I would go with analyzing the images programmatically, and extracting the data into something we can query.
Relevant Skills and Experience
I did programmatic image creation/manipulation many times before, I believe we should use HSL and not RGB, or any other color standard.
Proposed Milestones
$100 USD - initial
$150 USD - prior to handing over the code, after testing on my server
I suggest the analysis for each image to go as follows:
1) extract the average hue/value/saturation of the whole image
2) split the image in 3x3 grid, for each section extract the average h/s/v, these would be big sections
3) split the image in 100x100 grid, for each section extract average h/s/v, these would be small sections
We would have two tables in the DB, in which we would populate the data extracted via above method:
images
image_sections
Each big and small section would have an unique identifier, perhaps in form of [col]-[row].
Once all images have been analyzed (I can make the script loop through all files in a given folder), we should be able to query this data in a way which should give very accurate results.
I won't be able to go into more details here as I'm limited to 1000 chars in this form, but I hope to hear from you and we can discuss more.
Regards,
Julijan