While the social media is buzzing with inaccuracy in picture put out by Google search while people key-in for India's first Prime Minister, American search giant has received an Indian patent for an invention relating to identifying images using the face-recognition method.
According to reports, this patent application was filed in 2009 at the Mumbai patent office and after 10-years it was granted on April 16 by the government. Google claims the method includes identifying the named entity, retrieving images associated with the entity and performing face-detection on the retrieved images to detect faces in the retrieved images, where a face detection was performed using a face-detection algorithm.
To make the whole exercise foolproof, the method further includes identifying at least one representative face image from the retrieved images and using the representative face image it can identify one or more additional images of the entity, according to the patent document filed by Google.
Google, in its filing, pointed out that names of persons, especially those of the well-known ones, are popular queries in the large pool of contents on the web. In conventional search systems, the results of a query for a person could include images of the person identified in the query.
In order to identify and return appropriate content in response to a query, the content must be associated with the search query. For instance, to identify images associated with a search query, it requires that the images are to be keyed against the person's name, through various means including metadata, tags and similar methods.
“Unfortunately, images in large collections such as available through the web, including pictures and screens shots from video content, typically are not keyed and may not include information associated with a pictured person's name. Accordingly, conventional search systems may not be able to identify these un-keyed images responsive to a search query that includes an individual's name,” Google submitted.
The company said that the method can also include performing face-matching for the detected faces in the retrieved images, including comparing each detected face to each other detected face, to identify faces that match a particular face image associated with the named entity.
The number of faces that match the particular face image can also be determined. Moreover, the method can include comparing the determined number of faces that match the particular face image to a threshold value, and or determining the detected face in the retrieved images that matches the highest number of other detected faces in the retrieved images.