What does it mean to compute filter response averages for each cell? By my understanding I thought we were simply applying each of the N filters in the filter bank to each patch of the image, and collecting the image gradients in varying directions, acting as an edge detector.
This also lines up with the statement that the size of the descriptor is 4x4xN. So what do we mean by average the responses for each cell? For each cell do we average the responses to each of the filters to produce one response?
motoole2
The $N$ filters from the filter bank are applied to the entire image, producing $N$ images (i.e., one image for every filter). We then divide these images into 4x4 cells and compute the average response within each cell.
By averaging, we are averaging the response from the same filter. Given that I have 4x4 cells, and I have $N$ different filters, the descriptor is 4x4x$N$.
What does it mean to compute filter response averages for each cell? By my understanding I thought we were simply applying each of the N filters in the filter bank to each patch of the image, and collecting the image gradients in varying directions, acting as an edge detector.
This also lines up with the statement that the size of the descriptor is 4x4xN. So what do we mean by average the responses for each cell? For each cell do we average the responses to each of the filters to produce one response?
The $N$ filters from the filter bank are applied to the entire image, producing $N$ images (i.e., one image for every filter). We then divide these images into 4x4 cells and compute the average response within each cell.
By averaging, we are averaging the response from the same filter. Given that I have 4x4 cells, and I have $N$ different filters, the descriptor is 4x4x$N$.