Item validation based on visually imperceptible waterpark decoding
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for detecting and decoding a visually imperceptible or perceptible watermark. A watermark detection apparatus determines whether the particular image includes a visually imperceptible or perceptible watermark using detector a machine learning model. If the watermark detection apparatus detects a watermark, the particular image is routed to a watermark decoder. If the watermark detection apparatus cannot detect a watermark in the particular image, the particular image is filtered from further processing. The watermark decoder decodes the visually imperceptible or perceptible watermark detected in the particular image. After decoding, an item depicted in the particular image is validated based data extracted from the decoded visually imperceptible or perceptible watermark.
1. A computer-implemented method, comprising:
receiving, by a watermark detection apparatus, images;
for each particular image among the images:
determining, by the watermark detection apparatus, whether the particular image includes a visually imperceptible watermark using a detector machine learning model, wherein the watermark detection apparatus detects the visually imperceptible watermark in at least one of the images;
routing the particular image based on the determination whether the particular image includes the visually imperceptible watermark, including:
routing the particular image to a watermark decoder in response to the watermark detection apparatus detecting the visually imperceptible watermark in the particular image; and
filtering the particular image from further processing in response to the watermark detection apparatus not detecting the visually imperceptible watermark in the particular image;
decoding, by the watermark decoder, the visually imperceptible watermark detected in the particular image that was routed to the watermark decoder; and
validating an item depicted in the particular image based on data extracted from the decoded visually imperceptible watermark.
2. The computer-implemented method of claim 1 , wherein determining whether the particular image includes a visually imperceptible watermark using a detector machine learning model comprises determining whether the particular image includes the visually imperceptible watermark using a detector machine learning model trained to determine whether a region of an input image includes the visually imperceptible watermark based on a coarse analysis of the region that does not require a pixel by pixel analysis of the region.
3. The computer-implemented method of claim 2 , wherein determining whether the particular image includes the visually imperceptible watermark comprises:
determining, by the watermark detection apparatus, a set of encoded pixels in the region of the particular image; and
classifying the region of the particular image based on the set of encoded pixels in the region of the particular image, including:
classifying the region as a watermarked region in response to the set of encoded pixels meeting a watermark condition; and
classifying the region as a not watermarked region in response to the set of encoded pixels not meeting the watermark condition.
4. The computer-implemented method of claim 3 , further comprising:
outputting binary 1s for all pixels in the region to mark the region as including the visually imperceptible watermark; and
generating, using the binary 1s, a watermark map for the image, wherein the watermark map indicates regions of the map that contain the visually imperceptible watermark.
5. The computer-implemented method of claim 1 , wherein decoding the watermark comprises decoding the watermark using a decoder machine learning model trained to visually imperceptible watermarks in the particular image irrespective of a zoom level of an item depicted in the image.
6. The computer-implemented method of claim 5 , further comprising training the decoder machine learning model using a set of training images that includes multiple training images that depict watermarked items at different levels of zoom and with different image distortions.
7. The computer-implemented method of claim 6 , further comprising pre-processing the set of training images to prevent model performance deficiencies caused by training the decoder machine learning model in floating point numbers using images that are represented by RGB unsigned integers.
8. A system comprising:
a data storage device;
a watermark detection apparatus, including one or more processors, configured to perform operations comprising:
determining whether a particular image includes a visually imperceptible watermark using detector a machine learning model, wherein the watermark detection apparatus detects the visually imperceptible watermark in at least one of the images; and
routing the particular image based on the determination whether the particular image includes the visually imperceptible watermark, including:
routing the particular image to a watermark decoder in response to the watermark detection apparatus detecting the visually imperceptible watermark in the particular image; and
filtering the particular image from further processing in response to the watermark detection apparatus not detecting the visually imperceptible watermark in the particular image; and
a watermark decoder, including one or more processors, configured to perform operations comprising decoding, by the watermark decoder, the visually imperceptible watermark detected in the particular image that was routed to the watermark decoder, wherein the system is configured to perform operations comprising validating an item depicted in the particular image based on data extracted from the decoded visually imperceptible watermark.
9. The system of claim 8 , wherein determining whether the particular image includes a visually imperceptible watermark using a detector machine learning model comprises determining whether the particular image includes the visually imperceptible watermark using a detector machine learning model trained to determine whether a region of an input image includes the visually imperceptible watermark based on a coarse analysis of the region that does not require a pixel by pixel analysis of the region.
10. The system of claim 9 , wherein determining whether the particular image includes the visually imperceptible watermark comprises:
determining, by the watermark detection apparatus, a set of encoded pixels in the region of the particular image; and
classifying the region of the particular image based on the set of encoded pixels in the region of the particular image, including:
classifying the region as a watermarked region in response to the set of encoded pixels meeting a watermark condition; and
classifying the region as a not watermarked region in response to the set of encoded pixels not meeting the watermark condition.
11. The system of claim 10 , further comprising:
outputting binary 1s for all pixels in the region to mark the region as including the visually imperceptible watermark; and
generating, using the binary 1s, a watermark map for the image, wherein the watermark map indicates regions of the map that contain the visually imperceptible watermark.
12. The system of claim 8 , wherein decoding the watermark comprises decoding the watermark using a decoder machine learning model trained to visually imperceptible watermarks in the particular image irrespective of a zoom level of an item depicted in the particular image.
13. The system of claim 12 , further comprising training the decoder machine learning model using a set of training images that includes multiple training images that depict watermarked items at different levels of zoom and with different image distortions.
14. The system of claim 13 , further comprising pre-processing the set of training images to prevent model performance deficiencies caused by training the decoder machine learning model in floating point numbers using images that are represented by RGB unsigned integers.
15. A non-transitory computer readable medium storing instructions that, when executed by one or more data processing apparatus, cause the one or more data processing apparatus to perform operations comprising:
receiving, by a watermark detection apparatus, images;
for each particular image among the images:
determining, by the watermark detection apparatus, whether the particular image includes a visually imperceptible watermark using a detector machine learning model, wherein the watermark detection apparatus detects the visually imperceptible watermark in at least one of the images;
routing the particular image based on the determination whether the particular image includes the visually imperceptible watermark, including:
routing the particular image to a watermark decoder in response to the watermark detection apparatus detecting the visually imperceptible watermark in the particular image; and
filtering the particular image from further processing in response to the watermark detection apparatus not detecting the visually imperceptible watermark in the particular image;
decoding, by the watermark decoder, the visually imperceptible watermark detected in the particular image that was routed to the watermark decoder; and
validating an item depicted in the particular image based on data extracted from the decoded visually imperceptible watermark.
16. The non-transitory computer readable medium of claim 15 , wherein determining whether the particular image includes a visually imperceptible watermark using a detector machine learning model comprises determining whether the particular image includes the visually imperceptible watermark using a detector machine learning model trained to determine whether a region of an input image includes the visually imperceptible watermark based on a coarse analysis of the region that does not require a pixel by pixel analysis of the region.
17. The non-transitory computer readable medium of claim 16 , wherein determining whether the particular image includes the visually imperceptible watermark comprises:
determining, by the watermark detection apparatus, a set of encoded pixels in the region of the particular image; and
classifying the region of the particular image based on the set of encoded pixels in the region of the particular image, including:
classifying the region as a watermarked region in response to the set of encoded pixels meeting a watermark condition; and
classifying the region as a not watermarked region in response to the set of encoded pixels not meeting the watermark condition.
18. The non-transitory computer readable medium of claim 17 , further comprising:
outputting binary 1s for all pixels in the region to mark the region as including the visually imperceptible watermark; and
generating, using the binary 1s, a watermark map for the image, wherein the watermark map indicates regions of the map that contain the visually imperceptible watermark.
19. The non-transitory computer readable medium of claim 15 , wherein decoding the watermark comprises decoding the watermark using a decoder machine learning model trained to visually imperceptible watermarks in the particular image irrespective of a zoom level of an item depicted in the image.
20. The non-transitory computer readable medium of claim 19 , further comprising training the decoder machine learning model using a set of training images that includes multiple training images that depict watermarked items at different levels of zoom and with different image distortions.
21. The non-transitory computer readable medium of claim 19 , further comprising pre-processing the set of training images to prevent model performance deficiencies caused by training the decoder machine learning model in floating point numbers using images that are represented by RGB unsigned integers.