Cognitive Services - Vision Category

Computer Vision

Most of us will quickly bring image analysis in conjunction with high mathematical formulas and pattern recognition algorithms. The Computer Vision API but is based on a completely different level of abstraction. If one sends a picture of a squirrel on the service, “a squirrel is sitting on a rock” and a percentage chance comes in response, how much you can trust the statement. In addition, their position, as well as dominant color values can be queried tags and categories (animal, outdoor etc.), including faces. For portals, where images are uploaded by users, marking of potentially racist or adult content is helpful.

In addition to the - almost unspectacular in comparison - OCR recognition (optical character recognition), which is also offered as a service, a very useful API stands out: the thumbnail service. From an arbitrary image, a thumbnail of the desired size is generated and it tries to choose the best cut-out. Detected a logo, is the image fully preserved and will only shrink. Photos faces and dominant objects as well as possible in the image remain, while the rest will be truncated.

Emotion API

Who was already excited by the recognition of the age and sex, should have seen the Emotion API. Faces can be analyzed in terms of feeling expression: _Does the person look angry, shocked, scared, happy, neutral, sad or_surprised? The service works for pictures the same as for videos and is also available in a limited free version available (thirty thousand image requests, three hundred uploads per month). The premium packages are between 0.10 and 0.25 dollar per thousand transactions.

Face API

The Face API cares less about the expression as to the identification of faces. While earlier, the position of the face is detected, this API also provides information about the eyes and eyebrows, nose and corners of the mouth. It is recognized that the person wears a beard or is a glasses on has. This data is in turn relevant to identify faces. The face API can determine whether two photos show the same person or not - and how inferior everywhere with a probability. The detection works doing surprisingly well, even if the faces are partially covered or look into the camera.

Video API

The vision categoriy is finally completed by some video services. From image stabilizers and detecting faces, already described a thumbnail service to determine of possible meaningful thumbnails available is also here. In addition to using “Motiondetection” sequences are detected, in which a movement takes place - whether it reliable is enough for the homemade alarm system, you need to decide.