opencv/modules/imgproc/test/test_contours_new.cpp
Maksim Shabunin a25132986a
Merge pull request #25146 from mshabunin:cpp-contours
Reworked findContours to reduce C-API usage #25146

What is done:
* rewritten `findContours` and `icvApproximateChainTC89` using C++ data structures
* extracted LINK_RUNS mode to separate new public functions - `findContoursLinkRuns` (it uses completely different algorithm)
* ~added new public `cv::approximateChainTC89`~ - ** decided to hide it**
* enabled chain code output (method = 0, no public enum value for this in C++ yet)
* kept old function as `findContours_old` (exported, but not exposed to user)
* added more tests for findContours (`test_contours_new.cpp`), some tests compare results of old function with new one. Following tests have been added:
  * contours of random rectangle
  * contours of many small (1-2px) blobs
  * contours of random noise
  * backport of old accuracy test
  * separate test for LINK RUNS variant

What is left to be done (can be done now or later):
* improve tests: 
  * some tests have limited verification (e.g. only verify contour sizes)
  * perhaps reference data can be collected and stored
  * maybe more test variants can be added (?)
* add enum value for chain code output and a method of returning starting points (e.g. first 8 elements of returned `vector<uchar>` can represent 2 int point coordinates)
* add documentation for new functions - **✔️ DONE**
* check and improve performance (my experiment showed 0.7x-1.1x some time ago)
* remove old functions completely (?)
* change contour return order (BFS) or allow to select it (?)
* return result tree as-is (?) (new data structures should be exposed, bindings should adapt)
2024-04-09 09:37:49 +03:00

607 lines
20 KiB
C++

// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html
#include "opencv2/imgproc/types_c.h"
#include "test_precomp.hpp"
#include "opencv2/ts/ocl_test.hpp"
#include "opencv2/imgproc/detail/legacy.hpp"
#define CHECK_OLD 1
namespace opencv_test { namespace {
// debug function
template <typename T>
inline static void print_pts(const T& c)
{
for (const auto& one_pt : c)
{
cout << one_pt << " ";
}
cout << endl;
}
// debug function
template <typename T>
inline static void print_pts_2(vector<T>& cs)
{
int cnt = 0;
cout << "Contours:" << endl;
for (const auto& one_c : cs)
{
cout << cnt++ << " : ";
print_pts(one_c);
}
};
// draw 1-2 px blob with orientation defined by 'kind'
template <typename T>
inline static void drawSmallContour(Mat& img, Point pt, int kind, int color_)
{
const T color = static_cast<T>(color_);
img.at<T>(pt) = color;
switch (kind)
{
case 1: img.at<T>(pt + Point(1, 0)) = color; break;
case 2: img.at<T>(pt + Point(1, -1)) = color; break;
case 3: img.at<T>(pt + Point(0, -1)) = color; break;
case 4: img.at<T>(pt + Point(-1, -1)) = color; break;
case 5: img.at<T>(pt + Point(-1, 0)) = color; break;
case 6: img.at<T>(pt + Point(-1, 1)) = color; break;
case 7: img.at<T>(pt + Point(0, 1)) = color; break;
case 8: img.at<T>(pt + Point(1, 1)) = color; break;
default: break;
}
}
inline static void drawContours(Mat& img,
const vector<vector<Point>>& contours,
const Scalar& color = Scalar::all(255))
{
for (const auto& contour : contours)
{
for (size_t n = 0, end = contour.size(); n < end; ++n)
{
size_t m = n + 1;
if (n == end - 1)
m = 0;
line(img, contour[m], contour[n], color, 1, LINE_8);
}
}
}
//==================================================================================================
// Test parameters - mode + method
typedef testing::TestWithParam<tuple<int, int>> Imgproc_FindContours_Modes1;
// Draw random rectangle and find contours
//
TEST_P(Imgproc_FindContours_Modes1, rectangle)
{
const int mode = get<0>(GetParam());
const int method = get<1>(GetParam());
const size_t ITER = 100;
RNG rng = TS::ptr()->get_rng();
for (size_t i = 0; i < ITER; ++i)
{
SCOPED_TRACE(cv::format("i=%zu", i));
const Size sz(rng.uniform(640, 1920), rng.uniform(480, 1080));
Mat img(sz, CV_8UC1, Scalar::all(0));
Mat img32s(sz, CV_32SC1, Scalar::all(0));
const Rect r(Point(rng.uniform(1, sz.width / 2 - 1), rng.uniform(1, sz.height / 2)),
Point(rng.uniform(sz.width / 2 - 1, sz.width - 1),
rng.uniform(sz.height / 2 - 1, sz.height - 1)));
rectangle(img, r, Scalar::all(255));
rectangle(img32s, r, Scalar::all(255), FILLED);
const vector<Point> ext_ref {r.tl(),
r.tl() + Point(0, r.height - 1),
r.br() + Point(-1, -1),
r.tl() + Point(r.width - 1, 0)};
const vector<Point> int_ref {ext_ref[0] + Point(0, 1),
ext_ref[0] + Point(1, 0),
ext_ref[3] + Point(-1, 0),
ext_ref[3] + Point(0, 1),
ext_ref[2] + Point(0, -1),
ext_ref[2] + Point(-1, 0),
ext_ref[1] + Point(1, 0),
ext_ref[1] + Point(0, -1)};
const size_t ext_perimeter = r.width * 2 + r.height * 2;
const size_t int_perimeter = ext_perimeter - 4;
vector<vector<Point>> contours;
vector<vector<schar>> chains;
vector<Vec4i> hierarchy;
// run functionn
if (mode == RETR_FLOODFILL)
if (method == 0)
findContours(img32s, chains, hierarchy, mode, method);
else
findContours(img32s, contours, hierarchy, mode, method);
else if (method == 0)
findContours(img, chains, hierarchy, mode, method);
else
findContours(img, contours, hierarchy, mode, method);
// verify results
if (mode == RETR_EXTERNAL)
{
if (method == 0)
{
ASSERT_EQ(1U, chains.size());
}
else
{
ASSERT_EQ(1U, contours.size());
if (method == CHAIN_APPROX_NONE)
{
EXPECT_EQ(int_perimeter, contours[0].size());
}
else if (method == CHAIN_APPROX_SIMPLE)
{
EXPECT_MAT_NEAR(Mat(ext_ref), Mat(contours[0]), 0);
}
}
}
else
{
if (method == 0)
{
ASSERT_EQ(2U, chains.size());
}
else
{
ASSERT_EQ(2U, contours.size());
if (mode == RETR_LIST)
{
if (method == CHAIN_APPROX_NONE)
{
EXPECT_EQ(int_perimeter - 4, contours[0].size());
EXPECT_EQ(int_perimeter, contours[1].size());
}
else if (method == CHAIN_APPROX_SIMPLE)
{
EXPECT_MAT_NEAR(Mat(int_ref), Mat(contours[0]), 0);
EXPECT_MAT_NEAR(Mat(ext_ref), Mat(contours[1]), 0);
}
}
else if (mode == RETR_CCOMP || mode == RETR_TREE)
{
if (method == CHAIN_APPROX_NONE)
{
EXPECT_EQ(int_perimeter, contours[0].size());
EXPECT_EQ(int_perimeter - 4, contours[1].size());
}
else if (method == CHAIN_APPROX_SIMPLE)
{
EXPECT_MAT_NEAR(Mat(ext_ref), Mat(contours[0]), 0);
EXPECT_MAT_NEAR(Mat(int_ref), Mat(contours[1]), 0);
}
}
else if (mode == RETR_FLOODFILL)
{
if (method == CHAIN_APPROX_NONE)
{
EXPECT_EQ(int_perimeter + 4, contours[0].size());
}
else if (method == CHAIN_APPROX_SIMPLE)
{
EXPECT_EQ(int_ref.size(), contours[0].size());
EXPECT_MAT_NEAR(Mat(ext_ref), Mat(contours[1]), 0);
}
}
}
}
#if CHECK_OLD
if (method != 0) // old doesn't support chain codes
{
if (mode != RETR_FLOODFILL)
{
vector<vector<Point>> contours_o;
vector<Vec4i> hierarchy_o;
findContours_legacy(img, contours_o, hierarchy_o, mode, method);
ASSERT_EQ(contours.size(), contours_o.size());
for (size_t j = 0; j < contours.size(); ++j)
{
SCOPED_TRACE(format("contour %zu", j));
EXPECT_MAT_NEAR(Mat(contours[j]), Mat(contours_o[j]), 0);
}
EXPECT_MAT_NEAR(Mat(hierarchy), Mat(hierarchy_o), 0);
}
else
{
vector<vector<Point>> contours_o;
vector<Vec4i> hierarchy_o;
findContours_legacy(img32s, contours_o, hierarchy_o, mode, method);
ASSERT_EQ(contours.size(), contours_o.size());
for (size_t j = 0; j < contours.size(); ++j)
{
SCOPED_TRACE(format("contour %zu", j));
EXPECT_MAT_NEAR(Mat(contours[j]), Mat(contours_o[j]), 0);
}
EXPECT_MAT_NEAR(Mat(hierarchy), Mat(hierarchy_o), 0);
}
}
#endif
}
}
// Draw many small 1-2px blobs and find contours
//
TEST_P(Imgproc_FindContours_Modes1, small)
{
const int mode = get<0>(GetParam());
const int method = get<1>(GetParam());
const size_t DIM = 1000;
const Size sz(DIM, DIM);
const int num = (DIM / 10) * (DIM / 10); // number of 10x10 squares
Mat img(sz, CV_8UC1, Scalar::all(0));
Mat img32s(sz, CV_32SC1, Scalar::all(0));
vector<Point> pts;
int extra_contours_32s = 0;
for (int j = 0; j < num; ++j)
{
const int kind = j % 9;
Point pt {(j % 100) * 10 + 4, (j / 100) * 10 + 4};
drawSmallContour<uchar>(img, pt, kind, 255);
drawSmallContour<int>(img32s, pt, kind, j + 1);
pts.push_back(pt);
// NOTE: for some reason these small diagonal contours (NW, SE)
// result in 2 external contours for FLOODFILL mode
if (kind == 8 || kind == 4)
++extra_contours_32s;
}
{
vector<vector<Point>> contours;
vector<vector<schar>> chains;
vector<Vec4i> hierarchy;
if (mode == RETR_FLOODFILL)
{
if (method == 0)
{
findContours(img32s, chains, hierarchy, mode, method);
ASSERT_EQ(pts.size() * 2 + extra_contours_32s, chains.size());
}
else
{
findContours(img32s, contours, hierarchy, mode, method);
ASSERT_EQ(pts.size() * 2 + extra_contours_32s, contours.size());
#if CHECK_OLD
vector<vector<Point>> contours_o;
vector<Vec4i> hierarchy_o;
findContours_legacy(img32s, contours_o, hierarchy_o, mode, method);
ASSERT_EQ(contours.size(), contours_o.size());
for (size_t i = 0; i < contours.size(); ++i)
{
SCOPED_TRACE(format("contour %zu", i));
EXPECT_MAT_NEAR(Mat(contours[i]), Mat(contours_o[i]), 0);
}
EXPECT_MAT_NEAR(Mat(hierarchy), Mat(hierarchy_o), 0);
#endif
}
}
else
{
if (method == 0)
{
findContours(img, chains, hierarchy, mode, method);
ASSERT_EQ(pts.size(), chains.size());
}
else
{
findContours(img, contours, hierarchy, mode, method);
ASSERT_EQ(pts.size(), contours.size());
#if CHECK_OLD
vector<vector<Point>> contours_o;
vector<Vec4i> hierarchy_o;
findContours_legacy(img, contours_o, hierarchy_o, mode, method);
ASSERT_EQ(contours.size(), contours_o.size());
for (size_t i = 0; i < contours.size(); ++i)
{
SCOPED_TRACE(format("contour %zu", i));
EXPECT_MAT_NEAR(Mat(contours[i]), Mat(contours_o[i]), 0);
}
EXPECT_MAT_NEAR(Mat(hierarchy), Mat(hierarchy_o), 0);
#endif
}
}
}
}
// Draw many nested rectangles and find contours
//
TEST_P(Imgproc_FindContours_Modes1, deep)
{
const int mode = get<0>(GetParam());
const int method = get<1>(GetParam());
const size_t DIM = 1000;
const Size sz(DIM, DIM);
const size_t NUM = 249U;
Mat img(sz, CV_8UC1, Scalar::all(0));
Mat img32s(sz, CV_32SC1, Scalar::all(0));
Rect rect(1, 1, 998, 998);
for (size_t i = 0; i < NUM; ++i)
{
rectangle(img, rect, Scalar::all(255));
rectangle(img32s, rect, Scalar::all((double)i + 1), FILLED);
rect.x += 2;
rect.y += 2;
rect.width -= 4;
rect.height -= 4;
}
{
vector<vector<Point>> contours {{{0, 0}, {1, 1}}};
vector<vector<schar>> chains {{1, 2, 3}};
vector<Vec4i> hierarchy;
if (mode == RETR_FLOODFILL)
{
if (method == 0)
{
findContours(img32s, chains, hierarchy, mode, method);
ASSERT_EQ(2 * NUM, chains.size());
}
else
{
findContours(img32s, contours, hierarchy, mode, method);
ASSERT_EQ(2 * NUM, contours.size());
#if CHECK_OLD
vector<vector<Point>> contours_o;
vector<Vec4i> hierarchy_o;
findContours_legacy(img32s, contours_o, hierarchy_o, mode, method);
ASSERT_EQ(contours.size(), contours_o.size());
for (size_t i = 0; i < contours.size(); ++i)
{
SCOPED_TRACE(format("contour %zu", i));
EXPECT_MAT_NEAR(Mat(contours[i]), Mat(contours_o[i]), 0);
}
EXPECT_MAT_NEAR(Mat(hierarchy), Mat(hierarchy_o), 0);
#endif
}
}
else
{
const size_t expected_count = (mode == RETR_EXTERNAL) ? 1U : 2 * NUM;
if (method == 0)
{
findContours(img, chains, hierarchy, mode, method);
ASSERT_EQ(expected_count, chains.size());
}
else
{
findContours(img, contours, hierarchy, mode, method);
ASSERT_EQ(expected_count, contours.size());
#if CHECK_OLD
vector<vector<Point>> contours_o;
vector<Vec4i> hierarchy_o;
findContours_legacy(img, contours_o, hierarchy_o, mode, method);
ASSERT_EQ(contours.size(), contours_o.size());
for (size_t i = 0; i < contours.size(); ++i)
{
SCOPED_TRACE(format("contour %zu", i));
EXPECT_MAT_NEAR(Mat(contours[i]), Mat(contours_o[i]), 0);
}
EXPECT_MAT_NEAR(Mat(hierarchy), Mat(hierarchy_o), 0);
#endif
}
}
}
}
INSTANTIATE_TEST_CASE_P(
,
Imgproc_FindContours_Modes1,
testing::Combine(
testing::Values(RETR_EXTERNAL, RETR_LIST, RETR_CCOMP, RETR_TREE, RETR_FLOODFILL),
testing::Values(0,
CHAIN_APPROX_NONE,
CHAIN_APPROX_SIMPLE,
CHAIN_APPROX_TC89_L1,
CHAIN_APPROX_TC89_KCOS)));
//==================================================================================================
typedef testing::TestWithParam<tuple<int, int>> Imgproc_FindContours_Modes2;
// Very approximate backport of an old accuracy test
//
TEST_P(Imgproc_FindContours_Modes2, new_accuracy)
{
const int mode = get<0>(GetParam());
const int method = get<1>(GetParam());
RNG& rng = TS::ptr()->get_rng();
const int blob_count = rng.uniform(1, 10);
const Size sz(rng.uniform(640, 1920), rng.uniform(480, 1080));
const int blob_sz = 50;
// prepare image
Mat img(sz, CV_8UC1, Scalar::all(0));
vector<RotatedRect> rects;
for (int i = 0; i < blob_count; ++i)
{
const Point2f center((float)rng.uniform(blob_sz, sz.width - blob_sz),
(float)rng.uniform(blob_sz, sz.height - blob_sz));
const Size2f rsize((float)rng.uniform(1, blob_sz), (float)rng.uniform(1, blob_sz));
RotatedRect rect(center, rsize, rng.uniform(0.f, 180.f));
rects.push_back(rect);
ellipse(img, rect, Scalar::all(100), FILLED);
}
// draw contours manually
Mat cont_img(sz, CV_8UC1, Scalar::all(0));
for (int y = 1; y < sz.height - 1; ++y)
{
for (int x = 1; x < sz.width - 1; ++x)
{
if (img.at<uchar>(y, x) != 0 &&
((img.at<uchar>(y - 1, x) == 0) || (img.at<uchar>(y + 1, x) == 0) ||
(img.at<uchar>(y, x + 1) == 0) || (img.at<uchar>(y, x - 1) == 0)))
{
cont_img.at<uchar>(y, x) = 255;
}
}
}
// find contours
vector<vector<Point>> contours;
vector<Vec4i> hierarchy;
findContours(img, contours, hierarchy, mode, method);
// 0 < contours <= rects
EXPECT_GT(contours.size(), 0U);
EXPECT_GE(rects.size(), contours.size());
// draw contours
Mat res_img(sz, CV_8UC1, Scalar::all(0));
drawContours(res_img, contours);
// compare resulting drawn contours with manually drawn contours
const double diff1 = cvtest::norm(cont_img, res_img, NORM_L1) / 255;
if (method == CHAIN_APPROX_NONE || method == CHAIN_APPROX_SIMPLE)
{
EXPECT_EQ(0., diff1);
}
#if CHECK_OLD
vector<vector<Point>> contours_o;
vector<Vec4i> hierarchy_o;
findContours(img, contours_o, hierarchy_o, mode, method);
ASSERT_EQ(contours_o.size(), contours.size());
for (size_t i = 0; i < contours_o.size(); ++i)
{
SCOPED_TRACE(format("contour = %zu", i));
EXPECT_MAT_NEAR(Mat(contours_o[i]), Mat(contours[i]), 0);
}
EXPECT_MAT_NEAR(Mat(hierarchy_o), Mat(hierarchy), 0);
#endif
}
TEST_P(Imgproc_FindContours_Modes2, approx)
{
const int mode = get<0>(GetParam());
const int method = get<1>(GetParam());
const Size sz {500, 500};
Mat img = Mat::zeros(sz, CV_8UC1);
for (int c = 0; c < 4; ++c)
{
if (c != 0)
{
// noise + filter + threshold
RNG& rng = TS::ptr()->get_rng();
cvtest::randUni(rng, img, 0, 255);
Mat fimg;
boxFilter(img, fimg, CV_8U, Size(5, 5));
Mat timg;
const int level = 44 + c * 42;
// 'level' goes through:
// 86 - some black speckles on white
// 128 - 50/50 black/white
// 170 - some white speckles on black
cv::threshold(fimg, timg, level, 255, THRESH_BINARY);
}
else
{
// circle with cut
const Point center {250, 250};
const int r {20};
const Point cut {r, r};
circle(img, center, r, Scalar(255), FILLED);
rectangle(img, center, center + cut, Scalar(0), FILLED);
}
vector<vector<Point>> contours;
vector<Vec4i> hierarchy;
findContours(img, contours, hierarchy, mode, method);
#if CHECK_OLD
vector<vector<Point>> contours_o;
vector<Vec4i> hierarchy_o;
findContours_legacy(img, contours_o, hierarchy_o, mode, method);
ASSERT_EQ(contours_o.size(), contours.size());
for (size_t i = 0; i < contours_o.size(); ++i)
{
SCOPED_TRACE(format("c = %d, contour = %zu", c, i));
EXPECT_MAT_NEAR(Mat(contours_o[i]), Mat(contours[i]), 0);
}
EXPECT_MAT_NEAR(Mat(hierarchy_o), Mat(hierarchy), 0);
#endif
// TODO: check something
}
}
// TODO: offset test
// no RETR_FLOODFILL - no CV_32S input images
INSTANTIATE_TEST_CASE_P(
,
Imgproc_FindContours_Modes2,
testing::Combine(testing::Values(RETR_EXTERNAL, RETR_LIST, RETR_CCOMP, RETR_TREE),
testing::Values(CHAIN_APPROX_NONE,
CHAIN_APPROX_SIMPLE,
CHAIN_APPROX_TC89_L1,
CHAIN_APPROX_TC89_KCOS)));
TEST(Imgproc_FindContours, link_runs)
{
const Size sz {500, 500};
Mat img = Mat::zeros(sz, CV_8UC1);
// noise + filter + threshold
RNG& rng = TS::ptr()->get_rng();
cvtest::randUni(rng, img, 0, 255);
Mat fimg;
boxFilter(img, fimg, CV_8U, Size(5, 5));
const int level = 135;
cv::threshold(fimg, img, level, 255, THRESH_BINARY);
vector<vector<Point>> contours;
vector<Vec4i> hierarchy;
findContoursLinkRuns(img, contours, hierarchy);
if (cvtest::debugLevel >= 10)
{
print_pts_2(contours);
Mat res = Mat::zeros(sz, CV_8UC1);
drawContours(res, contours);
imshow("res", res);
imshow("img", img);
waitKey(0);
}
#if CHECK_OLD
vector<vector<Point>> contours_o;
vector<Vec4i> hierarchy_o;
findContours_legacy(img, contours_o, hierarchy_o, 0, 5); // CV_LINK_RUNS method
ASSERT_EQ(contours_o.size(), contours.size());
for (size_t i = 0; i < contours_o.size(); ++i)
{
SCOPED_TRACE(format("contour = %zu", i));
EXPECT_MAT_NEAR(Mat(contours_o[i]), Mat(contours[i]), 0);
}
EXPECT_MAT_NEAR(Mat(hierarchy_o), Mat(hierarchy), 0);
#endif
}
}} // namespace opencv_test