US 12,190,563 B2
Convolutional machine learning models using multiple tiling mechanisms
Rama Krishna Singh, Greater Noida (IN); Ravi Pande, Noida (IN); and Priyank Jain, Noida (IN)
Assigned to Optum, Inc., Minnetonka, MN (US)
Filed by Optum, Inc., Minnetonka, MN (US)
Filed on Jul. 13, 2022, as Appl. No. 17/812,308.
Prior Publication US 2024/0020946 A1, Jan. 18, 2024
Int. Cl. G06V 10/764 (2022.01); G06V 10/10 (2022.01); G06V 10/22 (2022.01); G06V 10/56 (2022.01)
CPC G06V 10/764 (2022.01) [G06V 10/16 (2022.01); G06V 10/22 (2022.01); G06V 10/56 (2022.01); G06V 2201/03 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method for generating a predictive output for a prediction input data object that comprises an input image comprising a plurality of image regions, the computer-implemented method comprising:
for each image region in the plurality of image regions, generating, by one or more processors, a color distribution ratio set that comprises: (i) a whitespace presence ratio for the image region, and (ii) one or more non-whitespace color range presence ratios for the image region with respect to one or more non-whitespace color ranges;
generating, by the one or more processors, a plurality of tiled images for the input image, wherein: (i) each tiled image in the plurality of tiled images is associated with a respective image tiling mechanism of a plurality of image tiling mechanisms, (ii) each tiled image in the plurality of tiled images is generated by selecting a respective N-sized tiled region subset of the plurality of image regions in accordance with the respective image tiling mechanism for the tiled image, (iii) Nis a region selection count hyper-parameter that is shared across the plurality of image tiling mechanisms, and (iv) the plurality of image tiling mechanisms comprises a greedy thresholding tiling mechanism that is configured to generate the respective N-sized tiled region subset by selecting, from an occurrence-based region sequence of the plurality of image regions that is ordered in accordance with each region position indicator for the plurality of image regions, first-occurring N image regions whose color distribution ratio sets satisfy one or more color distribution ratio thresholds;
generating, by the one or more processors, a composite-tiled image for the input image that comprises the plurality of tiled images;
generating, by the one or more processors and using a composite-tiled image prediction machine learning model, and based at least in part on the composite-tiled image, a composite-tiled image embedding for the input image;
generating, by the one or more processors, the predictive output based at least in part on the composite-tiled image embedding; and
performing, by the one or more processors, one or more prediction-based actions based at least in part on the predictive output.
 
8. A system for generating a predictive output for a prediction input data object that comprises an input image comprising a plurality of image regions, the system comprising one or more processors and memory including program code, the memory and the program code configured to, with the one or more processors, cause the system to at least:
for each image region in the plurality of image regions, generate a color distribution ratio set that comprises: (i) a whitespace presence ratio for the image region, and (ii) one or more non- whitespace color range presence ratios for the image region with respect to one or more non- whitespace color ranges;
generate a plurality of tiled images for the input image, wherein: (i) each tiled image in the plurality of tiled images is associated with a respective image tiling mechanism of a plurality of image tiling mechanisms, (ii) each tiled image in the plurality of tiled images is generated by selecting a respective N-sized tiled region subset of the plurality of image regions in accordance with the respective image tiling mechanism for the tiled image, (iii) Nis a region selection count hyper-parameter that is shared across the plurality of image tiling mechanisms, and (iv) the plurality of image tiling mechanisms comprises a greedy thresholding tiling mechanism that is configured to generate the respective N-sized tiled region subset by selecting, from an occurrence-based region sequence of the plurality of image regions that is ordered in accordance with each region position indicator for the plurality of image regions, first-occurring N image regions whose color distribution ratio sets satisfy one or more color distribution ratio thresholds;
generate a composite-tiled image for the input image that comprises the plurality of tiled images;
generate, using a composite-tiled image prediction machine learning model, and based at least in part on the composite-tiled image, a composite-tiled image embedding for the input image;
generate the predictive output based at least in part on the composite-tiled image embedding; and
perform one or more prediction-based actions based at least in part on the predictive output.
 
15. A computer program product for generating a predictive output for a prediction input data object that comprises an input image comprising a plurality of image regions, the computer program product comprising at least one non-transitory computer readable storage medium having computer-readable program code portions stored therein, the computer- readable program code portions configured to:
for each image region in the plurality of image regions, generate a color distribution ratio set that comprises: (i) a whitespace presence ratio for the image region, and (ii) one or more non- whitespace color range presence ratios for the image region with respect to one or more non- whitespace color ranges;
generate a plurality of tiled images for the input image, wherein: (i) each tiled image in the plurality of tiled images is associated with a respective image tiling mechanism of a plurality of image tiling mechanisms, (ii) each tiled image in the plurality of tiled images is generated by selecting a respective N-sized tiled region subset of the plurality of image regions in accordance with the respective image tiling mechanism for the tiled image, (iii) Nis a region selection count hyper-parameter that is shared across the plurality of image tiling mechanisms, and (iv) the plurality of image tiling mechanisms comprises a greedy thresholding tiling mechanism that is configured to generate the respective N-sized tiled region subset by selecting, from an occurrence-based region sequence of the plurality of image regions that is ordered in accordance with each region position indicator for the plurality of image regions, first-occurring N image regions whose color distribution ratio sets satisfy one or more color distribution ratio thresholds;
generate a composite-tiled image for the input image that comprises the plurality of tiled images;
generate, using a composite-tiled image prediction machine learning model, and based at least in part on the composite-tiled image, a composite-tiled image embedding for the input image;
generate the predictive output based at least in part on the composite-tiled image embedding; and
perform one or more prediction-based actions based at least in part on the predictive output.