Normally in all the transaction paper's related to 2D DCT. They are explaining 2DCT using row column decomposition method as follows:

1. Apply 1D DCT on all the rows.
2. Transpose of intermediate results.
3. finally applying 1D DCT on all columns. But in matlab if I follow the same procedure I am not getting the output values as inbuilt dct2 function output. If I follow the same procedure without intermediate transpose I will be getting the same value as inbuilt dct2 output.

I want to know the reason why intermediate transpose is not required in matlab implementation.