Our vision is to make all images on the internet accessible to blind and partially sighted people.
We have developed a web-service that provides meaningful descriptions for online images, with an emphasis on social image sharing applications (Flickr, Facebook etc.).
So using this webservice can determine that this photo has a boy and a girl, standing with a dog, on the sea beach with blue sky. Using social data where possbile we can mash up the names, relations and locations of people in the photos.
The service can also interface with screen reader, voice over services and embedded within browser extensions.
For this prototype, we are utilising cloud based image processing apis from rekognition.com, iqengines.com and mash social data from Facebook.com. Our long term goal is to utilize open source technology and develop the necessary infrastructure and services to provide open apis. We are hosting all the code on Github.
Detects and recognizes faces with guesstimations about gender, facial expressions and Recognizes common scenes - Nightlife, Beach, Urban etc.
See a working example at http://rekognition.com/demo/
objectQuery(): Send an image to the IQ Engines server, which will extract its visual content Parameters: Input image file name Returns: Unique query id associated with the input image
resultQuery(): Request classification information for an image using unique identifier that was generated when posting the image initially Parameters: Unique image indentifier Returns: Currenly plain json response from the server, containing all the useful bits
We currently use only the following calls:
queryApi: http://api.iqengines.com/v1.2/query/ docs: https://www.iqengines.com/apidocs/apis/query-api.html
resultApi: http://api.iqengines.com/v1.2/result/ docs https://www.iqengines.com/apidocs/apis/result-api.html
1. Each API call should be associated with a unique api_sig hash
2. api_sig is generated by hashing input variables in _alphabatical_ order
3. The api_sig that is computed while posting an image to the server, needs to be
tracked as it is used as a unique identifier to fetch the results on that
image in all subsequent calls.
4. Each and every api call should have its own unique api_sig.
5. The query sends image as HTML multipart POST
generateStory(): Request metadata from an image to train into a human readable sentence. Require further training the metadata from the community...
We are currently using built-in voiceOver function on mobile and desktop devices.