Using Text Search Ideas to Speed Up Image Search

Microsoft Research news: Text-Search Tricks Speak Volumes in Image Search, May 2007

Finding similar images on Web used to require prohibitively high computational cost. Now however, researchers use text-search ideas to make content-based image search commercially feasible. For each image, a set of features are detected, and each feature is represented by a vector describing its characteristics such as orientation and intensity. In this way, each image resembles a document in text-search, and each vector resembles a token. Vocabulary is generated from millions of tokens gathered, and an inverted index could be built. Thus, the speed of finding a similar image on the Web falls to around 0.1 seconds.

A project call “Photo2search in Beijing” has turns this idea into reality. Geo-tagged street view images with longitude and latitude are crawled from photo-sharing websites like Flickr and put into an inverted index. Imagine you get lost in Beijing. Just pull out your camera phone, shoot a photo of your surroundings, send it to the system, and you get a digital map with your position marked on it by matching your photo to the most similar geo-tagged street view images.


Relevant lecture:  

23. VECTOR MODELS (11/17)

27. MULTIMEDIA IR (12/1)

Comments (3)

An Army of Ones and Zeroes

“An Army of Ones and Zeroes: How I became a soldier in the Georgia-Russia cyberwar.” by Evgeny Morozov via

As it stands, no one can really dispute that the last decade has brought significant changes to our societal definitions of warfare. Most obvious among these changes is the shift from the nation-to-nation principals of the Clausewitzian era to a new 21st century battlefield of non-state actors and Radio Shack enhanced IT tactics. Meanwhile, our military schools and strategists redefine their tactics and goals as they struggle to keep up.

In Morozov’s journalistic experiment, the author channels Matthew Broderick’s cheekiness from “Wargames” while signing on to act as a cyber-soldier against Georgia in its recent military face-off with Putin’s Russia. Experimenting with simple page-reload scripts and DOS attacks, Morozov describes his exploits against Georgian government information sites using widely available, pre-built tools that made joining the ranks so easy that he was left with “concerns about the number of child soldiers who may just find it too fun and accessible to resist.”

Given that warring countries have always had very different “calls-to-arms” for their citizen militias, I wonder how technologically sophisticated societies will harness the power of their citizens in information warfare over the next decade. While it’s somewhat hard to imagine the United States asking its general population to militarize their home computers for an information assault on China, it’s not unrealistic to envision a war between the Korean states or between China and Taiwan being fought in-part by thousands of teenage, or even elderly patriots recruited and trained in advanced cyber-warfare using an advanced social network that uses internal feedback systems such as quests, rankings, and rewards to promote its soldiers . Warfare 2.0 FTW. Kinda scary.

What struck me about this article is that given the expanding toolkit of tracking and surveillance hardware installed throughout this country, Morozov mentions nothing about nation vs. civilian reprisals. If Georgia discovers you are attacking its infrastructure, how can it strike back? Is this perhaps why he didn’t choose to attack the Russians despite his statement that his “geopolitical sympathies…lie with Moscow’s counterparts.”

I think it is valid to make correlations between this article and concepts discussed by Vannevar Bush and “Operation Clean Data.” When your information is centralized, is it not also weakened from a security standpoint? How are government information systems designed to provide data unification, internal transparency, redundancy, modularization, and useability all at the same time? How do cyber-warfare techniques exploit these systems with automation and retrieval? Surely these issues are being weighed by information and security experts to anticipate the many changes in the future of information warfare.


Comments off

Data Fusion: The Ups and Downs of All-Encompassing Digital Profiles

The current issue of the magazine Scientific American includes several articles on the rise of digital information and its use, from RFID and biometrics to eavesdropping. I found an article on data fusion, aka data integration, as the most relevant for this discussion.

Data Fusion: The Ups and Downs of All-Encompass Digital Profiles

The article begins with the author’s reflection on his experience traveling internationally several years ago, when his credit card issuer blacklisted his card because the company’s anti-fraud data mining algorithm detected potential fraud. He had merely bought a latte and a cell phone SIM in England. The company knew he was in England, as he’d bought his ticket to England with the same credit card! Shouldn’t they have known it was him?

From this introduction the author traces the history of data mining efforts in the United States, focusing in particular on the challenges of integrating multiple data sources together in order to data mine effectively. He cites examples from DARPA’s counter-terrorism efforts and the Department of Health & Human Services anti-fraud efforts, exploring how data mining is viewed as an ultimate tool, however one that is still rife with inconsistency and errors.

Errors primarily arise from the difficulty normalizing data from varying sources with varying levels of detail and uncertainty. And perhaps most importantly, that once those data sources are aligned, how does one guarantee identity? Who’s who? Am I Andy Brooks, A.L Brooks, and/or Andrew Brooks?

After examining data mining’s shortfalls, the author turns to examples of more effective work done with data fusion and data mining. The winners? Casinos! Think of those wallet-sized perks cards casinos are so happy to give you. In order to counter the efforts of cheaters, casinos have long funded development of non-obvious relationship analysis techniques. The techniques attempt to normalize data across multiple sources in an evolving way, one that tolerates error and uncertainty, and strives to grow more intelligent over time.

The article concludes with the author’s most important point – that similar to the history of cryptography, the public is essentially left out of any discussions about the use of data mining and data fusion. We don’t really know when it’s used, by whom, and for what purpose. We often only know after the fact, such as when the author’s credit card was declined when trying to buy a train ticket in London.

Candidate Lectures:

11. Information Integration & Interoperability (10/6)

15. Personal Information Management (10/20)

Comments off