March 24, 2015
I read “Images That Fool Computer Vision Raise Security Concerns.” I found the write up a reminder that the marketing and venture capitalists’ hype are one thing. Real world software performance is another thing.
The article states:
Cornell researchers have found that computers, like humans, can be fooled by optical illusions, which raises security concerns and opens new avenues for research in computer vision.
The passage I highlighted in a mellow yellow says:
But computers don’t process images the way humans do, Yosinski [a Cornell wizard] said. “We realized that the neural nets did not encode knowledge necessary to produce an image of a fire truck, only the knowledge necessary to tell fire trucks apart from other classes,” he explained. Blobs of color and patterns of lines might be enough. For example, the computer might say “school bus” given just yellow and black stripes, or “computer keyboard” for a repeating array of roughly square shapes.
It turns out that this diagram looks exactly like a penguin.
The smart software sees the abstraction as what most grade school children know as a lovable penguin. I did not smell a penguin until after I left grade school. Someone should have warned me.
And the challenge? I have no comment about the expectations of a government professional who relies on image recognition as part of an on going investigation.
Stephen E Arnold, March 24, 2015
February 17, 2015
In order to build a fantastic Web site these days, you need eye-catching graphics. While creating a logo can be completed with Fiverr, making daily images for your content feed is a little bit more difficult. It is not cost efficient to hire a graphic designer for every image (unless you have deep pockets), so it helps to have an image library to retrieve images. The problem with typing in image library into a search engine means you have to sift through results and assess each possible source.
Graphic designer Ash Stallard-Phillips collected “25 Awesome Sites With Stunning Free Stock Photos.” He rounded up the image libraries, because:
“As a web designer myself, I always find it handy to have an image library that I can use for dummy images and testing. I have compiled a list of the best sites offering free stock photos that you can use for your projects. “
Ash evaluates each resource, listing the pros and cons. Many of the image Web sites he lists are ones we have not used before and will be useful as we create content. There is an increase in the number of articles like Ash’s on the Internet and they are not just for photo libraries. They are lists that have tons of helpful information that you would usually have to sift through search results for. It saves time on searching and the evaluation process.
January 19, 2015
Watson has been going to town in different industries, putting to use its massive artificial brain. It has been working in the medical field interpreting electronic medical record data. According to Open Health News, IBM has used its technology in other medical ways: “IBM Research Scientists Investigate Use Of Cognitive Computing-Based Visual Analytics For Skin Cancer Image Analysis.”
IBM partnered with Memorial Sloan Kettering to use cognitive computing to analyze dermatological images to help doctors identify cancerous states. The goal is to help doctors detect cancer earlier. Skin cancer is the most common type of cancer in the United States, but diagnostics expertise varies. It takes experience to be able to detect cancer, but cognitive computing might take out some of the guess work.
Using cognitive visual capabilities being developed at IBM, computers can be trained to identify specific patterns in images by gaining experience and knowledge through analysis of large collections of educational research data, and performing finely detailed measurements that would otherwise be too large and laborious for a doctor to perform. Such examples of finely detailed measurements include the objective quantification of visual features, such as color distributions, texture patterns, shape, and edge information.”
IBM is already a leader in visual analytics and the new skin cancer project has a 97% sensitivity and 95% specificity rate in preliminary tests. It translates to cognitive computing being accurate.
Could the cognitive computing be applied to identifying other cancer types?
January 8, 2015
Image search is a touchy subject. Copyright, royalties, privacy, and accuracy are a huge concern for image holders and searchers. People are scouring the Internet for images they can freely use without problems, but often times the images have a watermark or are so common they are mediocre. Killer Startups points to a great new startup that could revolutionize how people find pictures: “Today’s Killer Startup: Compfight.”
Compfight is an image search engine comparable to Flicker, except it is faster and uses features similar to the advanced search function on Google.
“The site also lets you specify if you’re looking only for Creative Commons licensed images or ones to use commercially. If you’re new to this kind of image use, Compfight even provides a handy little guide on how to cite your sources properly. Last and probably least, Compfight also provides access to professional stock photos, starting as low as $1 per image.”
Developers are still trying to create the perfect image search and while it is a work in progress, Compfight shows we’re on the right path.
November 29, 2014
The idea that numerical recipes can identify a person in video footage is a compelling one. I know one or two folks involved in law enforcement who would find a reliable, low cost, high speed solution very desirable.
The face on the left is a reverse engineered FR image. The chap on the right is the original Creature from the Black Lagoon. Toss in a tattoo and this Black Lagoon fellow would not warrant a second look at Best Buy on Cyber Monday.
I read “This Is What Happens When You Reverse Engineer Facial Recognition.” the internal data about an image is not a graduation photograph. The write up contains an interesting statement:
The resulting meshes were then 3D-printed, creating masks that could be worn by people, presenting cameras with an image that is no longer an actual face, yet still recognizable as one to software.
Does this statement point to a way to degrade the performance of today’s systems? A person wanting to be unrecognized could flip this reverse engineering process and create a mash up of two or more “faces.” Sure, the person would look a bit like the Creature from the Black Lagoon, but today’s facial recognition systems might be uncertain about who was wearing the mask.
Stephen E Arnold, November 29, 2014
October 16, 2014
Image search is touted as being intuitive and accurate. Users simply need to submit an image to the search engine and based off the picture analyzing algorithms similar images will be returned. It, however, is still in the works. Image search still proves to be a difficult task for search engines to master. Search Engine Watch brings us the news “Bing Unveils Responsive Design For Image Search” that the search engine is ramping up to improve its image search.
The newest improvements optimizes image search for touch screen mobile devices. Bing has changed the way uses can browse through images, making it simpler to explore and refine results. Pinterest board searches have been added and a mini-header that will slide with users as they scroll down will offer quick access to popular results. The image hover feature has also been updated.
Along with the updates, Bing has these tips to improve image search:
• “Quality: No matter what the user is searching for, Bing is focused on providing high-quality and relevant image search results.
• Suggestions: Users that are scrolling page after page are clearly having a difficult time finding what they are looking for. Bing maintains a set of search suggestions and collections to help users find what they need.
• Actions: There are many different ways to search and endless topics to search about. Bing has provided the tools necessary to filer results, create an image match, and create one-click access to Pinterest.”
These upgrades will improve image search, but it still has a long way to go.
July 30, 2014
Robots. Magic stuff like semantic concept lenses. Logical wackiness like software that delivers knowledge.
I read “Twitter Acquires Deep Learning Startup Madbits.” The write up points out to the drooling venture capitalists that Twitter’s purchase is “the latest in a spate of deep learning and computer vision acquisitions that also includes Google, Yahoo, Dropbox, and Pinterest.” What this means is that these oh-so-hot outfits are purchasing companies that purport to have software that can figure stuff out.
I recall a demonstration in Japan in the late 1990s. I was giving some talks in Osaka and Tokyo. One of the groups I met showed me a software system that could process a picture and spit out what was in the picture. I remember that the system was able to analyze a photo of a black and white cow standing in a green pasture.
The software nailed it. The system displayed in Japanese, ?. My hosts explained that the idiograph meant “cow.” High fives ensued. On other pictures, the system did not perform particularly well.
Flash forward 30 years. In a demonstration of image recognition at an intelligence conference, the system worked like a champ on clear images that allowed the algorithm to identify key points, compute distances, and then scurry off to match the numeric values of one face with those in the demo system’s index. The system, after decades of effort and massive computational horsepower increases, was batting about .500.
The problem is that different pictures have different looks. When the subject is wearing a baseball cap, has grown a beard, or is simply laughing, the system does not do particularly well.
You can see how Google performs. Navigate to Google Images, select a picture of a monument, and let Google find matches. Some are spot on. I use a perfect match example in my lectures about open source intelligence tools. I have some snaps in my presentatio0n that do not work particularly well. Here’s an example of a Google baffler:
This is a stuffed pony wearing a hat. Photo was taken in Santiago, Chile at an outdoor flea market.
This is the match that Google returned:
Notice that there were no stuffed horses in the retrieved data set. The “noise” in the original image does not fool a human. Google algorithms are another kettle of fish or booth filled with stuffed ponies.
The Twitter purchase of Madbits (the name suggests swarming or ant methods) delivers some smart folks who have, according to the write up, developed software that:
automatically understands, organizes and extracts relevant information from raw media. Understanding the content of an image, whether or not there are tags associated with that image, is a complex challenge. We developed our technology based on deep learning, an approach to statistical machine learning that involves stacking simple projections to form powerful hierarchical models of a signal.
Once some demonstrations of Twitter’s scaling of this interesting technology is available, I can run the radiation poisoning test. Math is wonderful except when it is not able to do what users expect, hope, or really want to get.
Marketing is good. Perhaps Twitter will allow me to track down this vendor of stuffed ponies. (Yep, it looked real to me.) I know, I know. This stuff works like a champ in the novels penned by Alastair Reynolds. Someday.
Stephen E Arnold, July 30, 2014
July 25, 2014
A new image-based search tool can take some of the research out of adopting a pet. Lifehacker turns our attention to the free iOS app in, “PetMatch Searches for an Adoptable Pet Based on Appearance.” Now, pet lovers who see their perfect pet on the street can take a picture and find local doppelgangers in need of homes. Perhaps this will help lower dog-napping rates. Reporter Dave Greenbaum notes:
“You should never adopt an animal solely based on looks, of course—you should research the personality of the breed you want—but looks are a factor. This app works great for mixed breed dogs when you aren’t sure what kind of dog you are looking at. I like the fact it will look at local adoption agencies to find a match, too. Online services like Petfinder.com help you find local pets to adopt, but you have to know which breed you are looking for first, and searching for mixed breed dogs (common at shelters) is difficult. This app makes it easy to do a reverse image search and do your research based on the results.”
Another point to note is that PetMatch includes a gallery of dog and cat breeds, so if the picture is in your head instead of your phone, you can still search for a look-alike. It’s a clever idea, and an innovative use of image search functionality.
Cynthia Murrell, July 25, 2014
June 17, 2014
The Metropolitan Museum of Art offers the new picture archive The Collection Online, with some 400,000 images of art searchable by artist, culture, method, material, geographic location, date or era, and even by department. From costumes to books to ceramics to the more obvious paintings and sculpture in limestone or bronze, this collection is incredible in its scope and detail. Searching, say, for a painting by Vincent Van Gogh yields a list of 124 records. Many of these were not works by Van Gogh, but there is an option to limit to the records by that painter. Click on The Potato Peeler and find not only the stats of the painting (oil on canvas from 1885, 16 x 12 1/2in.) but also where to find it in the Museum (Gallery 826) if available. Beneath the image there is some additional information available,
“This painting from February/March 1885, with its restricted palette of dark tones, coarse facture, and blocky drawing, is typical of the works Van Gogh painted in Nuenen the year before he left Holland for France. His peasant studies of 1885 culminated in his first important painting, The Potato Eaters (Van Gogh Museum, Amsterdam).”
You are also informed that on the reverse of the same canvas is a Van Gogh’s Self Portrait with a Straw Hat. Users have the possibility of registering for free, which would enable them to create their own assortment by saving images to an individual collection. The Met is no stranger to successful online endeavors, having just recently won a Webby for their Instagram account through the Academy of Digital Arts & Sciences.
Chelsea Kerwin, June 17, 2014
February 10, 2014
Here’s a new way to search from one of the minds that helped loose Twitter upon the world. The Los Angeles Times shares an interview with a Twitter co-founder in, “Biz Stone Answers our Questions About New Q&A App Jelly.” Forget algorithms; this app lets you take or upload a picture and pose a question about it to other humans, both within and outside your social-media circles.
Stone and co-founder, Ben Finkel, started with a question: if we were to design a search tool around today’s online landscape, as opposed to the one that existed about a decade ago, what would it look like? As the app’s website explains, “It’s not hard to imagine that the true promise of a connected society is people helping each other.” (Finkel, by the way, founded Q&A site Fluther.com and served as its CEO until that service was acquired by Twitter in 2010.)
One of Jelly‘s rules may annoy some: users cannot post a question without including an image. Writer Jessica Guynn asks Stone why he incorporated that requirement. He responds:
“We did a lot of testing and more often than not, an image very much deepens the context of a question. That’s why we made it so you can either take a picture with your camera and say, ‘What kind of tree is this?’ Or you can pull from the photo albums you already have. Or you can get [a photo] from the Web. Photos are what make mobile mobile. We are really taking advantage of the fact that this is a mobile native application…. Everyone is carrying around these great cameras. It’s a uniquely mobile experience to pair a short question with a photo. It might frustrate a few people in the long run but it will only end up with better quality for us. There is a higher bar to submitting a question.”
The image requirement is just one way Jelly differs from Twitter. The team also worked toward making the new app less conversational in order to avoid the clutter of non-answers. (And we thought 140 characters was limiting.) We’re curious to see how well users will warm to this unique service.
Cynthia Murrell, February 10, 2014