I am trying to understand what is there underhood of the convolutional layer in a conv-net.
I have read many tutorials and the blogs on medium and many things about conv-net but I found there was a lack of clarification on the convolutional process itself. I want to know what is the convolutional layer of the network is doing there we all know that there is a kernel(filter) that loop through whole image span and take the dot product of itself and this is all of its output but what does the actual meaning of this convolution process in terms of matrix what does it assigns to its output what is its physical significance of this particular number .
Just like using a Gaussian kernel we can interpret it as it is giving the central pixel a higher weight so that it can grow more in the blurred image and edges can become sharper so that they can be seen in the blurred image clearly all of this operation is done as we consider the Gaussian distribution of the focused window on image (same dimension of kernel) the probability value of the central pixel may become lower(if it is a minority in it's surrounding):_- This is all why we make a choice of Gaussian kernel for image blur and other image operation.
Now I want to create the same intuition as that of the Gaussian kernel in terms of surrounding pixel values.
Summarising my query, I want to draw an intuition for a convolutional layer in terms of operation made with filter and surrounding values within the image window that is in current focus.