Once a query image arrives at the API, it is analyzed and compared to the reference object collection, allowing returning recognition results.
When taking a query image, try to fit the object (or objects) into the frame as accurately as possible, although the recognition technology tolerates a certain level of imperfection here. It is crucial to avoid blurring, overexposure, darkness, and strong light reflections. Using flash lamp is not recommended. The API accepts query images in JPEG format.
The limitations for query pictures depend on the recognition mode:
|Recognition mode||Max file size||Min edge length||Min image surface||Max image surface|
|Single||2 MiB||100 pix||0.05 Mipx||0.50 Mipx|
|Multi||5 MiB||100 pix||0.10 Mipx||6.00 Mipx|
As the query pictures can contain several objects for some recognition modes, the limitations of query pictures apply to the pictures themselves and not to the objects they represent. The most general rule that can be successfully applied for every recognition mode says that a single object in the query picture should not have any dimension shorter than about 200-300 pixels. This allows you to estimate how big your query image should be to contain the expected number of objects. In other words, if you wish to fit more objects, your should increase the size of your query image to meet the above-mentoned rule for every single object contained in the image.
Before you are able to perform a recognition query, you need to acquire an image to act as your recognition request data. There are two general ways to obtain query images:
You can use your querying application to take the images right before sending them in your recognition requests. It is the best way to query from a mobile device or other equipment having a camera. This method is useful when you need to query the API with "live" images.
Recognition queries can be performed using previously-stored or already existing images. This is the best way to query from devices that do not have a camera or does not have to use it, like server applications, video stream analyzers, or web crawlers.
To execute a recognition query, you have to use a device or software which meets the following criteria: