how to calculate the accuracy of an image and display it in bar graph?

Example: display bar graph that show accuracy of the original images and also contrast image.

Thank you

i'm very appreciate for your help

Answer 1

[Image Analyst]

To know the accuracy of an image, or of anything, you need to know the absolute true value. Do you know that?

I don't know what you mean by "contrast image". Do you mean that you want to know how to increase or decrease he contrast of an image, or do you want to know the range or the gray levels, or perhaps the range divided by the mean, or something else?

[shaira ain]

im sorry.. im a newbie in matlab..

actually i want to know the accuracy of an image based on the range of the gray levels?

[Image Analyst]

This doesn't have anything to do with MATLAB, yet. How do you know the accuracy of your image is not 100%? Why do you suspect errors?

[shaira ain]

actually, im doing an image processing mini project for this subject.. im using image enhancement technique...that includes log transformation, power log transformation, gamma correction, contrast stretching, histogram equalization, BBHE and MMBBHE.

and my lecturer ask me to show the accuracy of each image throught a bar graph..

thank you..

[Walter Roberson]

In order to make this calculation, you would need a copy of the "real" image, the image as it would be if it was perfect. Do you have that?

[shaira ain]

do you mean an "original image"? i have it.

[Walter Roberson]

Did you start with an image and artificially degrade it and now are trying to use enhancement techniques to get closer to the first image? If you did, then that first image is what needs to be known.

If you started with a degraded image and are trying to get more out of it (e.g., trying to read a blurry license plate), then you would not have any way of measuring the accuracy.

[shaira ain]

yup...the first one

i start an image and artificially degrade it and now trying to use enhancement techniques to get closer to the first image.

do you know how??

[Walter Roberson]

Okay, we are getting somewhere.

Now you need to define "accuracy" for your purpose.

For example, which is better: that 100 pixels are one shade darker than they should be, or that one pixel is wildly completely wrong?

[shaira ain]

one pixel is wildly completely wrong is better... is it right??

[Walter Roberson]

There is no right or wrong for this, just how you define accuracy. There are some reasonable definitions of "accuracy" that would say that one pixel wildly wrong would be about equivalent to 38 pixels each wrong by the minimum possible difference. But if you have a human looking at the image, often a block that is one shade darker is completely unnoticeable without special study, whereas one pixel wildly wrong can attract a lot of attention to the human visual system (e.g., a bright red speck on a field of black.)

[shaira ain]

oh, ok.. i understand.. then, how to calculate the accuracy?