Learning OpenCV, 2nd Edition

double delta = 0, // Offset before assignment to dst

int borderType = cv::BORDER_DEFAULT // Border extrapolation to use

);

All the arguments of cv::sepFilter2D() are the same as those of cv::filter2D(), with the

exception of the replacement of the kernel argument with the rowKernel and columnKernel

arguments. The latter two are expected to be 𝑛

!

-by-1 and 1-by-𝑛

!

arrays (with 𝑛

!

not necessarily equal to

𝑛

!

).

The Filter Architecture

Almost all of the functions described in this chapter are based on a unified Filter Architecture. The Filter

Architecture is a class system that allows you (or the OpenCV developers) to create a new filter and

automatically reuse all of the things that every filter has in common (the convolution steps, handling of the

borders, etc.)

The Big Picture

The process of doing convolutions is complex because these are always very computationally intensive

operations. As a result, there is a great benefit to doing things very efficiently. On the other hand, writing

very efficient code is often very difficult to do in a modular way. As a result, OpenCV has an entire

architecture for handling just this problem. This architecture begins at the top with high-level operations

like cv::blur(). This function takes your image, smoothes it, and gives it back. Sounds simple, but this

is actually ...