detectSIFTFeatures only working for uint8

I = imread('cameraman.tif');
points = detectSIFTFeatures(I);

Gives me heaps of points as expected

I = imread('cameraman.tif');
points = detectSIFTFeatures(double(I));

gives no points. The same zero points retured for anything other than uint8, for every image I've tried. The same behaviour is true for SURF points.

Does anyone have any suggestions?

Answer 1

[Udit06]

Hi Michael,

If you want to use an image of double data type as input, you should first scale it down to the range [0, 1]. Doing so will give you the similar results that you are getting with uint8 data type. Here is the code for the same.

I = imread('cameraman.tif');
points1 = detectSIFTFeatures(I)

points1 = 
  274×1 SIFTPoints array with properties:

          Scale: [274×1 single]
    Orientation: [274×1 single]
         Octave: [274×1 int32]
          Layer: [274×1 int32]
       Location: [274×2 single]
         Metric: [274×1 single]
          Count: 274

I = imread('cameraman.tif');
points2 = detectSIFTFeatures(double(I)/255)

points2 = 
  274×1 SIFTPoints array with properties:

          Scale: [274×1 single]
    Orientation: [274×1 single]
         Octave: [274×1 int32]
          Layer: [274×1 int32]
       Location: [274×2 single]
         Metric: [274×1 single]
          Count: 274

You can refer to the following note of the MathWorks documentation for the same.

https://www.mathworks.com/help/matlab/ref/im2double.html#:~:text=input%20pixel%20values.-,Note,-Many%20MATLAB%C2%AE

I hope this helps.

[DGM]

https://www.mathworks.com/help/images/convert-image-data-between-classes.html

[Michael Jones]

Thanks!

Answer 2

[Qu Cao]

Use im2double:

I = imread('cameraman.tif');
points = detectSIFTFeatures(im2double(I))

points = 
  274x1 SIFTPoints array with properties:

          Scale: [274x1 single]
    Orientation: [274x1 single]
         Octave: [274x1 int32]
          Layer: [274x1 int32]
       Location: [274x2 single]
         Metric: [274x1 single]
          Count: 274