Image intensity as a function of pattern strength

Hi All

Attached are images of square wave gratings, a "barcode" and random noise. Please let us assume for the discussion, that all of the images have the same number of pixels and the same average luminance.

When we look at it, we can see stright away that the square wave gratings is most "patterns", the barcode has less initsirc order and the noise image is simply a noise.

I have tried several functions, but they keep showing that the noise image has the most clusters, and I also could not find a way to know how strong a pattern is in an image.

Many thanks

Ron

[Walter Roberson]

Will the lines always be vertical? If so that is information that can be used.

Are the patterns perfect, or could there be occasional missing pixels ? I notice what you posted is JPEG, which is prone to aliasing artifacts never every vertical line.

[Ron Meidan]

Hi Walter

The bars are just used as an example. As you mentioned the term "perfect", this is exactly what I was asking about. With our eyes, it is plain to see a perfect pattern, and I wondering if MATLAB could also tell how "perfect" an image is.

Many thanks

Ron

Answer 1

[Image Analyst]

There are several attributes for each image. You can measure all of them and then decide which of them corresponds to the "strength" of the pattern. For example you can measure the mean and standard deviation. You can measure the number of black blobs. You can measure the area and solidity of each blob, etc. For example for the 3 with different size bars in them, are they all the same strength or not? If not what determines strength?

Perhaps this will help:

https://www.mathworks.com/help/releases/R2021a/images/ref/brisque.html?doclanguage=en&prodfilter=ML+VP+DB+NN+IA+IP+CO+DM+SG+ST&addons=usbwebcams

[Ron Meidan]

Hi Image Analyst

Most thank you for your answer.

If you may, please look at the attached image, as an example.

Both of the pattern have roughly similar spatial frequency, amount of pixels, and total amout of elements.

However, the upper row seems to be very distinct in it's visual attributes, while the one beneath it seems to be somewhat more random and with less distinct features.It has less repetitions as the elements vary in their shapes and distances from one another.

Does MATLAB has a way to straight away tell us how "flawed" patterns such as the bottom one is, even without giving prior criteria?

Many thanks

Ron

[Image Analyst]

Again there are lots of ways to describe it - lots of properties/attributes. If you look at color, the top row and bottom row are identical. If you look at shape or y position they are not. And not all of the shapes in the top row are identical. They differ in their x value across the image. You cannot compare things in advance without saying what to compare. However you can measure everything you can think of and take the standard deviations of the properties. So for the top row, the SD of the shape, area, color etc. will be zero but the x values are different so there will be a non-zero stdev for x location. Then if you have two images you can compare the two sets of SDs to see if one set has SDs on average that are lower than the other set.

[Ron Meidan]

Thank you very much.