For this algorithm we’ll continue to use of the Vitis Visionlibraries.
As before, we’ll configure the library (in the hardware, via templates in our hardware source files) to process eight pixels per clock cycle. Functionally, our hardware algorithm is now equivalent to this in standard software OpenCV:
cv::resize(image, resize_ocv, cv::Size(out_width, out_height), 0, 0, CV_INTER_LINEAR);
cv::GaussianBlur(resize_ocv, result_ocv, cv::Size(7, 7), 3.0f, 3.0f, cv::BORDER_CONSTANT);
We’re resizing as in Example 7, and as we mentioned are applying a 7 × 7 window to the Gaussian blur function. We have also (arbitrarily) selected σx = σy =3.0