Systems and methods for redaction of screens
Inventors
Nguyen, Anh Tuan • Spitzer-Williams, Noah • HERSHFIELD, Jacob
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11930293-B2
Publication Date
2024-03-12
Expiration Date
Abstract
Systems, devices, and methods redact one or more light-emitting screens in data recorded on a recording device. The redaction may include receiving recorded data comprising a plurality of pixel values. The redaction may include detecting one or more light-emitting screens in the received image. The redaction may include redacting a subset of the pixel values from the recorded associated with the one or more detected light-emitting screens. The redaction may be commonly applied to multiple frames of recorded data through the use of a unique identifier assigned to a same detected light-emitting screen. A same light-emitting screen may be tracked across multiple non-sequential or sequential frames and assigned a same unique identifier.
Core Innovation
The invention relates to redacting light-emitting screens in image or video data recorded at an incident. A redaction computing device detects a light-emitting screen of a display included in the recorded image or captured in a received image, and generates information for redaction associated with the detected light-emitting screen. Pixels in the image associated with the detected light-emitting screen are redacted to extract redacted media comprising a redacted image.
The disclosure coordinates redaction across images and frames by transmitting an indication of the detected light-emitting screen to a client computing device and displaying a redaction mask associated with the detected light-emitting screen. The client transmits an instruction associated with the displayed redaction mask, and extracting the redacted media is performed in accordance with the transmitted instruction. A boundary between a light-emitting screen and a non-light-emitting bezel is annotated for training a detection engine, and metadata indicative of the light-emitting screen is generated, including one or more of a boundary, pixel location, or dimension.
The invention supports consistent identification of the light-emitting screen across sequential or non-sequential frames by tracking the same screen and assigning a unique track identifier for enforcement of consistent redaction over time. Mask segments are defined over a timeline and may be edited, with mask sizing that the mask area is larger than the detected screen region. Extracting redacted media includes redacting a subset of pixel values associated with the detected light-emitting screen and the mask, using a redaction filter to masked pixel subsets.
Claims Coverage
The document includes three independent claims (system, computer-implemented method, and non-transitory computer-readable medium). Across these independent claims, the inventive coverage centers on detecting a light-emitting screen in incident image/video, generating an associated mask and redacting pixels associated with the detected screen, coordinating mask-based extraction via transmitted indication/instruction, and using trained detection based on annotated boundaries between a light-emitting screen and a non-light-emitting bezel plus metadata generation.
Incident image system for redacting detected light-emitting screens
A system comprising a redaction computing device configured to detect a light-emitting screen in an image recorded at an incident, transmitting an indication of the detected light-emitting screen over a network, and extracting redacted media comprising a redacted image in which pixels associated with the detected light-emitting screen are redacted; and a client computing device configured to receive the indication, display a redaction mask associated with the detected light-emitting screen, and transmit an instruction associated with the displayed redaction mask, wherein extracting the redacted media is in accordance with the transmitted instruction.
Detection engine trained with annotated boundaries between screen and bezel
The detection engine is trained using training data comprising one or more images in which a boundary between a light-emitting screen and a non-light-emitting bezel of a display included in each of the one or more images has been annotated.
Method for concurrently detecting screen subsets across different video frames
A computer-implemented method comprising receiving an image of video data comprising a plurality of pixel values, detecting a light-emitting screen of a display captured in the received image and generating a mask associated with the detected light-emitting screen, redacting a subset of the pixel values associated with the detected light-emitting screen and the mask, providing an indication of the subset of the pixel values for display, and identifying another subset of the pixel values associated with the light-emitting screen in a second image of the video data, wherein the received image is different from the second image.
Mask shape corresponds to detected screen shape for pixel subset redaction
Redacting a subset of the pixel values associated with the detected light-emitting screen and the mask, wherein a shape of the mask corresponds to a shape of the detected light-emitting screen.
Non-transitory medium using neural network detection and metadata for redaction
A non-transitory computer-readable medium comprising computer-executable instructions that cause a processing circuit to detect a light-emitting screen of a display captured in an image of video data, generating a mask associated with the light-emitting screen, and extracting redacted media by redacting the image in accordance with the mask to provide a redacted image, wherein detecting the light-emitting screen comprises applying the video data to a neural network previously trained on separate images each comprising one or more annotated light-emitting screens each surrounded on one or more sides via one or more respective non-light-emitting bezels, and detecting the light-emitting screen comprises generating metadata indicative of the light-emitting screen including one or more of a boundary, pixel location, or dimension that indicates a set of pixel values of the image that represent the light-emitting screen.
Across the independent claims, the core coverage is the combination of detecting a light-emitting screen and identifying pixels representing it, generating and using an associated redaction mask to redact pixel subsets and extract redacted media, coordinating display of a redaction mask and applying an instruction tied to that mask, and improving detection by training with annotated boundaries between a light-emitting screen and a non-light-emitting bezel and generating metadata indicative of the light-emitting screen using a neural network approach.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?