Computer Learns Dogspeak: Programs Can Classify Dog Barks Better Than Humans, Study Shows

Daily Science Journal (Jan. 28, 2008) — Computer programs may be the most accurate tool for studying acoustic communications amongst animals, according to Csaba Molnár from Eötvös Loránd University in Hungary and his research team. Their research shows that a new piece of software is able to classify dog barks according to different situations and even identify barks from individual dogs, a task humans find challenging.

A new piece of software is able to classify dog barks according to different situations and even identify barks from individual dogs, according to research by Csaba Molnár from Eötvös Loránd University in Hungary and his team. (Credit: Image courtesy of Springer)

The aim of Molnár and colleagues’ experiments was to test a computer algorithm’s ability to identify and differentiate the acoustic features of dog barks, and classify them according to different contexts and individual dogs. The software analyzed more than 6000 barks from 14 Hungarian sheepdogs (Mudi breed) in six different situations: ‘stranger’, ‘fight’, ‘walk’, ‘alone’, ‘ball’ and ‘play’. The barks were recorded with a tape recorder before being transferred to the computer, where they were digitalized and individual bark sounds were coded, classified and evaluated.

In the first experiment looking at classification of barks into different situations, the software correctly classified the barks in 43 percent of cases. The best recognition rates were achieved for ‘fight’ and ‘stranger’ contexts, and the poorest rate was achieved when categorizing ‘play’ barks. These findings suggest that the different motivational states of dogs in aggressive, friendly or submissive contexts may result in acoustically different barks.

In the second experiment looking at the recognition of individual dogs, the algorithm correctly classified the barks in 52 percent of cases. The software could reliably discriminate among individual dogs while humans can not, which suggests that there are individual differences in barks of dogs even though humans are not able to recognise them.

The authors conclude by highlighting the value of their new methodology: “The use of advanced machine learning algorithms to classify and analyze animal sounds opens new perspectives for the understanding of animal communication… The promising results obtained strongly suggest that advanced machine learning approaches deserve to be considered as a new relevant tool for ethology*.”

* Ethology: the study of animal behavior, with a focus on behavioral patterns in natural environments.

Journal reference: Molnar C et al (2008). Classification of dog barks: a machine learning approach. Animal Cognition (DOI 10.1007/s10071-007-0129-9)

Adapted from materials provided by Springer.

Read the rest of this entry »

Face Recognition Made Possible With New Computer Program

Daily Science Journal (Jan. 25, 2008) — Hung-Son Le, Umeå University, Sweden, has developed in his PhD dissertation the algorithms that give a computer the possibility of recognising a face, even if only one picture is taken. The results can be used for safe and secure identity control or, on the light side, to find out to which famous persons you look alike.

If a non-authorised person gets access to your pin code and credit card, most likely your money will disappear from your account. Nevertheless, this would be impossible if the ATM could recognize your face as you look at a camera. Now, the algorithms to carry out this function, face recognition, exist. Face recognition can also be used in other functions, for instance in a dating service. Maybe the customer is interested in a man that looks like Brad Pitt or a woman that looks like Angelina Jolie.

Systems that can identify different faces are normally trained through a database with a large collection of face images in different illumination and pose. Nevertheless to collect such a large number of face images for each person is difficult and quite often expensive. Moreover these systems have problems due to the bad quality of the pictures, as well as facial expressions, the variety of angles and the different illuminations. These problems are now over.

The effective algorithms developed by Hung-Son Le make it possible to have a system that can identify a face even when there is only one picture in the database for each person. Moreover, the effectiveness of the system is a considerable improvement when taking into account light conditions, or facial expressions. His algorithms use a method than improves contrast in underexposed and overexposed pictures. Thus details can be made visible which otherwise would be difficult for a computer to identify. Given the method used (Hidden Markov Model, HMM), once the system is in place, it needs no time for retraining, when compared to existing HMM-based competitors, to “know” new pictures with different expressions taken under different illumination conditions.

The experiments carried out with the system and tested against international standards such as FERET and the Yale database, have demonstrated that it outperforms the leading competitors.

Commercial applications based on the PhD dissertation results are under development and will soon be presented. Among others, a face websearch engine is under final development phase.

Adapted from materials provided by Umeå University, via AlphaGalileo.

Read the rest of this entry »

Dogs Can Classify Complex Photos In Categories Like Humans Do

Daily Science Journal (Nov. 29, 2007) — Like us, our canine friends are able to form abstract concepts. Friederike Range and colleagues from the University of Vienna in Austria have shown for the first time that dogs can classify complex color photographs and place them into categories in the same way that humans do. And the dogs successfully demonstrate their learning through the use of computer automated touch-screens, eliminating potential human influence.

Researchers have shown for the first time that dogs can classify complex color photographs and place them into categories in the same way that humans do. (Credit: iStockphoto/Rami Ben Ami)

In order to test whether dogs can visually categorize pictures, and transfer their knowledge to new situations, four dogs were shown landscape and dog photographs, and expected to make a selection on a computer touch-screen.

In the training phase, the dogs were shown both the landscape and dog photographs simultaneously and were rewarded with a food pellet if they selected the dog picture (positive stimulus). The dogs then took part in two tests.

In the first test, the dogs were shown completely different dog and landscape pictures. They continued to reliably select the dog photographs, demonstrating that they could transfer their knowledge gained in the training phase to a new set of visual stimuli, even though they had never seen those particular pictures before.

In the second test, the dogs were shown new dog pictures pasted onto the landscape pictures used in the training phase, facing them with contradictory information: on the one hand, a new positive stimulus as the pictures contained dogs even though they were new dogs; on the other hand, a familiar negative stimulus in the form of the landscape.

When the dogs were faced with a choice between the new dog on the familiar landscape and a completely new landscape with no dog, they reliably selected the option with the dog. These results show that the dogs were able to form a concept i.e. ‘dog’, although the experiment cannot tell us whether they recognized the dog pictures as actual dogs.

The authors also draw some conclusions on the strength of their methodology: “Using touch-screen computers with dogs opens up a whole world of possibilities on how to test the cognitive abilities of dogs by basically completely controlling any influence from the owner or experimenter.” They add that the method can also be used to test a range of learning strategies and has the potential to allow researchers to compare the cognitive abilities of different species using a single method.

Journal reference: Range F et al (2007). Visual categorization of natural stimuli by domestic dogs (Canis familiaris). Animal Cognition (DOI 10.1007/s10071-007-0123-2).

Adapted from materials provided by Springer.

Read the rest of this entry »

How Well Do Dogs See At Night?

Daily Science Journal (Nov. 9, 2007) — A lot better than we do, says Paul Miller, clinical professor of comparative ophthalmology at University of Wisconsin-Madison.

Dogs have good night vision, due in large part to the tapetum, a mirror-like structure which reflects light, giving the retina a second chance to register light that has entered the eye. This is also what makes dogs eyes glow at night. The dog is holding a toy in her mouth. (Credit: Michele Hogan)

“Dogs have evolved to see well in both bright and dim light, whereas humans do best in bright light. No one is quite sure how much better a dog sees in dim light, but I would suspect that dogs are not quite as good as cats,” which can see in light that’s six times dimmer than our lower limit. Dogs, he says, “can probably see in light five times dimmer than a human can see in.”

Dogs have many adaptations for low-light vision, Miller says. A larger pupil lets in more light. The center of the retina has more of the light-sensitive cells (rods), which work better in dim light than the color-detecting cones. The light-sensitive compounds in the retina respond to lower light levels. And the lens is located closer to the retina, making the image on the retina brighter.

But the canine’s biggest advantage is called the tapetum. This mirror-like structure in the back of the eye reflects light, giving the retina a second chance to register light that has entered the eye. “Although the tapetum improves vision in dim light, it also scatters some light, degrading the dog’s vision from the 20:20 that you and I normally see to about 20:80,” Miller says.

The tapetum also causes dog eyes to glow at night.

Adapted from materials provided by University of Wisconsin - Madison.

Read the rest of this entry »

Reading Process Is Surprisingly Different That Previously Thought, Technology Shows

Daily Science Journal (Sep. 11, 2007) — Being able to read competently is one of the most important skills we need to function in today’s fast-paced society. Analysing the way we read can offer valuable insights into how we process visual information.

When we read, our eyes look at different letters in the same word and then combine the different images through a process known as fusion, researchers have found. (Credit: iStockphoto/Shannon Long)

Scientists have been interested in the movements of our eyes while reading for forty years. However, until now most assumed that when we read both eyes look at the same letter of a word concurrently.

Now ground-breaking research by cognitive psychologist Professor Simon Liversedge and his team at the University of Southampton has shown that this is not actually the case. They found that our eyes are actually up to something much more exciting when we read - our eyes look at different letters in the same word and then combine the different images through a process known as fusion.

The research Prof. Liversedge will present at the BA Festival of Science in York shows that the reading process is not as simple as one might think; it is rarely a case of the eyes scanning the page smoothly from left to right. Depending on what we are reading and how hard we are finding the information to digest our eyes make small jerky movements, that allow us to focus on a particularly difficult word or often re-read passages we didn’t get the first time. Analysing these eye movements enables psychologists to understand how our brain processes the sentence.

With sophisticated eye tracking equipment able to determine which letter of a font-size 14 word a person is looking at every millisecond from 1 metre away, Prof. Liversedge’s team went one further and looked at the letters within the word within the sentence. They were able to deduce that when our eyes are not looking at the same letter of the word, they are usually about two letters apart. Prof. Liversedge explains: ‘Although this difference might sound small, in fact it represents a very substantial difference in terms of the precise "picture" of the world that each eye delivers to the brain.'

So if our eyes are looking at different parts of the same word, thereby receiving different information from each eye, how is it that we are able to see the words clearly enough to read them? There are two ways the brain can do this; either the image from one of the eyes is blocked or the two different images are somehow fused together. To test how the latter mechanism might work, the team chose words that could easily split in two, such as cowboy, and presented half of the word to the left eye, and half to the right eye separately. They then analysed readers’ eye movements when reading sentences containing these particular words presented in this way.

‘We were able to clearly show that we experience a single, very clear and crisp visual representation due to fusion of the two different images from each eye,’ he explains. ‘Also when we decide which word to look at next we work out how far to move our eyes based on the fused visual representation built from the disparate signals of each eye.

‘A comprehensive understanding of the psychological processes underlying reading is vital if we are to develop better methods of teaching children to read and offer remedial treatments for those with reading disorders such as dyslexia. Our team are now measuring the range of visual disparities over which both adult and child readers can successfully fuse words.’

Professor Simon Liversedge will give his talk, ‘What our eyes get up to while we read’ as part of the session entitled ‘What eye movements tell us about the brain and language’ on 14 September at Vanbrugh V/045, University of York as part of the BA Festival of Science.

Adapted from materials provided by British Association for the Advancement of Science.

Read the rest of this entry »

The New Face Of Identity Protection: You

Daily Science Journal (Aug. 5, 2007) — Trying to remember dozens of personal identification numbers (PIN), passwords and credit card numbers may not be necessary for much longer, thanks to a University of Houston professor and his team.

Taking a radically new approach, UH Eckhard Pfeiffer Professor Ioannis Kakadiaris and his Computational Biomedicine Lab (CBL) developed the URxD face recognition software that uses a three-dimensional snapshot of a person’s face to create a unique identifier, a biometric. Shown in government testing to be tops in its field, URxD can be used for everything from gaining access to secure facilities to authorizing credit card purchases. The identification procedure is as effortless as taking a photograph.

URxD leads the pack for 3D face recognition solutions based on the face’s shape, according to the results of the Face Recognition Vendor Test (FRVT 2006). The National Institute of Standards and Technology conducted the rigorous testing for FRVT 2006, which was sponsored by several U.S. government agencies. FRVT 2006 is the first independent performance benchmark for 3-D face recognition technology.

“Accuracy is the name of the game in 3-D face recognition,” Kakadiaris said. “What makes our system so accurate is the strength of the variables that we use to describe a person’s face.

“Remembering dozens of personal identification numbers and passwords is not the solution to identity theft. PINs and passwords are not only inconvenient to memorize, but also are impractical to safeguard. In essence, they merely tie two pieces of information together; once the secret is compromised, the rest follows. The solution is to be able to tie your private information to your person in a way that cannot be compromised.”

The software and technology also could play a role in national security.

“With the growing concern for security at the personal, national and international level, the University of Houston is pleased that Dr. Kakadiaris and his team have demonstrated a very promising technology for personal identification,” said John Warren, UH associate general counsel for research and intellectual property management. “We look forward to its adoption by government and industry.”

URxD inventors are hoping for corporate interest in bringing the technology, now at the advanced prototype stage, to the marketplace.

“This technology will have a positive impact on some of today’s hottest issues,” Kakadiaris said. “Imagine a day when you simply sit in front of your computer, and it recognizes who you are. Everything will be both easier and more secure, from online purchases to parental control of what Web sites your children can visit.”

Note: Use of results from the Facial Recognition Vendor Test 2006 does not constitute the U.S. government’s endorsement of any particular system.

Adapted from materials provided by University of Houston.

Read the rest of this entry »

A Blueprint For 'Smart' Health Care

Daily Science Journal (Jul. 30, 2007) — Always on, connected, cheap and on sale everywhere.

A cell phone displays a patient's health information in this photo illustration of new technology for providing smart health care announced recently by the University of Florida and IBM. (Credit: Image courtesy of University of Florida)

What people have come to expect in cell phones and personal communicators may soon become common in health-care devices and products at home and in medical offices, thanks to new technology announced recently by the University of Florida and IBM.

The technology creates the first-ever roadmap for widespread commercial development of "smart" devices that, for example, take a person's blood pressure, temperature or respiration rate the minute a person steps into his or her house -- then transmit it immediately and automatically to doctors or family.

That could eliminate the need for many doctor's visits, which are often difficult for the elderly or sick. By enabling regular updates via text message or e-mail, the technology also could pave the way for people to share real-time information on their health or well-being with absent loved ones. And it could prove useful for doctors who need to keep tabs on many patients at one time by helping the doctors to prioritize whom to treat first.

"We call it quality-of-life engineering," said Sumi Helal, professor of computer engineering and the project's lead UF researcher. "It's really a change of mindset."

The idea of using technology to provide medical care at a distance is nothing new. Doctors have relied on "telemedicine" to communicate with specialists for years. More recently, telemedicine has been expanded to include, for example, surgeons performing robotic procedures on distant patients.

But the UF-IBM advance goes a step further: It provides the technological "stepstones" to make it easy for any company to manufacture and sell smart networked devices -- while also making them more user-friendly for consumers.

"UF and IBM both see the need and the opportunity to integrate the physical world of sensors and other devices directly into enterprise systems," said Richard Bakalar, Chief Medical Officer for IBM. "Doing so in an open environment will remove market inhibitors that impede innovation in critical industries like health care and open a broader device market that's fueled by uninterrupted networking."

Helal has devoted the past several years to developing smart devices for the elderly in a model home known as the "Gator Tech Smart Home" in Gainesville.

He and his students pioneered the "Smart Wave" microwave oven that can automatically determine how much time to cook a frozen meal or keep track of how much salt it contains. Among other devices, they also created an instrument that records how many steps a person takes, information that can tell absent caregivers how active its occupants are.

But these and other devices currently have a major shortcoming: They require "a team of engineers" to install them, Helal said. In a world where consumers are accustomed to electronics that require no more than a power outlet, that dramatically limits their appeal. "We decided to create a technology that self integrates," Helal said. "When you bring it in to the house and plug it in, it automatically provides its service and finds a path to the outside world."

With $60,000 in research funding from IBM, Helal designed "middleware," or software and hardware that glues together different systems, that can give his and any similar health-aid devices this independence and connectivity. Importantly, the software is based on open standards, or publicly available specifications useable by anyone, such as those now being made available by consortiums of technology companies including Eclipse, W3C and OSGi.

Open standards make it easy for product developers to tap the technology in any new smart assistive devices, Helal said. That, in turn, will make the devices more common.

The hardware component of the system is an inexpensive sensor platform about half the size of a business card. Developed at UF and licensed to Pervasa, a Gainesville-based UF spinoff company headed by Helal, the "Atlas" platform makes it easy to create a network of sensors and make their information available on a computer network.

The advance is crucial given the increasing number of elderly Americans. The number of people 85 and over is expected to rise from 4.2 million in 2000 to 6.1 million in 2010 and 9.6 million by 2030, according to federal government statistics. Meanwhile, the percentage of older Americans living alone will either remain high or continue to grow: About half of women and nearly a quarter of men aged 75 and older currently live alone.

But the UF-IBM technology may also prove useful in many other medical settings. For example, Helal said, it could help emergency rooms operate more safely. Rather than a standard waiting list, patients could be equipped with networked wireless monitors of their vital signs, allowing doctors to determine who in a waiting room needs the most immediate care.

Adapted from materials provided by University of Florida.

Read the rest of this entry »