Face recognition is used to identify individuals on our continuously incoming stream of data. By using deep Convolutional Neural Networks, we recognize the faces of all people that are known to our algorithms. With this, users can receive notifications when their favourite athletes, politicians or pop stars are on television, monitor the context they appear in, and receive personal recommendations.
Media Distillery is always using the latest speech recognition algorithms to ensure high-quality recognition optimized for the type of audio or video. The large vocabulary system based on deep learning is able to recognize a wide spectrum of languages. And our skilled data science team is able to create a speech model for each application to ensure high quality.
General object recognition gives additional value to our platform by providing more context to the analyzed content. By using our query language, users can combine the results of, for example, our face recognition with certain objects that are found. Our database of over a thousand objects enables very detailed context description to enrich the results retrieved by our system.
To recognize brands and logos at a large scale in real-time, we use the latest GPU-powered Deep Learning algorithms. In a multi-step approach, our algorithms first identify areas of interest using Region Proposal Networks and then determine if there is a logo visible. We can train the system by providing it with a number of example images. From then on, it will be able to recognize a new brand.
Our state-of-the-art text recognition technology can translate visible words in video streams into searchable text. This way you can search through text that appears in video the same way you can search through text that exists in documents. Our technology can help translate and interpret subtitles, identify brands and even index name tags.
To provide a more high-level entry to the raw data retrieved by all single analyses, we enable the user to query topics in our search engine by using topic detection. Topic detection uses models that create word embeddings based on N-grams, and with the help of these embeddings, we can train advanced machine learning algorithms to classify entire documents for topics.