google kupił firmę neven vision.
czym się firma zajmuje? na blogu googla jest mocno ogólnikowo, że rozpoznawaniem obrazu. strona firmy neven vision już nie działa. jest tylko informacja, że google ich przejął. i nic więcej.
ale cóż to dla mnie. chwila poszukania i oto:
Neven Vision's computer vision technology is the fastest in the industry and represents leading accomplishments in this field. Several U.S. patents have been granted to Neven Vision for its unique inventions. U.S. government and independent observers have recognized the underlying technology as most advanced with respect to accuracy, speed and efficiency.
The core building blocks of the company's vision-enabled products and services are our patented face and object recognition engines. Below is a simple representation of how each of these engines work.
- Faces in an image are automatically detected by a robust face finder component that determines the position and size of the face in real world conditions (varying illumination, pose and expression).
- A second processing stage determines the position of local features on the face, called “landmark finding�.
- Then a Gabor Wavelet transformation takes place to compute the “face-template� from the local features by extracting the template from an image that contains the essential, condensed facial information needed to determine a person's identity (a “face-template� is only about 1kByte in size; min <350bytes max 1.6KBytes). Two templates are then compared to yield a similarity. Templates belonging to the same person produce high similarity while templates from different persons produce low similarities.
Neven Vision's object recognition technology is based on a local feature approach:
- Objects are recognized by identifying characteristic points of an object, deriving feature vectors from the texture in the vicinity of the characteristic points.
- These feature vectors are then compared to a database of known objects to establish matching correspondences between the current object and the objects contained in the database.
- If a sufficient number of local feature correspondences are found, a positive identification is accepted.
brzmi bardzo interesujÄ…co. o ile pamiÄ™tam oprogramowanie tego typu byÅ‚o używane także w systemach monitorowania lotnisk – do wyÅ‚apywania terrorystów.
raczej wÄ…tpiÄ™ by google zaczÄ…Å‚ inwestować teraz w soft antyterrorystyczny, ale widzÄ™ interesujÄ…cÄ… możliwość – zapisujemy gdzieÅ› jakieÅ› zdjÄ™cie np. wujka jurka. potem każde zdjÄ™cie gdzie on siÄ™ pojawia dostaje automatycznie taga “wujek jurek". to “gdzieÅ›" jest oczywiste – picassa albums.
mam nadzieję, że technologia neven vision zostanie szybko zintegrowana z images.google.com dzięki czemu będę mógł szukać nie tylko po keywordach stron czy nazwach plików, ale po faktycznej zawartości obrazków.





