Bug with conv2

Here is a sample convolution performed with the built in function conv2. Unless I am completely crazy, this is not even close to the correct answer? Based on that image and kernel, I was expecting the output to be: [3 5 4; 9 0 6; 1 7 3]. Does anyone know of issues with conv2, or have insight to share?

Answer 1

[Ajay Pattassery]

The above output from conv2 with the argument as 'valid' is expected.

If you are computing in the paper the kernel needs to be flipped and then slid over the image I.

The resultant kernel will be

     0     0     0
     0     0     0
     0     0     1

Then the output will be,

     3     7     8
     7     2     9
     3     8     0

Answer 2

[Bruno Luong]

The answer is correct. May be you expect the result with sliding sum with unflip kernel which is of course wrong.