IBM Watson Visual Recognition

The main articles are:

• Video analytics (terms, applications, technologies)
• Computer vision (machine vision)
• Pattern recognition

2022: Pattern recognition testing failure

Artificial intelligence recognizes images worse than humans.

Computer vision does not have the physiological features that a person has, so it recognizes images worse. This conclusion was reached by scientists from the Higher School of Economics and Moscow Polytechnic University. The HSE announced this on September 7, 2022.

To understand how machine perception of images differs from human, Russian scientists have uploaded images of classic visual illusions to IBM Watson's online image recognition service Visual Recognition. Most of them were geometric silhouettes, partially hidden by geometric forms of the background color. The system tried to determine what represented the incoming image and indicated the degree of confidence in its response.

It turned out that artificial intelligence is not able to recognize any imaginary figure. The exception was a painted imaginary triangle. Due to the high contrast with the background, it was recognized correctly.

Objects similar to those that we used during the experiment are found in real life, "said study author Vladimir Vinnikov, an analyst at the HSE Department of Scientifically Computer Science's Big Data Analysis Methods Research Laboratory. - For example, a trailer trailer or a radio tower, which at night are indicated only by dimensional lights, autopilot car or plane perceives in the same way as we are imaginary geometric figures.

The human eye constantly moves involuntarily, and the photosensitive surface of its retina has the shape of a hemisphere. For a person to see an illusion, it is enough for the image to be vector - to consist of reference points and curves connecting them. The human imagination will complete the picture thanks to the physiological feature of vision - constant eye movement.

In optoelectronic systems, everything is arranged differently. Their photosensitive matrix has a flat, usually rectangular, shape, and the lens system itself is far from as free to move as the human eye. Therefore, artificial intelligence cannot complete imaginary lines that connect fragments of geometric illusion. Machine vision sees only what is actually depicted, while a person builds in the imagination a complete image in its outlines.

Neural network pattern recognition systems are actively distributed in the commercial sector. However, the question of how accurately the machine recognizes the image is still open. Human lives may depend on the accuracy of its recognition. For example, if the autopilot of a car or aircraft does not recognize an object with low contrast relative to the background and does not have time to evade the obstacle in time, a disaster may occur.

Scientists believe that the shortcomings of machine pattern recognition can be corrected.

For example, to supplement the recognition of raster images, which are a grid of pixels, with imitation of physiological features of eye movement, which allow the eye to see two-dimensional and three-dimensional scenes. An alternative is to add a vector description of the images. It will allow you to program the machine to bypass the image along the trajectories specified by the vectors.

Imaginary shapes should definitely be used as tests in systems that depend on the recognition of photo and video streams. For example, in autopilots of cars or unmanned aerial vehicles. This will help avoid risks associated with the use of machine intelligence systems in industry and transport systems, Vladimir Vinnikov believes.