Neural-network-based prediction of the stopping moment for text recognition in a video stream
AssigneeSmart Engines Service, LLC
InventorsKseniya Stanislavovna KRYUCHKOVA, Aleksandr Vladimirovich SHESHKUS, Konstantin Bulatovich BULATOV
Life of this patent
Estimating the term…
Abstract
When performing text recognition on a video stream, it is beneficial to stop the recognition process as soon as an acceptable recognition result is obtained. In an embodiment, a method for determining when to stop the recognition process method for a text field utilizes a predictive model, such as a neural network, to classify a sequence of calculated errors between text-recognition results for a set of successive frames in the video stream. The classification is used to determine whether to stop the recognition process at the current frame, or to continue the recognition process on the next frame.
Claims20
A method comprising using at least one hardware processor to, for each of one or more fields in a plurality of frames from a video stream for which a recognition process is being performed:
- perform text recognition on the field to produce a current text-recognition result;
- combine the current text-recognition result with a preceding text-recognition result for the field from one or more prior frames;
- calculate an error between the combined text-recognition result and the preceding text-recognition result;
- add the calculated error to an accumulated feature set;
- and,
- when the accumulated feature set has reached a predefined size, wherein the predefined size is greater than one,
- apply a predictive model to the accumulated feature set of the predefined size to classify the accumulated feature set into one of a plurality of classes, wherein the plurality of classes comprises a first class that indicates continuation of the recognition process, and a second class that indicates stopping of the recognition process,
- when the one class is the first class, continue the recognition process for the field, and,
- when the one class is the second class, stop the recognition process for the field.
A system comprising:
- at least one hardware processor;
- and
- one or more software modules that are configured to, when executed by the at least one hardware processor,
- perform text recognition on the field to produce a current text-recognition result,
- combine the current text-recognition result with a preceding text-recognition result for the field from one or more prior frames,
- calculate an error between the combined text-recognition result and the preceding text-recognition result,
- add the calculated error to an accumulated feature set, and,
- when the accumulated feature set has reached a predefined size, wherein the predefined size is greater than one,
- apply a predictive model to the accumulated feature set of the predefined size to classify the accumulated feature set into one of a plurality of classes, wherein the plurality of classes comprises a first class that indicates continuation of the recognition process, and a second class that indicates stopping of the recognition process,
- when the one class is the first class, continue the recognition process for the field, and,
- when the one class is the second class, stop the recognition process for the field.
A non-transitory computer-readable medium having instructions stored therein, wherein the instructions, when executed by a processor, cause the processor to:
- perform text recognition on the field to produce a current text-recognition result;
- combine the current text-recognition result with a preceding text-recognition result for the field from one or more prior frames;
- calculate an error between the combined text-recognition result and the preceding text-recognition result;
- add the calculated error to an accumulated feature set;
- and,
- when the accumulated feature set has reached a predefined size, wherein the predefined size is greater than one,
- apply a predictive model to the accumulated feature set of the predefined size to classify the accumulated feature set into one of a plurality of classes, wherein the plurality of classes comprises a first class that indicates continuation of the recognition process, and a second class that indicates stopping of the recognition process,
- when the one class is the first class, continue the recognition process for the field, and,
- when the one class is the second class, stop the recognition process for the field.
Dependencies are inferred from the claim text.