5

I market a product with color #2A4095, and want to give it a product name using the name of an object that has the same color.

So, I want to find what objects (flowers, fruits, cloth, human objects) have color #2A4095 approximately.

The pictures could come from the Flickr set. The search engine's algorithm has to remove background/shade/reflections to estimate each object's actual color. It's rather difficult because several photography styles results in various kinds of color alterations.

For instance this should come for #d85133, not grey nor black nor white: Fuji apple


For this purpose I use Multicolr, but it is far from perfect:

  • It just takes the average color of a picture. A better tool would:
    • only process pictures that have a main object (ex: no drawings or landscape pictures)
    • detect the main object's contour, and don't take the color of the background into account.
    • don't take the object's shade and light reflection into account.
  • It does not allow to search for #2A4095, only a few colors are available

Multicolr color search

  • Have you tried PicItUp tools? – user416 Dec 1 '14 at 14:19
  • Or Palette? I noticed that searching for search pictures with color finds candidates – user416 Dec 1 '14 at 14:22
  • @JanDoggen: Palette has the exact same problems as Multicolr, see above. – Nicolas Raoul Dec 1 '14 at 14:23
  • @JanDoggen: Sorry my PicItUp comment was mistaken, removed. I tried the PicItUp tools too, none is a good solution: "Multicolor Flickr search" is the closest they have and it too has the exact same problems as Multicolr. – Nicolas Raoul Dec 1 '14 at 14:27
  • It looks like navy blue to me (colorhexa.com/2a4095) or not far from it. Why not just settle for searching for navy blue things? It seems much easier. – Mawg says reinstate Monica Jul 30 '15 at 9:01
1

You could use python to generate a histogram or each image with buckets of a given size, (the colour tolerance), about your desired colour then if the image contains more than a given percentage of the desired colour offer it up as a suggestion.

There are a number of python libraries that can create a histogram of an image, the bucket size would go some way to reducing the false rejects due to the style of photography and the threshold should try to reject images where a tiny area of the image matches.

You could possibly combine this with the python scapy library to mine images from a given site.

|improve this answer|||||

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.