Using double instead of im2double for images?

I want to convert an input image into floating point, so that interpolation using interp2 will be possible. I want to achieve this without using the Image Processing Toolbox and without the im2double function. So I tried the syntax I=double(I), however, the resulting image looked very different than the original.

Here is a comparison of the resulting images using im2double(I) and double(I) respectively:

What is happening here? It looks like the double(I) code eliminates the higher frequencies from the spectrum of the image:

When I plot double(I) using imagesc instead of imshow I get the right image displayed, but the spectrum is still shrunken.

So, how can I implement what im2double does without using the pre-defined function? Any explanation is greatly appreciated.

Answer 1

[Stephen23]

"What is happening here?"

You input image is of class uint8, which means its values are from [0-255]. When you call double, you get the same range of values [0-255]. In contrast calling im2double scales the output to [0-1] (have a look in the variable viewer, and you will see that their output values are different).

imshow by default assumes that its inputs are [0-1]. So when you give it data [0-255] all of the values above 1 are shown as white. There are two easy solutions:

1. scale the data yourself: out = double(inp)/255
2. tell imshow the range to use: imshow(inp,[0,255])

Answer 2

[Mohammad Anas]

From your question, I can guess the solution to your problem is very simple. You probably may have to scale back the resulting image by multiplying with a factor of 255 in order to see the same image that would result by using im2double. Hope that helps.