Why do we transfer the image into double?

I saw a lot of image processing examples and before conducting any analysis the image is transfered into double. Is it essential? and what is the benefit from doing so?

Answer 1

[Image Analyst]

No, it's usually not. But sometimes it is, like if you want to use conv2(). It has to do with how MATLAB likes to multiply variables of different classes.

[Muhamed Sewidan]

Could you please give me some cases in which I should transfer the image into double before carrying out any processes?

[Image Analyst]

Other than conv2(), I can't think of any off the top of my head. But you obviously must have had some cases in mind because you're the one who brought up the question. Care to share those cases?

Answer 2

[Paul Hoffrichter]

MATLAB when using mixed mode arithmetic, will try to save memory.

250 + 10 = 260, right?

>> x =double(250) + uint8(10);
>> x
x =
  uint8
   255

In other languages, like C/C++, the rule is to maximize precision. Avoid mixed mode arithmetic unless you need it. There is no error or warning provided, by default, in the above code.

[Muhamed Sewidan]

in the normal case, all the data (like numbers) are stored in double or uint8?

[Paul Hoffrichter]

Normally, data in MATLAB defaults to double unless specified by a function to be otherwise.

When using imread, it is usual to get a integer type, but not necessarily, uint8. Here you also see uint16, and others.

https://www.mathworks.com/help/matlab/ref/imread.html#btnecad-5

In fact, I see that TIFF can be returned as single floating point number:

"imread reads and converts TIFF images as follows:

• YCbCr images are converted into the RGB colorspace.
• All grayscale images are read as if black = 0, white = largest value.
• 1-bit images are returned as class logical.
• 16-bit floating-point images are returned as class single."

[Muhamed Sewidan]

thank you

[Paul Hoffrichter]

Glad to help.