Automatic method to generate product attributes based solely on product images
View Patent ↗Disclosed is a method to generate data describing a product. The method includes the steps of comparing a digitized query image of the product to digitized pre-existing product images in a pre-existing product database. The pre-existing product database is organized using a taxonomy and an ontology. The pre-existing product images are linked to a corresponding node in the taxonomy and are also linked to attribute data and attribute value data in the ontology. At least one pre-existing product image is then retrieved that most closely matches the query image based on at least one matching criterion selected in whole or in part by a user. From the pre-existing product database is extracted the node in the taxonomy, the attribute data, or the attribute value data linked to the pre-existing product image retrieved earlier. In this fashion, product data relevant to the item depicted in the query image can be generated automatically from existing product data.
1. A method to generate data describing a product, the method comprising:
in a special-purpose computer or a suitably programmed general-purpose computer:
(a) comparing a digitized query image of the product to digitized pre-existing product images in a pre-existing product database, wherein the pre-existing product database comprises a taxonomy and an ontology, and the pre-existing product images are linked to a corresponding node in the taxonomy and also linked to attribute data and attribute value data in the ontology for each pre-existing product depicted in the pre-existing product images, wherein the taxonomy is organized according to attributes and attribute values relevant to appearance of the depicted pre-existing products and attributes and attribute values having no relevance to the appearance of the depicted pre-existing products; then
(b) retrieving at least one pre-existing product image that most closely matches the query image based on at least one matching criterion selected in whole or in part by a user; and then
(c) extracting from the pre-existing product database the node in the taxonomy, the attribute data, or the attribute value data linked to the pre-existing product image retrieved in step (b).
2. The method of claim 1 , wherein step (c) comprises extracting the node in the taxonomy linked to the pre-existing product image retrieved in step (b).
3. The method of claim 2 , wherein step (a) comprises comparing the query image to all of the pre-existing product images in a pre-existing product database; and further comprising, after step (c):
(d) comparing the query image to pre-existing product images in the pre-existing product database that are indexed at the taxonomic node extracted in step (c); and then
(e) retrieving and ranking by mathematical distance from the query image a number of pre-existing product images “n” from the taxonomic node extracted in step (c); and then
(f) extracting from the “n” pre-existing product images retrieved and ranked in step (e) the attribute data and the attribute value data linked to the “n” pre-existing product images.
4. The method of claim 1 , wherein step (c) comprises extracting the attribute data linked to the pre-existing product image retrieved in step (b).
5. The method of claim 1 , wherein step (c) comprises extracting the attribute value data linked to the pre-existing product image retrieved in step (b).
6. The method of claim 1 , wherein step (c) comprises extracting the node in the taxonomy, the attribute data, and the attribute value data linked to the pre-existing product image retrieved in step (b).
7. The method of claim 1 , wherein step (b) comprises retrieving the pre-existing product image based on a matching criterion selected from the group consisting of color histogram and edge detection histogram.
8. The method of claim 1 , further comprising, after step (c):
(d) associating the node in the taxonomy, the attribute data, or the attribute value data retrieved in step (b) with the query image.
9. The method of claim 8 , further comprising, after step (d):
(e) saving the query image within the pre-existing product database at the associated node from step (d).
10. The method of claim 1 , further comprising, after step (c):
(d) associating the node in the taxonomy retrieved in step (b) with the query image.
11. The method of claim 1 , further comprising, after step (c):
(d) associating the attribute data retrieved in step (b) with the query image.
12. The method of claim 1 , further comprising, after step (c):
(d) associating the attribute value data retrieved in step (b) with the query image.
13. The method of claim 1 , further comprising, after step (c):
(d) associating the node in the taxonomy, the attribute data, and the attribute value data retrieved in step (b) with the query image.
14. The method of claim 13 , further comprising, after step (d)
(e) saving the query image and its associated attribute data and attribute value data within the pre-existing product database at the associated node from step (d).
15. The method of claim 1 , wherein the attribute data in the ontology comprise data pertaining to attributes selected from the group consisting of internal size, number of drawers, number of shelves, whether a depicted pre-existing product has an ice-maker, whether a depicted pre-existing product has a water dispenser, whether a depicted pre-existing product has a product manual, and whether a depicted pre-existing product has an extended warranty.
16. A method to generate data describing a product, the method comprising:
in a special-purpose computer or a suitably programmed general-purpose computer:
(a) comparing a digitized query image of the product to digitized pre-existing product images in a pre-existing product database, wherein the pre-existing product database comprises a taxonomy and an ontology, and the pre-existing product images are linked to a corresponding node in the taxonomy and also linked to attribute data and attribute value data in the ontology for each pre-existing product depicted in the pre-existing product images, wherein the pre-existing products depicted in the pre-existing product images are selected from the group consisting of a power tool, a hand tool, a vehicle, a padlock, a lighting fixture, a lamp, a musical instrument, a kitchen appliance, a coffee maker, a television, a mirror, furniture, a video game controller, a bathroom fixture, and a cooking tool or utensil; then
(b) retrieving at least one pre-existing product image that most closely matches the query image based on at least one matching criterion selected in whole or in part by a user; then
(c) extracting from the pre-existing product database the node in the taxonomy, the attribute data, and the attribute value data linked to the pre-existing product image retrieved in step (b); then
(d) associating the node in the taxonomy, the attribute data, and the attribute value data retrieved in step (b) with the query image; and then
(e) saving the query image and its associated attribute data and attribute value data within the pre-existing product database at the associated node from step (d).
17. The method of claim 16 , wherein step (b) comprises retrieving the pre-existing product image based on a matching criterion selected from the group consisting of color histogram and edge detection histogram.
18. The method of claim 16 , wherein the attribute data in the ontology comprise data pertaining to attributes selected from the group consisting of internal size, number of drawers, number of shelves, whether a depicted pre-existing product has an ice-maker, whether a depicted pre-existing product has a water dispenser, whether a depicted pre-existing product has a product manual, and whether a depicted pre-existing product has an extended warranty.
19. A method to generate data describing a product, the method comprising:
in a special-purpose computer or a suitably programmed general-purpose computer:
(a) comparing a digitized query image of the product to digitized pre-existing product images in a pre-existing product database, wherein the pre-existing product database comprises a taxonomy and an ontology, and the pre-existing product images are linked to a corresponding node in the taxonomy and also linked to attribute data and attribute value data in the ontology for each pre-existing product depicted in the pre-existing product images, wherein the taxonomy is organized according to attributes and attribute values relevant to appearance of the depicted pre-existing products and attributes and attribute values having no relevance to the appearance of the depicted pre-existing products; then
(b) retrieving at least one pre-existing product image that most closely matches the query image based on at least one matching criterion selected in whole or in part by a user; then
(c) extracting from the pre-existing product database the node in the taxonomy linked to the pre-existing product image retrieved in step (b); then
(d) comparing the query image to pre-existing product images in the pre-existing product database that are indexed at the taxonomic node extracted in step (c); then
(e) retrieving and ranking by mathematical distance from the query image a number of pre-existing product images “n” from the taxonomic node extracted in step (c); and then
(f) extracting from the “n” pre-existing product images retrieved and ranked in step (e) the attribute data and the attribute value data linked to the “n” pre-existing product images.
20. The method of claim 19 , further comprising, after step (f):
(g) associating the node in the taxonomy, the attribute data, and the attribute value data retrieved in step (b) with the query image; and then
(h) saving the query image and its associated attribute data and attribute value data within the pre-existing product database at the associated node from step (g).