I have a variable "x" that takes real values between 1 to 6. I have 6 data points in this interval which each of them corresponds to an image. I read the histogram of the images for finding out the changes in each data point. I want to find a relationship between the histograms and the variable x. So that later on I can estimate (and interpolate) the value of the variable "x" by analyzing a new image. Any suggestions on reading material and/or algorithm?
I don't fully understand the question, however I will try to help a little bit. Presented method is rather deterministic, so if you prefer more NN aimed solution, stop reading :)
First of all make sure that you know what information you want to assess from the part of image you are going to test. Are you going to go with solely RGB intensity and some statistics from that, or you want to work on eg. histogram distribution.
Once you have variables that depict the state of the picture you are analyzing you can try to apply simple correlation function and perform statistical hypothesis testing. If all goes well you can go to the next step ie. formalization of rules that would lead to assessing "x" values based on variable you have selected. However, if the statistics are bad, go to the first point and look for better features and properties.
I wish I could give you more detailed answer, however your question was a bit ambiguous and thus its all I can do. Try to revise the question or provide more details and clear explanation for better answer to come.
To me, this sounds like a regression problem. Have you tried linear regression or Support Vector Regression? Perhaps you can be more specific. Here is a link that might help: https://en.wikipedia.org/wiki/Regression_analysis
OK will try : 6 datapoint , each datapoint spread out , identified as regions of interest . You take histogram -global or local ??
If local , worst case say take average min/max value and try to correlate to the datapoint corresponding to that region (assuming a pixel or coordinate value)
If global , use either Euclidean distance or peaks of the histogram and overlap or check covariance between the two segments (worst case)
OR use logistic regression to check where they lie in the given space !
It looks like Histogram matching problem. You have six points in interval 1 to 6 then modified image. Now you want to map input image histogram to modified image's histogram. If your problem is like this ten it is classical problem of digital image processing. You can follow any DIP book.
- Badges
- Science topic
- Similar topics
- Computer Science
- Images