4

Sorry, i use the word "good"...it's relative. But, in my project i want to find pictures, which where not under or overexposed and have a great dynamic with a lot of colors.

okay :

  • for the first problems i can take a V-histogram in the HSV-colorspace with a rude count of bins. So i can detect under/overexposed images
  • for the count of colors i can use a H/V-Diagram, maybe in a cumulative histogram?
  • For the dynamic (difference between the rate of colors) i can choose the H-Diagram, and calculate the avg of the bins.
  • Also, i want to detect global peaks. Where a color is erodes. Sure, I can get the maximum of a V-Histogram and look if the value/brightness is 255.

These points presents my ideas to solve the problems. But maybe there are other, effective solutions or ideas from any papers.

In my project i must explore a large count of image and i think it take a long time to create all this histograms.

Greetings

Abid Rahman K
  • 51,886
  • 31
  • 146
  • 157
501 - not implemented
  • 2,638
  • 4
  • 39
  • 74

1 Answers1

1

Try Color quantalization - I'm sure it should help you. Look at this StackOverflow discussion.

Community
  • 1
  • 1
ArtemStorozhuk
  • 8,715
  • 4
  • 35
  • 53
  • jeah, i quantizie it with the count of bins at the histogram to get a better performance. Or what are you think? – 501 - not implemented Jul 20 '12 at 11:48
  • @destiny I don't get your question :) You have to: 1) Convert image to Lab color space because in this color space Euclidain distance makes sence. 2) Do clusterization (see second link). 3) Compare colors in different color clusters. If distance is big between all of them than you have many different colors on image. Actally all this info is written in that wiki (first) link. – ArtemStorozhuk Jul 20 '12 at 13:58