This document discusses the future of visual search and image recognition technologies. It begins with a brief history of visual search and outlines key areas like object, scene and logo identification. Google and Pinterest are highlighted as leaders in allowing visual searches of images. The document also explores emerging technologies like visual search capabilities from Slyce, Blippar and Houzz. It concludes by discussing potential applications of visual analytics and image data for audiences, brands and businesses.
Digital Marketing in 5G Era - Digital Transformation in 5G Age
A picture's worth visual search: How image recognition will power analytics
1. A picture’s worth a
thousand hashtags:
How image recognition
will power the future of
analytics
David Berkowitz
Chief Strategy Officer
Sysomos
dberkowitz@sysomos.com
www.sysomos.com
@sysomos / @dberkowitz
2. About this presentation
This talk was presented to Marketing Week Live in London in March 2017. A
Texan version of this was delivered to W2O Group’s Pre-Commerce Summit
during SXSW that month. If you prefer Frito pie to bangers and mash, I will
gladly send you the W2O edition.
Sources for information shown in slides are presented as links in the bottom-
left corner. Further details about sources, where applicable, are also
included in the notes field when downloading this presentation.
To share your feedback or discuss this further, please contact me at
dberkowitz@sysomos.com.
Thank you.
David
3. And now for something completely standard: an agenda
• A brief history of nearly everything visual search
• Why visual search matters
• How Google, Pinterest, and others are deploying it
• How marketers can use it
• Numerous gratuitous references to all things British
8. Mona Lisa #Hashtags
• #MonaLisa
• #art
• #painting
• #woman
• #lady
• #smile
• #smug
• #Italian
• #DaVinci
• #epic
• #outdoors
• #masterpiece
• #sky
• #France
• #LaGioconda
• #badhairday
• #beautiful
• #Louvre
• #Renaissance
• #portrait
This is a futile exercise. One can’t simply capture the Mona
Lisa in hashtags. It points to the need for better ways to
identify and analyze visual content. Text and hashtags alone
don’t cut it.
10. Drop in a bucket visual
The
Visual
Data
GapRecall how pictures are worth a thousand words? There is
so much more data in images (‘the sun’) than there is in text
(represented metaphorically by the planets).
13. A challenge of Shakespearean proportions:
“His reasons are as two grains of wheat hid in
two bushels of chaff: you shall seek all day ere
you find them, and when you have them, they
are not worth the search.”
-Bassanio, The Merchant of Venice
14. Number of object categories out there
15,000There are
15,000object categories
Source: IEEE
15. Source: Computer Vision by Richard Szeliski
For fun, I included a few
examples of early
attempts at machine-
powered object
recognition.
17. An inflection point waiting to happen
Jason Goldberg, Razorfish:
“I’m strongly bullish on visual
search. It solves a real problem
consumers have… In the not-
too-distant future, it’ll become a
heavily used mainstream
feature. I think the inflection
point is at least a year away, but
not two years." We’re approaching the
inflection point, but it has
taken longer than expected.
This report is from
November 2014.
19. This is simply an
enlarged, cropped
version of the
highlights from the
previous chart.
20. You can’t always get what you want (with text search)
• 74% of consumers say text-based keyword
searches are inefficient for helping them find
the right products online
• 67% of consumers say quality of product
images is very important in selecting and
purchasing products
• 90% of information transmitted to the brain is
visual
• Visual information is processed 60,000x
faster than text
Source: Slyce
22. The dress that inspired Google Image Search in 2001
“…People wanted more than just text. This first became
apparent after the 2000 Grammy Awards, where Jennifer
Lopez wore a green dress that, well, caught the world’s
attention. At the time, it was the most popular search
query we had ever seen. But we had no surefire way of
getting users exactly what they wanted: JLo wearing that
dress. Google Image Search was born.”
-Eric Schmidt, Executive Chairman, Google
Source: Project Syndicate
23. Solving the Clarissa problem My wife gave me this reference. As a kid, she always
wanted to identify and shop for whatever Clarissa wore.
27. Layers of image recognition
A Deep
Learning
algorithm is
presented with
the images
made up of
simple pixels.
The algorithm
discovers simple
regularities that
are present
across many/all
images, like
curves & lines.
The algorithm
discovers
how these
regularities are
related to form
higher-level
concepts
The system gains
a high-level
understanding
of the original
image… all
automatically
Source: GazeMetrix
28. A framework for visual search
Scene
Identification Intelligence
Object
Identification Intelligence
Logo
Identification Intelligence
Image
Identification Intelligence
Category
Identification Intelligence
This notes some of the most important processes within
visual search. Also note that identification and intelligence
are two separate approaches. Examples follow.
29. What follows is an
example using a real
photo from Agnes, a
Chinese tourist to the UK.
30. Here’s her photo. In each subsequent
slide, you can see how the framework
plays out and a sample finding that can
be derived. Note the intelligence
examples that follow are for illustrative
purposes only; feel free to cite the
framework, but not the data itself.
32. Logo Identification:
There is a Fanta logo, and
the text in the top-right says
Starbucks
Logo Intelligence:
Fanta logos are rarely paired
with Starbucks; Fanta logos
are most often seen with
Coca-Cola and Adidas
33. Object Identification:
This looks like fish and chips
with a can of Fanta Lemon
Object Intelligence:
Fanta Lemon is the fourth
most popular soda when
paired with fish & chips
34. Image Identification:
This is the same photo that
appears on Agnes_Cin’s
Flickr and public Facebook
pages
Image Intelligence:
This image hasn’t been
shared in any media outlets
and hasn’t been shared
publicly
35. Scene Identification:
This photo seems to be taken
outdoors during the day
Scene Intelligence:
94% of photos at the British
Museum are shot indoors,
compared to 87% of museum
photos worldwide
43. Pinterest: one of the world’s biggest search engines
• 150 million monthly users
• 75 billion pins
• 2 billion searches/month
• 97% of searches are unbranded
Source: Pinterest
44. Pinterest: search pins from real-world images
Source: Pinterest
Examples from Pinterest’s new Visual
Discovery follow. In the downloadable
version of this talk, the next few visuals
play as GIFs, and you can read more at
the source below.
53. Toys R Us offers Slyce image detection for its catalog
Source: Slyce
54. Visual search to complement textual search
eMarketer: Do you think [visual search]
will replace some types of searches, or
do you think it will augment existing
searches?
Gierhart: It will probably augment. It’s
adding a new utility to what was there
before... There will still be contexts for
both.
Source: eMarketer (see a related video on YouTube)
57. TheTake uses AI to identify products, locations in video
Source: TheTake
58. What’s Possible
with
Image Analytics
The images that follow are sample
reports drawn from Sysomos. The data is
again for illustrative purposes. Reach out
if you want to dive deeper into any of
this.
59. Visual analysis: understanding visual characteristics
Logo recognition
Object recognition
Scene recognition
Food recognition
Color detection
OCR: Search text within
images
60. Visual analysis: understanding visual characteristics
Logo recognition
Object recognition
Scene recognition
Food recognition
Color detection
OCR: Search text within
images
61. Visual analysis: understanding visual characteristics
Logo recognition
Object recognition
Scene recognition
Food recognition
Color detection
OCR: Search text within
images
67. So long,
and thanks
for all the fish.
Let’s take tea!
David Berkowitz
Chief Strategy Officer
Sysomos
dberkowitz@sysomos.com
www.sysomos.com
@sysomos / @dberkowitz
Editor's Notes
A picture’s worth a thousand hashtags: How image recognition will power the future of analytics
• Understand how artificial intelligence, the surge of visual content creation, and other trends lead to visual search• Discover what you can find by searching images that doesn’t show up in text search• Explore the kinds of insights that you can unearth from those visual search results• See how visual search can then fit in with your overall marketing plan, from content marketing to customer service
The amount of data available from visuals versus text is like comparing the Earth to the sun
A picture’s worth a thousand hashtags: How image recognition will power the future of analytics
• Understand how artificial intelligence, the surge of visual content creation, and other trends lead to visual search• Discover what you can find by searching images that doesn’t show up in text search• Explore the kinds of insights that you can unearth from those visual search results• See how visual search can then fit in with your overall marketing plan, from content marketing to customer service
And now, some history…
https://en.wikipedia.org/wiki/Bill_Bryson
The challenge with image search – courtesy of William shakespeare
http://www.rhymezone.com/r/gwic.cgi?Path=shakespeare/comedies/merchantofvenice/i_i//&Word=have+them,+they+are+not+worth+the+search.#w
His reasons are as two grains of wheat hid in two bushels of chaff: you shall seek all day ere you find them, and when you have them, they are not worth the search.
http://www.rhymezone.com/r/gwic.cgi?Path=shakespeare/comedies/merchantofvenice/i_i//&Word=have+them,+they+are+not+worth+the+search.#w
Image: https://www.biography.com/.image/c_fill,cs_srgb,dpr_1.0,g_face,h_300,q_80,w_300/MTE1ODA0OTcxNzgzMzkwNzMz/william-shakespeare-194895-1-402.jpg
http://ieeexplore.ieee.org/document/7298599/
Computer Vision: Algorithms and Applications
Richard Szeliski
Springer Science & Business Media, Sep 30, 2010 - Computers - 812 pages
https://books.google.com/books?id=bXzAlkODwa8C
https://books.google.com/books?id=bXzAlkODwa8C&printsec=frontcover&source=gbs_ge_summary_r&cad=0#v=onepage&q&f=true
A lot goes into this whole image recognition thing
Computer Vision: Algorithms and Applications
Richard Szeliski
Springer Science & Business Media, Sep 30, 2010 - Computers - 812 pages
https://books.google.com/books?id=bXzAlkODwa8C
http://blog.slyce.it/wp-content/uploads/2015/11/Visual_Search_Technology_and_Market.pdf
Note Rolling Stones reference (British)
Source: Slyce, summer 2015
After last bullet: shouldn’t this whole talk be in emoji?
http://www.vogue.co.uk/article/j-lo-green-versace-dress-responsible-for-google-image-search
“In an interview charting the evolution of the internet-service provider, Schmidt revealed that following the singer's appearance on the red carpet at the 2000 Grammy Awards wearing that diaphanous, slashed-to-the-navel Versace dress, the relatively new Google search engine was inundated with people wanting to see it - except that they couldn't.”
----
So our co-founders Larry Page and Sergey Brin – like all other successful inventors – kept iterating. They started with images. After all, people wanted more than just text. This first became apparent after the 2000 Grammy Awards, where Jennifer Lopez wore a green dress that, well, caught the world’s attention. At the time, it was the most popular search query we had ever seen. But we had no surefire way of getting users exactly what they wanted: JLo wearing that dress. Google Image Search was born.
http://www.vogue.co.uk/article/j-lo-green-versace-dress-responsible-for-google-image-search
“In an interview charting the evolution of the internet-service provider, Schmidt revealed that following the singer's appearance on the red carpet at the 2000 Grammy Awards wearing that diaphanous, slashed-to-the-navel Versace dress, the relatively new Google search engine was inundated with people wanting to see it - except that they couldn't.”
----
So our co-founders Larry Page and Sergey Brin – like all other successful inventors – kept iterating. They started with images. After all, people wanted more than just text. This first became apparent after the 2000 Grammy Awards, where Jennifer Lopez wore a green dress that, well, caught the world’s attention. At the time, it was the most popular search query we had ever seen. But we had no surefire way of getting users exactly what they wanted: JLo wearing that dress. Google Image Search was born.
It’s an exciting time – Facebook is literally giving its technology away
http://www.theverge.com/2016/8/25/12630850/facebook-fair-deepmask-sharpmask-ai-image-recognition
A picture’s worth a thousand hashtags: How image recognition will power the future of analytics
• Understand how artificial intelligence, the surge of visual content creation, and other trends lead to visual search• Discover what you can find by searching images that doesn’t show up in text search• Explore the kinds of insights that you can unearth from those visual search results• See how visual search can then fit in with your overall marketing plan, from content marketing to customer service
https://www.gazemetrix.com/
Visual Search Framework
Category identification: that is a green dress
Image identification: that is a photo of Jennifer Lopez wearing a green dress from the 2000 Grammy Awards
Object identification: that is a Versace green dress, and here it is offered by various retailers and shown on people other than J-Lo
Category intelligence: green is the fifth most popular color of dresses this year
Image identification: that image was shared by 29,400 people and media outlets
Object identification:
Object intelligence
Scene identification: those are trees and mountains and it’s daytime and there are seven people all in this photo
Scene intelligence: 29% of photos of that green dress are shown in daytime, and 8% are paired with luxury handbag brands
Logo can apply to text as well
A picture’s worth a thousand hashtags: How image recognition will power the future of analytics
• Understand how artificial intelligence, the surge of visual content creation, and other trends lead to visual search• Discover what you can find by searching images that doesn’t show up in text search• Explore the kinds of insights that you can unearth from those visual search results• See how visual search can then fit in with your overall marketing plan, from content marketing to customer service
A picture’s worth a thousand hashtags: How image recognition will power the future of analytics
• Understand how artificial intelligence, the surge of visual content creation, and other trends lead to visual search• Discover what you can find by searching images that doesn’t show up in text search• Explore the kinds of insights that you can unearth from those visual search results• See how visual search can then fit in with your overall marketing plan, from content marketing to customer service
When you’re ready to buy, you can check out right on Pinterest, or get things directly from your favorite brands. CB2, Macy's, Neiman Marcus, Target, Wayfair and other brands will all be offering shoppable looks, thanks to Pinterest’s new partnerships with Curalate, Olapic, Project September, Refinery29 and ShopStyle.
Bangers and Mash – Pinterest Visual Discovery
Prince George Pinterest Visual Search
Prince George Pinterest Visual Search
Big Ben Pinterest Visual Search
Big Ben Pinterest Visual Search
A picture’s worth a thousand hashtags: How image recognition will power the future of analytics
• Understand how artificial intelligence, the surge of visual content creation, and other trends lead to visual search• Discover what you can find by searching images that doesn’t show up in text search• Explore the kinds of insights that you can unearth from those visual search results• See how visual search can then fit in with your overall marketing plan, from content marketing to customer service
eMarketer: Do you think [visual search] will replace some types of searches, or do you think it will augment existing searches?
Gierhart: It will probably augment. It’s adding a new utility to what was there before. There are probably several situations where you still are not going to be able to find that thing you’re thinking of, so you’re going to want to search. There will still be contexts for both.
eMarketer: What prompted you to decide to pull the trigger on visual search now?
Gierhart: We wanted to be first to market. I think visual search is the next thing that everyone’s going to start rolling out in the next year. So we just wanted to get out there and start trying it with our customers so we could figure out how to improve it.
- See more at: https://www.emarketer.com/Article/CMO-One-to-One-Neiman-Marcus-Readies-Visual-Search-Shoes-Handbags/1011580#sthash.ym76vUpl.dpuf
https://blippar.com/en/showroom/#planet-earth-ii
Get closer to the natural world with Blippar and Planet Earth. Scan one of seventeen everyday objects to reveal one of the incredible animals featured in the series. See the creature up close and learn some surprising facts. Try it yourself now by scanning your hand, a rose or a fork. Selected regions only.
www.blippar.com
Houzz Visual Match
http://blog.houzz.com/post/150412447953/houzz-introduces-visual-match-making-it-even
eMarketer interview with Liza Hausman
http://totalaccess.emarketer.com/Interview.aspx?R=6002271&dsNav=Ntk:relevance%7cliza+hausman%7c1%7c,Ro:0,Nr:NOT(Type%3aComparative+Estimate)
https://www.emarketer.com/Interview/Use-of-Augmented-Reality-Image-Search-Boost-Likelihood-of-Online-Home-Product-Purchases/6002271
eMarketer: You recently introduced another tool called Visual Match. What is it?
Hausman: Think about Visual Match as an automatic search. For any image you see on Houzz, it’s automatically looking for matches and will start recommending similar photos.
eMarketer: What has been the reaction to Visual Match?
Hausman: It’s been phenomenal. One of the early popular features on the Houzz app was these green tags that professionals and brands could add to their photos, to identify the products and materials in the photos. Houzz users loved the ability to not only see a photo and get excited about it, but also find out what’s in the photo and where to get it.
A lot of our development lately is focused on making that easier. Both View in My Room and Visual Match came out of this desire—they help people not just discover what they love, but take the next step and transact.
https://techcrunch.com/2017/02/13/working-with-major-studios-thetake-launches-ai-image-recognition-engine-for-businesses/amp/
Focus is now a B2B model to partner with studios, networks
https://thetake.com/products?mId=2791&mName=Fifty%20Shades%20Darker&mImage=https%3A%2F%2Fimg.thetake.com%2Fmovie_images%2F9c15cd823da4514f36ca1ac80cbdcb86df39d8b52ca16d813ffb2d02061ea8fc.jpeg
A picture’s worth a thousand hashtags: How image recognition will power the future of analytics
• Understand how artificial intelligence, the surge of visual content creation, and other trends lead to visual search• Discover what you can find by searching images that doesn’t show up in text search• Explore the kinds of insights that you can unearth from those visual search results• See how visual search can then fit in with your overall marketing plan, from content marketing to customer service
Multi select of Objects / Scenes/ Food etc.
Understanding visual characteristics
TASK: Find me all photos with “Coca Cola logo” taken at a beach with a ball OR balloon in the picture
Task: Find me all photos with “Coca Cola logo” taken at a beach with a ball OR balloon in the picture
Task: How often do people drink Coca Cola with “Burgers” vs. “Pizza” in Europe
Task: Most popular ‘colors’ of Coke ‘cans’ at different ‘scenes’
Task: Upload a list of user segments and understand their conversations, preferences and behavior patterns
Task: Identify partnership opportunities – co-branding and join promotion
Task: More intuitive and granular reporting along with geographic and behavior insights.
Creative optimization
Influencer marketing
Rights management
Leak prevention
Creative – Buster of John Lewis: https://s3-eu-central-1.amazonaws.com/centaur-wp/creativereview/prod/content/uploads/2016/11/Buster-The-Boxer-John-Lewis-Christmas-Ad-07.jpg
Influencer: Emma Willis of The Voice, Big brother
Rights management – Saul Goodman – Better Call Saul - http://static.tvtome.com/images/genie_images/news_hub/uploaded/TimSpotnews137893011780/breakingbad_saul.jpg
Help image: http://images6.fanpop.com/image/photos/36600000/The-Beatles-image-the-beatles-36630507-500-422.jpg
Partnership ideation image: Sherlock / Watson - http://i.dailymail.co.uk/i/pix/2014/07/28/article-2708469-0C0AAB1D000005DC-912_634x395.jpg
Competitive intelligence: Jessica Ennis-Hill, Olympian - https://static.independent.co.uk/s3fs-public/styles/story_medium/public/thumbnails/image/2015/12/05/09/Jessica-Ennis-Hill.jpg
Customer service – British Airways flight attendant - http://i4.getwestlondon.co.uk/incoming/article9033228.ece/ALTERNATES/s615/JS60931715.jpg
Summary slide (didn’t use)
Technology progress around image recognition is moving faster than consumer behavior is shifting
There are many interlocking components to visual search, and progress is happening everywhere
We mammals are better adapted to visual than textual search
There will always be a gap between what’s shared textually and what’s shared visually
The J-Lo/Clarissa problems have been practically solved; marketers must prioritize consumer education
There are many applications for visual search data: research, customer service, influencer marketing, creative optimization crisis management, and many more
A picture’s worth a thousand hashtags: How image recognition will power the future of analytics
• Understand how artificial intelligence, the surge of visual content creation, and other trends lead to visual search• Discover what you can find by searching images that doesn’t show up in text search• Explore the kinds of insights that you can unearth from those visual search results• See how visual search can then fit in with your overall marketing plan, from content marketing to customer service