Machine Learning is a branch of artificial intelligence that builds systems which can learn from data. These systems can adapt to real-world usage and perform better than classic systems in many situations.
Recent advances in Machine Learning technology allows us to extract features embedded in images or videos and use them as input parameters for machine learning algorithms. These algorithms learn from the input parameters and make assessments, classifications, identifications, decisions or predictions.
During this learning process, a human trainer tags areas within multiple images and labels them according to his or her needs and goals.
Based on mathematical and statistical analysis, the software creates a learning database it can use later to label objects in a new set of images. This labeling, also known as recognition reflects the statistical likelihood that a given visual element has matched a label in the learning database.
This learning process is useful for cases where the appearance of the features is not an exact match to a reference image even after the images have been processed and normalized.
On the other hand, recognition of exact matches after normalization does not require Machine Learning.
Machine Learning can identify objects in one specific image or recognize actions based on changes in the object's state over time.
The combination of learning algorithms, and features to be extracted, and the dataset used for learning is important to achieve the desired results because each type of classification can require its own features to be extracted and its own images for training.
For example, a system can learn to identify dogs and differentiate them from cats. Or, classify dogs by breed such as Chihuahua, Bulldog or Poodle. Or classify dogs and cats as four-legged animals. Or identify an individual dog.
We create practical ComputerVision applications using the latest research and breakthroughs in Machine Learning.