0

I want to know more about selective selection of a part of an image but there are just a few papers related to this in internet. It is strange that I can not find nothing good on Selective Coloring of images. I am trying to develope something about this but I am stacked and I can not continue. I`ve tried something but it is not working nothing. These days I am trying to follow these steps: Change pixels color

but I don`t know how to do those steps exactly

Community
  • 1
  • 1
Drill
  • 169
  • 2
  • 3
  • 12
  • One of the reasons you might not be getting good results on your searches is because the term(s) you are searching for are meaningless. I myself have never seen this problem described as you did here. You are looking for ways to cluster and divide colors, one relatively common way to do this is using the k-means algorithm. There are a lot of papers on this (too many to read), and the main problem is to define the distance between colors. – mmgp Dec 17 '12 at 19:29
  • @mmgp Well we can say it is a clustering issue and yes the main problem is the distance between colors. Anyway I am trying to calculate standard deviation of selected part from user and then I am iterating throw every pixel and I check for every pixel: if hue and saturation are smaller then standard deviation of selected part from user then I am trying to recolor it with a target color, otherwise I skip it and do nothing. I am not getting good results with this unfortunately :S – Drill Dec 18 '12 at 10:10
  • You need to better understand the problem. Your ad-hoc solution is not going to cut it. Did you google for k-means for finding k-dominant colors yet ? – mmgp Dec 18 '12 at 13:18
  • @mmgp no I didn`t but I think this clustering problem can be done by Segmentation of the image, can it? I think I understand the problem. Look I am trying to do this http://stackoverflow.com/questions/13325559/change-pixels-color but I have some problems with moving coordinates to target value.I can not find appropriate papers in recoloring of images or selective color of image. In the link I just put here I have described the problem and maybe from there you can suggest me what to do. I am not requiring the solution but just the fields where I should study to achieve this. Thank you indeed. – Drill Dec 18 '12 at 13:42
  • Image segmentation is hard, are you aware of that ? Also, clustering colors is a way to segment too. I don't understand at all why you are fully ignoring what I'm telling you. – mmgp Dec 18 '12 at 13:46
  • @mmgp I am not ignorig but I thought image segmentation is easier then clustering. If you are saying me that it is easier with clustering then I can use any algorithm for clustering colors of an image and based on this I will be able to distinguish colors, if you say so. – Drill Dec 18 '12 at 13:48
  • Clustering is so much easier... because all you have to do is cluster according to some criteria and it is perfectly clustered. Now, image segmentation expects a final output that is hardly met by simple criteria. This doesn't mean segmentation is slower than clustering, commonly it is the contrary. – mmgp Dec 18 '12 at 14:05
  • @mmgp yes right segmentation is faster (I know some basics in image processing) but for know I don`t care for performance. I just want to achieve this, or at least near to this issue and then I can work on increasing performance. So you are suggesting me first to cluster colors and after to try moving the selected color from user to the target color based on the distance between colors? Bthw did you maybe read the link I put in previous comment? Thank you for reading and trying to help me :) – Drill Dec 18 '12 at 14:10
  • I quickly read it, and it is going in the wrong way too (the answer present there is off the track too). Google for dominant colors using k-means. – mmgp Dec 18 '12 at 14:14
  • @mmgp Ok no problem. I more was interested in you to read that post just for understanding the problem and see if you are suggesting me in the right way because as you said it seems that the answer is not sending me to an appropriate solution. ok thank you again and I will continue now reading in clustering. I hope I will be able to do something on this. – Drill Dec 18 '12 at 14:31
  • @mmgp sorry to write again but it seems that K-mean needs to know in advance number of clusters (number of colors in the image) and of course in my case the number of clusters should be estimated automatically...I think it is difficult to guess K because it is dependet on the image which I am processing... – Drill Dec 18 '12 at 14:41
  • @mmgp Ok you are right, sorry – Drill Dec 18 '12 at 14:53
  • @mmgp I just finished clustering with K-Means and OK clustering is helping in distinguishing colors between them but the probles is that it works just with images that contains objects that have a large diffenrence of colors between them. I tried it with HSV and RGB(it is working better). – Drill Dec 20 '12 at 17:40
  • @mmgp I have to work with images which contains textures like carpits for example and there it is difficult to achieve recoloring with K-Mean. Instead I am thinking to do texture segmentation and then to see what is it going to happen :S. Thank you very much. I learned a new thing(clustering of colors with K-Mean algorithm and this is great). If you have any idea on helping me I appreciate it, otherwise thankyou indeed for this help. – Drill Dec 20 '12 at 17:41
  • There are many ways to go through texture classification for segmentation, none are perfect. You need to post more details. – mmgp Dec 20 '12 at 17:45
  • @mmgp Well here is an example of a possible image - a carpet http://www.mediafire.com/view/?wccepff3jkbgu4j. Here if a user selects a part of the image and then he also selects another color in a color picker then I should try to recolor in all the image the parts that are same as the selected part. In the sample image, if he selects the circle part then as you can see there are a mixture of colors and clustering of colors will not work. Instead I should try something like segmenting image based on texture and then trying to recolor some parts. – Drill Dec 20 '12 at 18:34
  • It seems like you could first apply k-means in the selected circle, so you determine its dominant color. – mmgp Dec 20 '12 at 18:46
  • @mmgp ...and then chage this color to a new one (the color that the user have selected from color picker)? I can do this with that circle but what about the others. I was thinking to determine similar shapes based on color distribution...anyway as I can see I need to make a hard research these days and I will try whatever people suggest and whatever I think it will make the issue solved. Thnx for all man. A you can see I need just some ideas because I am new in image processing, and for implementation i dont have problems – Drill Dec 20 '12 at 19:02
  • possible duplicate of [Selective Color of image](http://stackoverflow.com/questions/13845973/selective-color-of-image) – mmgp Jan 29 '13 at 17:33

0 Answers0