Technology

Kuznech Visual Search & Recognition Technology was created by a team of devoted and persistent scientists, machine engineers, and developers. We pulled only the best from world-class scientific research on deep learning performed over the last two decades, synthesized that knowledge with the research we have constantly ongoing, and created a system that successfully handles a wide range of visual search tasks: from face detection and recognition to video search and brand marking.
Deep learning is a field of computer science that seeks to mimic the human brain with hardware and software. Kuznech Visual Search and Recognition Engine is powered by convolutional neural networks, a type of deep-learning technology. We use convolutional neural nets because they are useful to recognize visual patterns from pixel images and videos with minimal preprocessing and can recognize patterns with extreme variability and robustness to distortions.
Deep architecture model is also unique because it can learn over time in much the same way people do: the technology can be taught to process, detect and recognize almost any type of object, simulating the intelligence of the human brain.
A deep architecture model at Kuznech’s technology core can be represented as an ensemble of decision trees. Every tree features several decision units split into two types:
  • fast detectors with binary feature extractors and
  • robust classifiers based on a visual cortex model.
Our engineers effectively combined hand-crafted methods with self-organized neural structures to provide fast and accurate pattern classification applicable to a wide range of recognition tasks.
nvidiaKuznech Visual Search & Recognition Technology uses the NVIDIA GPU technology that makes it possible to cut processing times to a fraction of a second for images and a couple seconds for videos – a huge boost over the traditional client-server setup. Kuznech has partnered with NVIDIA to provide unique, multi-GPU configured solutions which enhance productivity and increase performance for CUDA™-based parallel processing applications.
Visual recognition demands accurate localization and classification of objects visible in images and videos. Kuznech cherry picked the best state-of-the-art neural networks and improved them by applying its own engineering methods, resulting in highly accurate recognition. This approach has already come through with flying colors for face detection and recognition, content filtering, and brand recognition. With the help of NVIDIA GPU, our team has designed high-performance cloud systems for various image recognition services.
Since people are moving more and more into mobile in their way of online search, mobile visual search has become one of Kuznech big targets. Kuznech Mobile Visual Search & Recognition Technology has modeled the human visual cortex in its propensity to pick out many points of interest of the object and create a “fingerprint of a target” – and not scan it from top to bottom or from one side to another. After the “fingerprint” of an input image or video has been created, Kuznech technology matches it against databases of images created by retailers or advertisers. Kuznech Mobile Engine can effectively deal with such technical challenges, typical for mobile image search, like different quality cameras, shifted color balance, blurring, and over-exposure.
Kuznech Visual Search & Recognition Technology is patented in the US, with several patents pending.
We at Kuznech are hard at work developing the world’s leading visual search engine, opening up new horizons for scientific societies, developers, and all kinds of image- and video-related businesses. We fully utilize the latest advances in modern GPU computing, object recognition and machine learning to make visual search a part of everyday life.