Feng-GUI attention heatmap accuracy

We’ve been asked several times about the accuracy of Feng-GUI attention heatmap and why is it not reaching 100%.
Feng-GUI attention heatmap algorithm, which we call the ViewFinder , is a composition of algorithms aggregating over 30 years of research. Research from the fields of neuro-science studies, visual attention, statistics of eye-tracking sessions, perception and cognition of humans.
We improve the ViewFinder algorithm in order to achieve the highest accuracy, but is it possible to get to 100% ?

Let us answer by asking a question:
How accurate is the "Magic Wand" in Photoshop? (also referred as the “Fuzzy select” in Gimp)



The Magic Wand tool is designed to select areas of the image by finding edges of image segments.



Part of the answer to accuracy can be found at the "image segmentation and boundary detection" benchmark conducted by Berkley.
It’s a 1000 photos test suite for algorithms that try to find the boundaries of elements inside pictures.
In the test suite results you can see how real people and computer algorithms, marked the boundaries of segments inside photos and how accurate they are.
The current score is:
Real people: 79%
Machines algorithms: 70%


original image

edge detection algorithm

edge detection by human

http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/segbench/
results:
http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/segbench/bench/html/algorithms.html

You can see that edge detection algorithms are not 100% accurate and will never be considering the hardware used be personal computers.
To solve this limitation, GIMP supports six different selection tools. In order of appearance in the tool box they are: rectangle, ellipse, free, fuzzy, bezier, and intelligent.
Each tool has its advantages and disadvantages--no one tool is clearly superior to the others. In general, a combination of tools proves to be the optimal strategy for complex selection tasks.

We conducted a comparison benchmark between Feng-GUI results and eye-tracking results of 40 people, 90 images, 15 seconds per image.
http://www.feng-gui.com/faq.htm#quality
Comparing the regions of interests (ROI) in the images, results with more than 70% of similarities.
If you compare Feng-GUI results to random groups of 10 people (out of the 40 people), for 5 seconds, the similarities go over 80%.

To enable designers and advertisers integrate Feng-GUI’s attention heatmap into their work process, we keep on improving the ViewFinder algorithm while maintaining two important requirements: Accuracy as well as Speed.

happy new year!

Google patents media layout analysis

Google just published two patent applications. US2008/0107337 is “Methods and Systems for Analyzing Data in Media Material Having Layout” and US2008/0107338 is “Media Material Analysis of Continuing Article Portions”. You can view them at USPTO.

Both inventions, to which Google is the assignee, pertain to figuring out what’s important and what’s not on Web pages. Companies that scan hard copy and convert those images to machine-readable ASCII use some tricks but a great deal of brute force to figure out what’s information and what’s advertising or other dross.

thanks to Stephen Arnold for the information.

Contextual vs Behavior Targeting

An Eye Tracking study designed and analyzed by Next Century Media was carried out by PreTesting and TACODA.
That study shows that Behavioral Targeting averages 17% more Ad Looks than Contextual Targeting, and this average increases to 54% more Ad Looks after the first exposure.
In other words, as frequency builds, the same ad increasingly gets filtered out in the Contextual environment, but apparently as a result of surprise, the level of Ad Looking stays up in Behavioral.

The first Eye Tracking study ever conducted comparing Behavioral Targeting to Contextual Targeting finds that, on average, the same ads receive +17% more Looks when seen in unrelated-content sites, than when seen in sites specializing in the sale of those types of products.

After the first exposure, this +17% advantage for Behavioral over Contextual Targeting increases dramatically to +54%.

The fact that Behavioral Targeting, unlike Contextual Targeting, tends to build with frequency, suggests that there is a Surprise effect and that it does not wear out, and because of this repeat Surprise effect, the ad is given more chances to engage the user.

you can view the full report at
http://www.tacoda.com/success/research.php

TACODA conducted primary research to identify the demographics and behaviors of online 'heavy clickers.'
Who are the "Natural Born Clickers" and what is "the Invisible Hand" ?

The invisible hand is a metaphor coined by the economist Adam Smith to illustrate how those who follow their individual self-interests inadvertently stimulate the economy and assist society as a whole. There has been an invisible hand (or set of hands) that through self-interest have guided the development of online advertising, stimulated it through their innate curiosity and sometimes impulsive natures. This is the tale of a group of people who have had a profound impact on our lives as media planners and buyers. We've never known who they are or where they come from. We just knew that they were there, invisible, hiding within the numbers that make up our campaign report

you can view the video presentation and download the report at
http://www.tacoda.com/success/research.php

GuitarHeroNoid at LeWeb3 France

Kathy brooks from Six Apart and Loic Le Meur invited the GuitarHeroNoid to perform at LeWeb3 conference France.
Our man Tal Chalozin, the heronoid Puppeteer, went there and presented the garageGeeks activities.


Here is a video of the show

GuitarHeroNoid v2 at VON 2007 Boston

Mr Jeff Pulver invited the human-size robot playing guitar hero aka the GuitarHeroNoid to preform in VON Boston 2007 (Video Over the Net) conference in Boston.

Tal Chalozin and Yuval Tal have improved the robot version 2, adding some more features like the Controller-Controller and the Penis Guitar Holder.



Pictures from VON

Tangible Kabuki at GeekCon 2007

This was Feng-GUI GeekCon2007 project. Thanks to:
Project members: Rafael Mizrahi, Dani Vardi, Tal Yaniv and Eyal "Person" Shachar.
Performers: Dror Gill, Ayelet Yagil, Zvi Devir and Jeff Pulver.
Photographers: Yaniv Golan and Alex Sirota.

1. The TangibLaptop is a collaborative electronic music instrument with a laptop tangible multi-touch interface built by the GarageGeeks (Rafael Mizrahi, Ohad Pressman and Eyal "Person"). inspired by the reactable http://en.wikipedia.org/wiki/ReacTable



2. The GeekCon un-convention get-together is a creative gathering happening with a goal to form a critical mass of technically oriented creative talented people that will think up, create and deploy ideas. Or it’s a short summer camp for geeks… Take your pick.




1+2=3. Taking it one step further, instead of using physical small hand size objects, YOU will be the players in a musical part by wearing masks. Together, we create a musical part where you can take part by wearing fiducial masks detected by the Tangible laptop.




Reflections:
More Tangible Kabuki images at Flikr tagged with "Tangible Kabuki"

More GeekCon 2007 images at Flikr tagged with geekcon2007

Feature integration theory

The most influential psychological models of human visual attention.

http://en.wikipedia.org/wiki/Feature_integration_theory

ViewFinder FireFox AddOn

Create webpage heatmaps directly from your FireFox.

Install Extension Here
or Download the ViewFinder heatmap extension for FireFox.

sample heatmap of firefox homepage

GuitarHeroNoid

The reason I haven't been blogging for some time, is this robot I built with the GarageGeeks guys. Bringing our image processing and visual attention skills from http://www.Feng-GUI.com to the brain of the GuitarHeroNoid, a robot who play the PlayStation GuitarHero game.

more images at flickr
geek power!








Guitar HeroNoid first live show at KinnerNet2007 dining room.

ViewFinder heatmap for videos

Feng-GUI lab is pleased to announce on the video edition of the ViewFinder heatmap.

for example, see how the following videos are being heated by ViewFinder attention heatmap.

Coca-Cola GTA


The Matrix II trailer


Heineken commercial


Mission Impossible trailer


more info at the research page

ViewFinder - Web page Heat Map

Here is an example of running the ViewFinder heatmap algorithm over some web pages.

web page snapshot as input


web page heatmap as output

ViewFinder HeatMap - Animal Testing

Hold your horses :) Don't run and sign a petition.
"No Animals Were Harmed" during the process of the algorithm.
Image processing tests are painful, only to the one who implements them.

input


output

ViewFinder algorithm - Heat Map

Another enhancement to the ViewFinder algorithm is the HeatMap.
Merging the HeatMap over the original image emphasis the "heat" representing the Regions Of Interest (ROI).

Segmentation

In image analysis, segmentation is the partitioning of a digital image into multiple regions (sets of pixels), according to a given criterion. The goal of segmentation is typically to locate objects of interest and is sometimes considered a computer vision problem. Unfortunately, many important segmentation algorithms are too simple to solve this problem accurately: they compensate for this limitation with their predictability, generality, and efficiency.

wiki on Segmentation

ViewFinder Algorithm - Look for the Red spot

ViewFinder algorithm using Graph based segmentation and weights of the segments which are in an attention threshold.

Visual Attention

“…the visual system makes the assumption that the external world is, by default,
static and unchanging. By making this assumption, an internal representation of the
entire external world can be avoided in preference of a representation that focuses on one or a few pieces of information at a time. … In a dynamic world, if changing
information is not processed immediately, this information will be forever lost. Static information is continuously available and therefore the processing of it can be delayed with little consequence.” [Parkhurst 2002]