how convert pixel to mm for a mat file using Matlab?

I was wondering how can I convert pixel to the mm while my image is a mat file (uint 8). For example, I have a kernel size of 11 * 11 and another one a kernel size of 21 * 21, and I need to know what is equivalent of these sizes in mm?

Answer 1

[Walter Roberson]

Pixels have no inherent size. 11x11 matrix of pixels could be representing picometers or gigaparsecs. You need some kind of reference information.

[Neda]

how can I get reference information from a mat file? The image size is 580*682

[Walter Roberson]

• perhaps you were given information such as that the image represents 217mm by 256 mm
• perhaps you were given information such as that the image is 1/2 degree arc at 3.8 AU
• in cases where you have an image file, sometimes there is header information such as EXIF information about what the pixel resolution is. Occasionally that information is not an outright lie, but unless you are using a scientific instrument such as a microscope, you should assume that the information probably has little relationship to reality
• in cases where you have an image file, sometimes there is header information about aperature and lens length setting, or aperature and focal distance, that can permit you to calculate pixel resolution. Higher end cameras try hard to get the right focal distance because it is important for autofocus to get clear pictures, consumer cameras just get somewhere in the right range. However, most auto-focus is only accurate to about 3-ish meters; I see one of the Canon prosumer cameras indicating it has about an 8 meter autofocus range. There is also the problem that for any given lens settings, there is a distance beyond which any object effectively becomes at "infinity"
• DICOM header information from a calibrated medical device tends to be pretty accurate
• perhaps there is something in the image which is of exactly known size
• perhaps there is something in the image which is of estimatable size. For example you can make out the pixel distance between the centers of the pupils of a person being photographed, and you are able to estimate their age and gender, you can use reference statistical tables to estimate the real-world distance represented

[Neda]

@Walter Roberson thank you for spending time and explain different cases. In the folder, I could see info file which includes, info.x and info.y which image (mat file) should rescale based on these values. (there isn’t any header file)For example, info.x is included (only five rows display here)

0              0.246761819514584         0.493523639029168         0.740285458543752         0.987047278058337

And info.y is included (only five rows display here)

0              0.246825559975158         0.493651119950316         0.740476679925474         0.987302239900632

Could you please point me to the right direction how I can convert the pixel to the mm based on these reference information?

[Walter Roberson]

Unfortunately without further information those values do not explain anything.

The values are at least consistent with the X all being 0.246761819514584 apart to within round-off error, and the Y all being the slightly larger 0.246825559975158 apart to within round-off error. However, those values do not mean anything to me -- they are not, for example, conversion factors for decameters to inches (0.254)

Is it possible that the folder has 25 images arranged on a grid, and that the info you are looking at is the coordinates of the individual images on the grid?

[Neda]

Those info.x an y needs when I want to display the points on the image(basically there is some points indicate the gold standard and in order to display those points over the image, I should rescale them based on those info values). So you mean I can't use these info as a reference information to convert the pixel to mm?

Image and info are attached

[Walter Roberson]

The information there is not enough to convert pixels to mm.

It looks to me as if you have a reconstructed ultrasound image with minor variations in coordinates, perhaps due to round-off in the reconstruction. There does appear to be a small scale difference between x and y.

Unfortunately, ultrasound is a technology used on a variety of scales, including for small scale welding, and large scale welding, and fetal monitoring. Just knowing the technology involved does not give us any scale information.

You could guess that the coordinates are in mm and see whether it gives you reasonable conclusions. Is it plausible that you are looking at something that is about 6.6 inches by 5.6 inches?

[Walter Roberson]

You could multiply number of pixels by mean(diff(info.x)) or mean(diff(info.y)) to get "units" for which there is still a missing scaling factor to convert to millimeters, with it being plausible that the coordinates reflect millimeters directly.

However if the small differences in coordinates is important, you could get a more accurate area, except for the pixels on the right or bottom edge: you can deduce pixel widths by looking at differences in adjacent coordinates, but you do not have the coordinates for what is beyond the right or bottom edge in order to deduce the widths of those pixels.