative episodes.
`These segments can be reused in different situations and
`combined
`in different ways, depending upon
`the
`interaction between storyteller and audience. The model
`of usage is of two or more people sitting
`together by a
`computer, much in the same way that people sit together
`and go through photo albums. An alternative model of
`usage is one wherein the storyteller shares groupings of
`photos and annotations over the Internet.
`
`49%
`
`Meta Platforms, Inc.
`Exhibit 1011
`Page 003
`
`

`

`C H I 9 9 1 5 - 2 0 M A Y 1 9 9 9
`
`P a p e r s
`
`Building on the metaphor of a scrapbook, we call thesesmall groupings of photos scraplets (shown in Figure 2).”A scraplet can be assigned a name and other properties,thus providing annotation for a grouping that can beuseful in retrieving the grouping at a later time. Webelieve that such grouping and lightweight annotating willfit naturally within the activity of preparing a story, thusproviding a more enjoyable mechanism for elicitingmetadata from consumers.Moreover, use of voiceannotation may bring additional emotional power tostories that are shared over the Internet.The selection of photos for grouping into scraplets isbased upon two assumptions.
`
`First,
`the user should have apersonal memory of the events depicted in the photos.
`Second,
`
`chronological ordering of events is a dominantorganization principle of human episodic memory [11].Using the same photos in multiple scraplets links themimplicitly. The links are displayed during album playbackto indicate to the user multiple possible story lines.
`
`Benefits of Automated Feature Extraction
`
`FotoFile
`
`to generate some of the annotation that wouldotherwise have to be manually entered.It also providesnovel capabilities for content creation and organization.
`
`Face Recognition
`
`The black rectangular highlights on the pictures of Davidin Figure 1 denote faces that have been recognized by aface detection and recognition system [12] [13].Information about recognized faces appears in the
`
`Content
`Index
`
`in an identical manner to metadata gathered byhuman annotation. This is one example of the integrationof automated and human annotation in our approach, andit results in a hybrid system where the user guides themechanisms.When given photos that contain faces of new people, theface recognition system attempts to match the identity ofthe face (see Figure 3). The user either corrects orconfirms the choice; the system then can more accurately
`
`Meta Platforms, Inc.
`Exhibit 1011
`Page 004
`
`The use of automated feature extraction tools enables
`

`

`match faces to their correct identities in subsequentphotos. Once a face is matched to a name, that name willbe assigned as an annotation to all subsequently seenphotos that contain faces that match the original. Tohandle the false positives and false negatives of the facerecognition system, a user must confirm face matches (seeFigure 4) before the annotations associated with thesefaces are validated (i.e., added to the
`
`Users view the matched identities of faces through tooltipsdisplayed when the mouse sprite enters the: rectangularhighlight surrounding a face.
`
`Content Index).
`
`999
`
`Meta Platforms, Inc.
`Exhibit 1011
`Page 005
`
`

`

`C H I 9 9 15-20 MAY 1999
`
`P a p e r s
`
`Video Shot Detection
`FotoFile
`
`FotoFile
`
`Video Keyframe Extraction
`
`During the shot detection process, a keyframe extractionalgorithm [15] is used to generate a set of video frames(still images) which best represent the content of each shot.These keyframes attempt to represent abrupt changes invideo content as well as slower, ordered changes like pansand zooms. Each resulting keyframe is associated with avideo clip that starts with that frame and continues to theend of the shot. The set of these keyframes imposes an extrastructure on shots which help users fine-tune their selectionand manipulation of video clips and shots.Keyframe extraction is also used to derive a representativepicture for each video imported into
`user can automatically generate “albums” ofvideo clips extracted from longer video sequences using thevideo shot detection and keyframe extraction algorithms[14][15]. Video shot detection is the process of detectingboundaries between consecutive shots so that sequences ofinterrelated video frames can be grouped together.Examples of shot boundaries include abrupt shot changescaused by turning the camera off, as well as moresophisticated shot transitions like fades, dissolves, andwipes. A user can easily create an album that contains aseries of video clips that comprise a video (see Figure 5).Each clip represents a playable segment of video. Sinceeach video segment is itself a media object, it can berearranged, or placed in different albums-just like any
`
`FotoFile.
`
`Browsing and Visualization of the Content Space
`
`We believe that consumers’ information-seeking activitiesdiffer from those of information retrieval professionals,and that this is particularly true when the informationinvolves home media such as photos or videos.In these settings, directed search may be less frequent,whereas riffling and browsing through collections ofmaterials becomes the norm (and serendipity is expected).We provide support for these activities by integratingvisualization and browsing tools into
`
`FotoFile,
`Hyperbolic Tree
`
`package from Inxight Software [16][17].Figure 6 shows a hyperbolic tree built from the attributesand values in the
`
`Content Index.
`
`Hyperbolic Tree
`
`[21] was that items on the outside rim ofthe display tended to group strongly, with users oftenassuming that they belonged in the same category. It wassuggested that careful use of alternative perceptual codingfor semantic categories could alleviate this problem. Wehave achieved this by providing additional views based on
`
`Meta Platforms, Inc.
`Exhibit 1011
`Page 006
`
`The
`media object.
`such as the
`One problem observed in usability studies of the
`

`

`Papers
`
`Hyperbolic Tree.
`
`the use of automated image feature-extraction software[ 181.Image content is analyzed to extract measures forcolor distribution and texture, and a clustering algorithm[19] recursively partitions the collection of media objects toform the tree model displayed by the
`Inthis way, media objects that are visually similar to eachother will appear closer to each other in the visualizationspace.This adds structure to the browsing activity,enabling the user to visualize related clusters of materials inan intuitive manner.
`
`DISCUSSION
`FotoFile,
`
`we have attempted to balance tradeoffsacross two dimensions of information-seeking behaviors:Based upon our findings from user research, we haveattempted to integrate these capabilities in a way that issuited to the needs of the consumer environment. In orderto provide an integration that is easily understandable andusable, we need to emphasize certain capabilities more thanothers.To determine the appropriate balance points,additional user research is needed. In particular, we need todetermine:
`
`502
`
`The degree to which consumers wil!l performannotation if the benefits are significant andmeaningful.
`
`l
`
`l
`
`The usability and usefulness to consumers of browsingand visualization environments.One challenge in designing credible studies of this nature isin defining the right metrics for data analysis. Consumerinformation-seeking behavior is different from that ofspecialists performing directed searches in textualdatabases, where large numbers of people are searchingover large information spaces for materials indexed bysome unknown person.The characteristics of aninformation-seeking environment for consumers involverelatively few people searching (e.g., immediate familymembers) over a small amount of information (less thanseveral thousand items in a collection) that they havepersonally indexed, or that was indexed by someone theyknow.In many cases, serendipitous discovery is asignificant (but often unstated) goal. The traditional metricsof
`
`precision
`
`recall
`
`may not be as applicable.Alternative measures might include the level of goalattainment, the efficiency (number of actions) to reach agoal, the utility of the information found, which annotationsand features are used for later retrieval by both novices and
`
`Meta Platforms, Inc.
`Exhibit 1011
`Page 007
`
`With
`and
`

`

`CHI 99 15-20
`
`MAY
`
`1999
`
`-
`
`Papers
`
`CONCLUSION
`We have built an experimental multimedia organization and
`retrieval system that attempts to balance tradeoffs between
`(1) human annotation versus automated feature extraction,
`and (2) directed search versus exploratory browsing and
`visualization.
`The ultimate goal is to make multimedia
`content accessible to non-expert users.
`that address
`Photography and home movies are activities
`deep human needs; the need for creative expression, the
`need to preserve memories, the need to build personal
`relationships with others. Digital photography and digital
`video can provide powerful and novel ways for people to
`express, preserve, and connect. However, new technologies
`often raise new problems;
`the problem of multimedia
`organization and retrieval
`is brought about by the very
`technology that makes it possible for people to create and
`access ever-increasing amounts of content, from a widening
`diversity of sources.
`By helping consumers to better manage content, we hope to
`enable people
`to
`take full advantage of
`the benefits
`provided by digital media technologies.
`
`ACKNOWLEDGMENTS
`We owe a great debt to HongJiang Zhang, John Wang, and
`Wei-Ying Ma for their excellent work
`in content-based
`indexing and retrieval, which has been incorporated
`into
`our prototypes. Thanks and praise to Rick Steffens, Mike
`Krause, and others at HP’s Colorado Memory Systems
`Division
`for their support, encouragement, and inspiration.
`Ella Tallyn made substantial contributions
`to both the visual
`design of FotoFile and the conceptual design of our use of
`narrative structures in FotoFile.
`
`2.
`
`3.
`
`REFERENCES
`1.
`is Not Enough: Content
`Kuchinsky, A., Bit Velocity
`and Service
`Issues
`for Broadband Residential
`Information
`Services,
`IEEE
`3rd
`International
`Workshop on Community Networking,
`Antwerp,
`Belgium, May, 1996.
`Extensis Corporation, httn://www.extensis.cornl.
`Imspace Systems Corporation, httu://imspace.com/.
`Canto Software, http:Nwww.canto-software.coml.
`Digital Now, httv://www.diaitalnow.com/.
`Furht, B. Smoliar, S., Zhang, H., and Furht, B. Video
`and Image Processing in Multimedia Systems, Kluwer
`Academic Publishers, 1995. Conger., S., and Loch,
`K.D. (eds.).
`
`4.
`
`5.
`
`6.
`
`7.
`
`8.
`
`9.
`
`10.
`
`11.
`
`12.
`
`13.
`
`14.
`
`1.5
`
`16,
`
`17.
`18.
`
`19.
`
`20.
`
`21.
`
`Virage Incorporated, httn://www.virage.com/.
`IBM, httn://www-i.almaden.ibm.com/cs/showtell/obic/.
`a
`Chang, S.J.,
`and Rice, R.E.,
`Browsing:
`Multidimensional Framework, in Williams, M.E. (ed),
`Annual
`Review
`Information
`Science
`and
`of
`Technology, Vol. 28, pp. 231-276, Medford, NJ, 1993..
`Chalfen, R. Snapshot Versions of Life. Bowling Green
`State University Press, Bowling Green, Ohio, 1987.
`Tulving, E. Elements of Episodic Memory. Oxford,
`UK: Oxford University Press, 1983.
`Turk, M., and Pentland, A. Eigenfaces for Recognition.
`Journal of Cognitive Neuroscience, Vol. 3, No. 1, pp.
`71-86, 1991.
`H.A. Rowley, S. Baluja and T. Kanade. Neural
`network-based face detection. IEEE Trans. on Pattern
`Analysis and Machine Intelligence, vol. 20, no. 1, pp.
`23-38, Jan. 1998.
`H.J. Zhang, C. Y. Low and S. W. Smoliar. Video
`parsing
`and browsing
`using
`compressed data.
`Multimedia Tools and Applications, vol. 1, pp. 89- 111,
`1995.
`H.J. Zhang, et al. An integrated system for content-
`based
`video
`retrieval
`and
`browsing.
`Pattern
`Recognition, Pergomon Press/Pattern Recognition
`Society, May 1997.
`John Lamping, Ramana Rao, and Peter Pirolli. A
`focus+context technique based on hyperbolic geometry
`for visualizing
`large hierarchies. In Proceedings of the
`ACM SIGCHI Conference on Human Factors
`in
`Computing Systems (May I995), ACM.
`Inxight Software, Inc., httn://www.inxiaht.com.
`image
`W.Y. Ma and H.J. Zhang. Content-based
`indexing and retrieval. Chapter 13, The Handbook of
`Multimedia Computing, edited by Borko Furht, CRC
`Press LLC, 1998.
`R. Duda and P. Hart Pattern Classification and Scene
`Analysis. Wiley Publications: NY, 1973.
`Exploration
`Wilson,
`K. Evaluating
`Information
`Interfaces, position paper for Workshop on Innovation
`in
`Information Exploration Environments, CHI’98
`Conference on Human Factors in Computing Systems,
`http://www.fxoal.com/CHI98IE/.
`Czerwinski, M. and Larson, K., Trends in Future Web
`Designs: What’s Next for the HCI Professional?, ACM
`Interactions, November-December, 1998, pp. 9-14.
`
`503
`
`Meta Platforms, Inc.
`Exhibit 1011
`Page 008
`
`