Image processing system for deep fashion color recognition
A system and method are disclosed for image processing of one or more items in an inventory of one or more supply chain entities. Embodiments include receiving an initial set of images of at least two items in the inventory, identifying color distributions from the initial set of images using two encoders, and grouping colors of the at least two items based on similarities of the identified color distributions using a color coding model.
1. A system, comprising:
an image processing system comprising a server and configured to:
receive an initial set of images of at least two items;
identify color distributions from the initial set of images using two encoders;
group colors of the at least two items based on similarities of the identified color distributions using a color coding model;
receive an image of at least one additional item;
identify a color distribution from the image of the at least one additional item using the two encoders;
assign a color to the at least one additional item based, at least in part, on the identified color distribution from the image of the least one additional item and the grouped colors of the at least two items; and
modify a color attribute value associated with the at least one additional item based, at least in part, on the assigned color and a uniform color code.
2. The system of claim 1 , wherein the image processing system preprocesses an image of an item by:
localizing an item in an image by a bounding box detection algorithm;
resizing the image to a uniform size; and
applying an image mask.
3. The system of claim 1 , wherein the two encoders comprise at least one of a box encoder and an iterative cell encoder.
4. The system of claim 1 , wherein the image processing system is further configured to:
identify customer segments based, at least in part, on the modified color attribute value and purchase history of one or more customers.
5. A method, comprising:
receiving, by an image processing system comprising a server, an initial set of images of at least two items;
identifying, by the image processing system, color distributions from the initial set of images using two encoders;
grouping, by the image processing system, colors of the at least two items based on similarities of the identified color distributions using a color coding model;
receiving, by the image processing system, an image of at least one additional item;
identifying, by the image processing system, a color distribution from the image of the at least one additional item using the two encoders;
assigning, by the image processing system, a color to the at least one additional item based, at least in part, on the identified color distribution from the image of the least one additional item and the grouped colors of the at least two items; and
modifying, by the image processing system, a color attribute value associated with the at least one additional item based, at least in part, on the assigned color and a uniform color code.
6. The method of claim 5 , further comprising:
preprocessing an image of an item by:
localizing, by the image processing system, an item in an image by a bounding box detection algorithm;
resizing, by the image processing system, the image to a uniform size; and
applying, by the image processing system, an image mask.
7. The method of claim 5 , wherein the two encoders comprise at least one of a box encoder and an iterative cell encoder.
8. The method of claim 7 , wherein the box encoder extracts a histogram of color distributions from non-overlapping boxes that extend from the center of an image and grow larger to the perimeter of the image and concatenation of box features define a feature vector.
9. The method of claim 5 , further comprising:
identifying, by the image processing system, customer segments based, at least in part, on the modified color attribute value and purchase history of one or more customers.
10. A non-tangible computer-readable medium embodied with software, the software when executed:
receives an initial set of images of at least two items;
identifies color distributions from the initial set of images using two encoders;
groups colors of the at least two items based on similarities of the identified color distributions using a color coding model;
receives an image of at least one additional item;
identifies a color distribution from the image of the at least one additional item using the two encoders;
assigns a color to the at least one additional item based, at least in part, on the identified color distribution from the image of the least one additional item and the grouped colors of the at least two items; and
modifies a color attribute value associated with the at least one additional item based, at least in part, on the assigned color and a uniform color code.
11. The non-tangible computer-readable medium of claim 10 , wherein the software when executed is configured to preprocess an image of an item by:
localizing an item in an image by a bounding box detection algorithm;
resizing the image to a uniform size; and
applying an image mask.
12. The non-tangible computer-readable medium of claim 10 , wherein the two encoders comprise at least one of a box encoder and an iterative cell encoder.
13. The non-tangible computer-readable medium of claim 12 , wherein the box encoder extracts a histogram of color distributions from non-overlapping boxes that extend from the center of an image and grow larger to the perimeter of the image and concatenation of box features define a feature vector.
14. The non-tangible computer-readable medium of claim 10 , wherein the software when executed further:
identifies customer segments based, at least in part, on the modified color attribute value and purchase history of one or more customers.