I have used 5 available background extraction techniques (M1, M2, M3, M4, M5) on the image. From the image one can suggest visually that method 4 is best. But how to validate it statistically.
I don't have the ground truth for the same. Please suggest
To be able to apply statistical inference I think that you have to replicate the experiment a number of times, measure the appropriate responses and apply analysis of variance (ANOVA). If the ANOVA is significant, then you can use a multiple comparison method like Duncan's multiple range test to find out whether M4 is indeed the best.
Thank you for the suggestion Dr. ETuk .
It would be of great help if you can suggest or provide some references on how i can apply ANOVA on image
You need to find/define some univariate measure of the "quality" of your approach. This might be - for instance - the proportion of correctly assigned pixels (or, I don't know... something entirely different; however you aim to define "quality").
This measure is obtained for a collection (sample) of different images and for each method. It should be possible to identify some sensible distribution model for these measures. In case of the "proportion of correctly assigned pixels" one may think of a beta-distribution. Having the data and the distribution model you can test if the expected value of your measure is different between the methods (depends on the method). In case you assume that the normal distribution is an adequate model for the distribution of the measure, this test will be an ANOVA (or, more precisely: a linear model of the normal distribution family; depending on your distributional assumption this may be a different family as well - what is not called "ANOVA").
You may even consider covariates in the model, like: number of foreground objects, size of the foreground objects, color space, whatsoever... This would require to analyse more images spanning the ranges of possible values of these covariates. This may reveal thing like, for instance, that M1 is be better for large objects, whereas M3 is superior for many small objects, or that M4 is the best throughout, etc.
- Badges
- Science topic
- Similar topics
- Analytical Chemistry
- Extraction