tesseract/unittest/linlsq_test.cc

116 lines
4.0 KiB
C++
Raw Normal View History

// (C) Copyright 2017, Google Inc.
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
// http://www.apache.org/licenses/LICENSE-2.0
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "linlsq.h"
#include "include_gunit.h"
namespace {
class LLSQTest : public testing::Test {
public:
void SetUp() {}
void TearDown() {}
void ExpectCorrectLine(const LLSQ& llsq, double m, double c, double rms,
double pearson, double tolerance) {
EXPECT_NEAR(m, llsq.m(), tolerance);
EXPECT_NEAR(c, llsq.c(llsq.m()), tolerance);
EXPECT_NEAR(rms, llsq.rms(llsq.m(), llsq.c(llsq.m())), tolerance);
EXPECT_NEAR(pearson, llsq.pearson(), tolerance);
}
FCOORD PtsMean(const std::vector<FCOORD>& pts) {
FCOORD total(0, 0);
for (int i = 0; i < pts.size(); i++) {
total += pts[i];
}
return (pts.size() > 0) ? total / pts.size() : total;
}
void VerifyRmsOrth(const std::vector<FCOORD>& pts, const FCOORD& orth) {
LLSQ llsq;
FCOORD xavg = PtsMean(pts);
FCOORD nvec = !orth;
nvec.normalise();
double expected_answer = 0;
for (int i = 0; i < pts.size(); i++) {
llsq.add(pts[i].x(), pts[i].y());
double dot = nvec % (pts[i] - xavg);
expected_answer += dot * dot;
}
expected_answer /= pts.size();
expected_answer = sqrt(expected_answer);
EXPECT_NEAR(expected_answer, llsq.rms_orth(orth), 0.0001);
}
void ExpectCorrectVector(const LLSQ& llsq, FCOORD correct_mean_pt,
FCOORD correct_vector, float tolerance) {
FCOORD mean_pt = llsq.mean_point();
FCOORD vector = llsq.vector_fit();
EXPECT_NEAR(correct_mean_pt.x(), mean_pt.x(), tolerance);
EXPECT_NEAR(correct_mean_pt.y(), mean_pt.y(), tolerance);
EXPECT_NEAR(correct_vector.x(), vector.x(), tolerance);
EXPECT_NEAR(correct_vector.y(), vector.y(), tolerance);
}
};
// Tests a simple baseline-style normalization.
TEST_F(LLSQTest, BasicLines) {
LLSQ llsq;
llsq.add(1.0, 1.0);
llsq.add(2.0, 2.0);
ExpectCorrectLine(llsq, 1.0, 0.0, 0.0, 1.0, 1e-6);
float half_root_2 = sqrt(2.0) / 2.0f;
ExpectCorrectVector(llsq, FCOORD(1.5f, 1.5f),
FCOORD(half_root_2, half_root_2), 1e-6);
llsq.remove(2.0, 2.0);
llsq.add(1.0, 2.0);
llsq.add(10.0, 1.0);
llsq.add(-8.0, 1.0);
// The point at 1,2 pulls the result away from what would otherwise be a
// perfect fit to a horizontal line by 0.25 unit, with rms error of 0.433.
ExpectCorrectLine(llsq, 0.0, 1.25, 0.433, 0.0, 1e-2);
ExpectCorrectVector(llsq, FCOORD(1.0f, 1.25f), FCOORD(1.0f, 0.0f), 1e-3);
llsq.add(1.0, 2.0, 10.0);
// With a heavy weight, the point at 1,2 pulls the line nearer.
ExpectCorrectLine(llsq, 0.0, 1.786, 0.41, 0.0, 1e-2);
ExpectCorrectVector(llsq, FCOORD(1.0f, 1.786f), FCOORD(1.0f, 0.0f), 1e-3);
}
// Tests a simple baseline-style normalization with a rotation.
TEST_F(LLSQTest, Vectors) {
LLSQ llsq;
llsq.add(1.0, 1.0);
llsq.add(1.0, -1.0);
ExpectCorrectVector(llsq, FCOORD(1.0f, 0.0f), FCOORD(0.0f, 1.0f), 1e-6);
llsq.add(0.9, -2.0);
llsq.add(1.1, -3.0);
llsq.add(0.9, 2.0);
llsq.add(1.10001, 3.0);
ExpectCorrectVector(llsq, FCOORD(1.0f, 0.0f), FCOORD(0.0f, 1.0f), 1e-3);
}
// Verify that rms_orth() actually calculates:
// sqrt( sum (!nvec * (x_i - x_avg))^2 / n)
TEST_F(LLSQTest, RmsOrthWorksAsIntended) {
std::vector<FCOORD> pts;
pts.push_back(FCOORD(0.56, 0.95));
pts.push_back(FCOORD(0.09, 0.09));
pts.push_back(FCOORD(0.13, 0.77));
pts.push_back(FCOORD(0.16, 0.83));
pts.push_back(FCOORD(0.45, 0.79));
VerifyRmsOrth(pts, FCOORD(1, 0));
VerifyRmsOrth(pts, FCOORD(1, 1));
VerifyRmsOrth(pts, FCOORD(1, 2));
VerifyRmsOrth(pts, FCOORD(2, 1));
}
} // namespace.