Merge pull request #18053 from Yosshi999:bit-exact-resizeNN
authorYosshi999 <Yosshi999@users.noreply.github.com>
Fri, 28 Aug 2020 18:20:05 +0000 (03:20 +0900)
committerGitHub <noreply@github.com>
Fri, 28 Aug 2020 18:20:05 +0000 (21:20 +0300)
commit7495a4722f5e23ed16e4d1ef6d4db9483aec31d8
treeab58794cb72710e25894e6bac37deca4b7ee7774
parent7ce56b3a47ad7c4719a584f4b0156a28d86aacf5
Merge pull request #18053 from Yosshi999:bit-exact-resizeNN

Bit-exact Nearest Neighbor Resizing

* bit exact resizeNN

* change the value of method enum

* add bitexact-nn to ResizeExactTest

* test to compare with non-exact version

* add perf for bit-exact resizenn

* use cvFloor-equivalent

* 1/3 scaling is not stable for floating calculation

* stricter test

* bugfix: broken data in case of 6 or 12bytes elements

* bugfix: broken data in default pix_size

* stricter threshold

* use raw() for floor

* use double instead of int

* follow code reviews

* fewer cases in perf test

* center pixel convention
modules/imgproc/include/opencv2/imgproc.hpp
modules/imgproc/perf/perf_resize.cpp
modules/imgproc/src/fixedpoint.inl.hpp
modules/imgproc/src/resize.cpp
modules/imgproc/test/test_imgwarp.cpp
modules/imgproc/test/test_resize_bitexact.cpp