lda.cpp 38 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
/*
 * Copyright (c) 2011. Philipp Wagner <bytefish[at]gmx[dot]de>.
 * Released to public domain under terms of the BSD Simplified license.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
 *   * Redistributions of source code must retain the above copyright
 *     notice, this list of conditions and the following disclaimer.
 *   * Redistributions in binary form must reproduce the above copyright
 *     notice, this list of conditions and the following disclaimer in the
 *     documentation and/or other materials provided with the distribution.
 *   * Neither the name of the organization nor the names of its contributors
 *     may be used to endorse or promote products derived from this software
 *     without specific prior written permission.
 *
 *   See <http://www.opensource.org/licenses/bsd-license>
 */

#include "precomp.hpp"
#include <iostream>
#include <map>
#include <set>

namespace cv
{

// Removes duplicate elements in a given vector.
template<typename _Tp>
29 30 31 32
inline std::vector<_Tp> remove_dups(const std::vector<_Tp>& src) {
    typedef typename std::set<_Tp>::const_iterator constSetIterator;
    typedef typename std::vector<_Tp>::const_iterator constVecIterator;
    std::set<_Tp> set_elems;
33 34
    for (constVecIterator it = src.begin(); it != src.end(); ++it)
        set_elems.insert(*it);
35
    std::vector<_Tp> elems;
36 37 38 39
    for (constSetIterator it = set_elems.begin(); it != set_elems.end(); ++it)
        elems.push_back(*it);
    return elems;
}
40

41 42 43
static Mat argsort(InputArray _src, bool ascending=true)
{
    Mat src = _src.getMat();
44
    if (src.rows != 1 && src.cols != 1) {
45
        String error_message = "Wrong shape of input matrix! Expected a matrix with one row or column.";
46
        CV_Error(Error::StsBadArg, error_message);
47
    }
48
    int flags = SORT_EVERY_ROW | (ascending ? SORT_ASCENDING : SORT_DESCENDING);
49 50 51 52 53
    Mat sorted_indices;
    sortIdx(src.reshape(1,1),sorted_indices,flags);
    return sorted_indices;
}

54 55 56
static Mat asRowMatrix(InputArrayOfArrays src, int rtype, double alpha=1, double beta=0) {
    // make sure the input data is a vector of matrices or vector of vector
    if(src.kind() != _InputArray::STD_VECTOR_MAT && src.kind() != _InputArray::STD_VECTOR_VECTOR) {
57
        String error_message = "The data is expected as InputArray::STD_VECTOR_MAT (a std::vector<Mat>) or _InputArray::STD_VECTOR_VECTOR (a std::vector< std::vector<...> >).";
58
        CV_Error(Error::StsBadArg, error_message);
59
    }
60
    // number of samples
61 62
    size_t n = src.total();
    // return empty matrix if no matrices given
63 64
    if(n == 0)
        return Mat();
65 66
    // dimensionality of (reshaped) samples
    size_t d = src.getMat(0).total();
67
    // create data matrix
68
    Mat data((int)n, (int)d, rtype);
69
    // now copy data
70
    for(int i = 0; i < (int)n; i++) {
71 72
        // make sure data can be reshaped, throw exception if not!
        if(src.getMat(i).total() != d) {
73
            String error_message = format("Wrong number of elements in matrix #%d! Expected %d was %d.", i, (int)d, (int)src.getMat(i).total());
74
            CV_Error(Error::StsBadArg, error_message);
75 76
        }
        // get a hold of the current row
77
        Mat xi = data.row(i);
78 79 80 81 82 83
        // make reshape happy by cloning for non-continuous matrices
        if(src.getMat(i).isContinuous()) {
            src.getMat(i).reshape(1, 1).convertTo(xi, rtype, alpha, beta);
        } else {
            src.getMat(i).clone().reshape(1, 1).convertTo(xi, rtype, alpha, beta);
        }
84 85 86
    }
    return data;
}
87 88

static void sortMatrixColumnsByIndices(InputArray _src, InputArray _indices, OutputArray _dst) {
89
    if(_indices.getMat().type() != CV_32SC1) {
90
        CV_Error(Error::StsUnsupportedFormat, "cv::sortColumnsByIndices only works on integer indices!");
91
    }
92
    Mat src = _src.getMat();
93
    std::vector<int> indices = _indices.getMat();
94 95
    _dst.create(src.rows, src.cols, src.type());
    Mat dst = _dst.getMat();
96
    for(size_t idx = 0; idx < indices.size(); idx++) {
97
        Mat originalCol = src.col(indices[idx]);
98
        Mat sortedCol = dst.col((int)idx);
99 100 101 102
        originalCol.copyTo(sortedCol);
    }
}

103
static Mat sortMatrixColumnsByIndices(InputArray src, InputArray indices) {
104 105 106 107
    Mat dst;
    sortMatrixColumnsByIndices(src, indices, dst);
    return dst;
}
108 109


110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166
template<typename _Tp> static bool
isSymmetric_(InputArray src) {
    Mat _src = src.getMat();
    if(_src.cols != _src.rows)
        return false;
    for (int i = 0; i < _src.rows; i++) {
        for (int j = 0; j < _src.cols; j++) {
            _Tp a = _src.at<_Tp> (i, j);
            _Tp b = _src.at<_Tp> (j, i);
            if (a != b) {
                return false;
            }
        }
    }
    return true;
}

template<typename _Tp> static bool
isSymmetric_(InputArray src, double eps) {
    Mat _src = src.getMat();
    if(_src.cols != _src.rows)
        return false;
    for (int i = 0; i < _src.rows; i++) {
        for (int j = 0; j < _src.cols; j++) {
            _Tp a = _src.at<_Tp> (i, j);
            _Tp b = _src.at<_Tp> (j, i);
            if (std::abs(a - b) > eps) {
                return false;
            }
        }
    }
    return true;
}

static bool isSymmetric(InputArray src, double eps=1e-16)
{
    Mat m = src.getMat();
    switch (m.type()) {
        case CV_8SC1: return isSymmetric_<char>(m); break;
        case CV_8UC1:
            return isSymmetric_<unsigned char>(m); break;
        case CV_16SC1:
            return isSymmetric_<short>(m); break;
        case CV_16UC1:
            return isSymmetric_<unsigned short>(m); break;
        case CV_32SC1:
            return isSymmetric_<int>(m); break;
        case CV_32FC1:
            return isSymmetric_<float>(m, eps); break;
        case CV_64FC1:
            return isSymmetric_<double>(m, eps); break;
        default:
            break;
    }
    return false;
}

167

168
//------------------------------------------------------------------------------
169
// cv::subspaceProject
170
//------------------------------------------------------------------------------
171
Mat subspaceProject(InputArray _W, InputArray _mean, InputArray _src) {
172 173 174 175
    // get data matrices
    Mat W = _W.getMat();
    Mat mean = _mean.getMat();
    Mat src = _src.getMat();
176 177 178 179 180
    // get number of samples and dimension
    int n = src.rows;
    int d = src.cols;
    // make sure the data has the correct shape
    if(W.rows != d) {
181
        String error_message = format("Wrong shapes for given matrices. Was size(src) = (%d,%d), size(W) = (%d,%d).", src.rows, src.cols, W.rows, W.cols);
182
        CV_Error(Error::StsBadArg, error_message);
183 184 185
    }
    // make sure mean is correct if not empty
    if(!mean.empty() && (mean.total() != (size_t) d)) {
186
        String error_message = format("Wrong mean shape for the given data matrix. Expected %d, but was %d.", d, mean.total());
187
        CV_Error(Error::StsBadArg, error_message);
188
    }
189 190
    // create temporary matrices
    Mat X, Y;
191
    // make sure you operate on correct type
192
    src.convertTo(X, W.type());
193 194
    // safe to do, because of above assertion
    if(!mean.empty()) {
195 196 197 198
        for(int i=0; i<n; i++) {
            Mat r_i = X.row(i);
            subtract(r_i, mean.reshape(1,1), r_i);
        }
199
    }
200 201 202 203 204 205
    // finally calculate projection as Y = (X-mean)*W
    gemm(X, W, 1.0, Mat(), 0.0, Y);
    return Y;
}

//------------------------------------------------------------------------------
206
// cv::subspaceReconstruct
207 208 209 210 211 212 213
//------------------------------------------------------------------------------
Mat subspaceReconstruct(InputArray _W, InputArray _mean, InputArray _src)
{
    // get data matrices
    Mat W = _W.getMat();
    Mat mean = _mean.getMat();
    Mat src = _src.getMat();
214
    // get number of samples and dimension
215
    int n = src.rows;
216 217 218
    int d = src.cols;
    // make sure the data has the correct shape
    if(W.cols != d) {
219
        String error_message = format("Wrong shapes for given matrices. Was size(src) = (%d,%d), size(W) = (%d,%d).", src.rows, src.cols, W.rows, W.cols);
220
        CV_Error(Error::StsBadArg, error_message);
221 222 223
    }
    // make sure mean is correct if not empty
    if(!mean.empty() && (mean.total() != (size_t) W.rows)) {
224
        String error_message = format("Wrong mean shape for the given eigenvector matrix. Expected %d, but was %d.", W.cols, mean.total());
225
        CV_Error(Error::StsBadArg, error_message);
226
    }
Ilya Lavrenov's avatar
Ilya Lavrenov committed
227
    // initialize temporary matrices
228 229 230 231
    Mat X, Y;
    // copy data & make sure we are using the correct type
    src.convertTo(Y, W.type());
    // calculate the reconstruction
232
    gemm(Y, W, 1.0, Mat(), 0.0, X, GEMM_2_T);
233 234
    // safe to do because of above assertion
    if(!mean.empty()) {
235 236 237 238
        for(int i=0; i<n; i++) {
            Mat r_i = X.row(i);
            add(r_i, mean.reshape(1,1), r_i);
        }
239
    }
240 241 242
    return X;
}

243

244 245
class EigenvalueDecomposition {
private:
246

247 248
    // Holds the data dimension.
    int n;
249

250 251
    // Stores real/imag part of a complex division.
    double cdivr, cdivi;
252

253 254 255
    // Pointer to internal memory.
    double *d, *e, *ort;
    double **V, **H;
256

257 258
    // Holds the computed eigenvalues.
    Mat _eigenvalues;
259

260 261
    // Holds the computed eigenvectors.
    Mat _eigenvectors;
262

263 264 265 266 267
    // Allocates memory.
    template<typename _Tp>
    _Tp *alloc_1d(int m) {
        return new _Tp[m];
    }
268

269 270 271 272 273 274 275 276
    // Allocates memory.
    template<typename _Tp>
    _Tp *alloc_1d(int m, _Tp val) {
        _Tp *arr = alloc_1d<_Tp> (m);
        for (int i = 0; i < m; i++)
            arr[i] = val;
        return arr;
    }
277

278 279
    // Allocates memory.
    template<typename _Tp>
Andrey Kamaev's avatar
Andrey Kamaev committed
280
    _Tp **alloc_2d(int m, int _n) {
281 282
        _Tp **arr = new _Tp*[m];
        for (int i = 0; i < m; i++)
Andrey Kamaev's avatar
Andrey Kamaev committed
283
            arr[i] = new _Tp[_n];
284 285
        return arr;
    }
286

287 288
    // Allocates memory.
    template<typename _Tp>
Andrey Kamaev's avatar
Andrey Kamaev committed
289 290
    _Tp **alloc_2d(int m, int _n, _Tp val) {
        _Tp **arr = alloc_2d<_Tp> (m, _n);
291
        for (int i = 0; i < m; i++) {
Andrey Kamaev's avatar
Andrey Kamaev committed
292
            for (int j = 0; j < _n; j++) {
293 294 295 296 297
                arr[i][j] = val;
            }
        }
        return arr;
    }
298

299
    void cdiv(double xr, double xi, double yr, double yi) {
Andrey Kamaev's avatar
Andrey Kamaev committed
300
        double r, dv;
301 302
        if (std::abs(yr) > std::abs(yi)) {
            r = yi / yr;
Andrey Kamaev's avatar
Andrey Kamaev committed
303 304 305
            dv = yr + r * yi;
            cdivr = (xr + r * xi) / dv;
            cdivi = (xi - r * xr) / dv;
306 307
        } else {
            r = yr / yi;
Andrey Kamaev's avatar
Andrey Kamaev committed
308 309 310
            dv = yi + r * yr;
            cdivr = (r * xr + xi) / dv;
            cdivi = (r * xi - xr) / dv;
311 312
        }
    }
313

314
    // Nonsymmetric reduction from Hessenberg to real Schur form.
315

316
    void hqr2() {
317

318 319 320 321
        //  This is derived from the Algol procedure hqr2,
        //  by Martin and Wilkinson, Handbook for Auto. Comp.,
        //  Vol.ii-Linear Algebra, and the corresponding
        //  Fortran subroutine in EISPACK.
322

323 324
        // Initialize
        int nn = this->n;
Andrey Kamaev's avatar
Andrey Kamaev committed
325
        int n1 = nn - 1;
326 327
        int low = 0;
        int high = nn - 1;
328
        double eps = std::pow(2.0, -52.0);
329 330
        double exshift = 0.0;
        double p = 0, q = 0, r = 0, s = 0, z = 0, t, w, x, y;
331

332
        // Store roots isolated by balanc and compute matrix norm
333

334 335
        double norm = 0.0;
        for (int i = 0; i < nn; i++) {
336
            if (i < low || i > high) {
337 338 339
                d[i] = H[i][i];
                e[i] = 0.0;
            }
340
            for (int j = std::max(i - 1, 0); j < nn; j++) {
341 342 343
                norm = norm + std::abs(H[i][j]);
            }
        }
344

345 346
        // Outer loop over eigenvalue index
        int iter = 0;
Andrey Kamaev's avatar
Andrey Kamaev committed
347
        while (n1 >= low) {
348

349
            // Look for single small sub-diagonal element
Andrey Kamaev's avatar
Andrey Kamaev committed
350
            int l = n1;
351 352 353 354 355 356 357 358 359 360
            while (l > low) {
                s = std::abs(H[l - 1][l - 1]) + std::abs(H[l][l]);
                if (s == 0.0) {
                    s = norm;
                }
                if (std::abs(H[l][l - 1]) < eps * s) {
                    break;
                }
                l--;
            }
361

362 363
            // Check for convergence
            // One root found
364

Andrey Kamaev's avatar
Andrey Kamaev committed
365 366 367 368 369
            if (l == n1) {
                H[n1][n1] = H[n1][n1] + exshift;
                d[n1] = H[n1][n1];
                e[n1] = 0.0;
                n1--;
370
                iter = 0;
371

372
                // Two roots found
373

Andrey Kamaev's avatar
Andrey Kamaev committed
374 375 376
            } else if (l == n1 - 1) {
                w = H[n1][n1 - 1] * H[n1 - 1][n1];
                p = (H[n1 - 1][n1 - 1] - H[n1][n1]) / 2.0;
377
                q = p * p + w;
378
                z = std::sqrt(std::abs(q));
Andrey Kamaev's avatar
Andrey Kamaev committed
379 380 381
                H[n1][n1] = H[n1][n1] + exshift;
                H[n1 - 1][n1 - 1] = H[n1 - 1][n1 - 1] + exshift;
                x = H[n1][n1];
382

383
                // Real pair
384

385 386 387 388 389 390
                if (q >= 0) {
                    if (p >= 0) {
                        z = p + z;
                    } else {
                        z = p - z;
                    }
Andrey Kamaev's avatar
Andrey Kamaev committed
391 392
                    d[n1 - 1] = x + z;
                    d[n1] = d[n1 - 1];
393
                    if (z != 0.0) {
Andrey Kamaev's avatar
Andrey Kamaev committed
394
                        d[n1] = x - w / z;
395
                    }
Andrey Kamaev's avatar
Andrey Kamaev committed
396 397 398
                    e[n1 - 1] = 0.0;
                    e[n1] = 0.0;
                    x = H[n1][n1 - 1];
399 400 401
                    s = std::abs(x) + std::abs(z);
                    p = x / s;
                    q = z / s;
402
                    r = std::sqrt(p * p + q * q);
403 404
                    p = p / r;
                    q = q / r;
405

406
                    // Row modification
407

Andrey Kamaev's avatar
Andrey Kamaev committed
408 409 410 411
                    for (int j = n1 - 1; j < nn; j++) {
                        z = H[n1 - 1][j];
                        H[n1 - 1][j] = q * z + p * H[n1][j];
                        H[n1][j] = q * H[n1][j] - p * z;
412
                    }
413

414
                    // Column modification
415

Andrey Kamaev's avatar
Andrey Kamaev committed
416 417 418 419
                    for (int i = 0; i <= n1; i++) {
                        z = H[i][n1 - 1];
                        H[i][n1 - 1] = q * z + p * H[i][n1];
                        H[i][n1] = q * H[i][n1] - p * z;
420
                    }
421

422
                    // Accumulate transformations
423

424
                    for (int i = low; i <= high; i++) {
Andrey Kamaev's avatar
Andrey Kamaev committed
425 426 427
                        z = V[i][n1 - 1];
                        V[i][n1 - 1] = q * z + p * V[i][n1];
                        V[i][n1] = q * V[i][n1] - p * z;
428
                    }
429

430
                    // Complex pair
431

432
                } else {
Andrey Kamaev's avatar
Andrey Kamaev committed
433 434 435 436
                    d[n1 - 1] = x + p;
                    d[n1] = x + p;
                    e[n1 - 1] = z;
                    e[n1] = -z;
437
                }
Andrey Kamaev's avatar
Andrey Kamaev committed
438
                n1 = n1 - 2;
439
                iter = 0;
440

441
                // No convergence yet
442

443
            } else {
444

445
                // Form shift
446

Andrey Kamaev's avatar
Andrey Kamaev committed
447
                x = H[n1][n1];
448 449
                y = 0.0;
                w = 0.0;
Andrey Kamaev's avatar
Andrey Kamaev committed
450 451 452
                if (l < n1) {
                    y = H[n1 - 1][n1 - 1];
                    w = H[n1][n1 - 1] * H[n1 - 1][n1];
453
                }
454

455
                // Wilkinson's original ad hoc shift
456

457 458
                if (iter == 10) {
                    exshift += x;
Andrey Kamaev's avatar
Andrey Kamaev committed
459
                    for (int i = low; i <= n1; i++) {
460 461
                        H[i][i] -= x;
                    }
Andrey Kamaev's avatar
Andrey Kamaev committed
462
                    s = std::abs(H[n1][n1 - 1]) + std::abs(H[n1 - 1][n1 - 2]);
463 464 465
                    x = y = 0.75 * s;
                    w = -0.4375 * s * s;
                }
466

467
                // MATLAB's new ad hoc shift
468

469 470 471 472
                if (iter == 30) {
                    s = (y - x) / 2.0;
                    s = s * s + w;
                    if (s > 0) {
473
                        s = std::sqrt(s);
474 475 476 477
                        if (y < x) {
                            s = -s;
                        }
                        s = x - w / ((y - x) / 2.0 + s);
Andrey Kamaev's avatar
Andrey Kamaev committed
478
                        for (int i = low; i <= n1; i++) {
479 480 481 482 483 484
                            H[i][i] -= s;
                        }
                        exshift += s;
                        x = y = w = 0.964;
                    }
                }
485

486
                iter = iter + 1; // (Could check iteration count here.)
487

488
                // Look for two consecutive small sub-diagonal elements
Andrey Kamaev's avatar
Andrey Kamaev committed
489
                int m = n1 - 2;
490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510
                while (m >= l) {
                    z = H[m][m];
                    r = x - z;
                    s = y - z;
                    p = (r * s - w) / H[m + 1][m] + H[m][m + 1];
                    q = H[m + 1][m + 1] - z - r - s;
                    r = H[m + 2][m + 1];
                    s = std::abs(p) + std::abs(q) + std::abs(r);
                    p = p / s;
                    q = q / s;
                    r = r / s;
                    if (m == l) {
                        break;
                    }
                    if (std::abs(H[m][m - 1]) * (std::abs(q) + std::abs(r)) < eps * (std::abs(p)
                                                                                     * (std::abs(H[m - 1][m - 1]) + std::abs(z) + std::abs(
                                                                                                                                           H[m + 1][m + 1])))) {
                        break;
                    }
                    m--;
                }
511

Andrey Kamaev's avatar
Andrey Kamaev committed
512
                for (int i = m + 2; i <= n1; i++) {
513 514 515 516 517
                    H[i][i - 2] = 0.0;
                    if (i > m + 2) {
                        H[i][i - 3] = 0.0;
                    }
                }
518

519
                // Double QR step involving rows l:n and columns m:n
520

Andrey Kamaev's avatar
Andrey Kamaev committed
521 522
                for (int k = m; k <= n1 - 1; k++) {
                    bool notlast = (k != n1 - 1);
523 524 525 526 527 528 529 530 531 532 533 534 535 536
                    if (k != m) {
                        p = H[k][k - 1];
                        q = H[k + 1][k - 1];
                        r = (notlast ? H[k + 2][k - 1] : 0.0);
                        x = std::abs(p) + std::abs(q) + std::abs(r);
                        if (x != 0.0) {
                            p = p / x;
                            q = q / x;
                            r = r / x;
                        }
                    }
                    if (x == 0.0) {
                        break;
                    }
537
                    s = std::sqrt(p * p + q * q + r * r);
538 539 540 541 542 543 544 545 546 547 548 549 550 551 552
                    if (p < 0) {
                        s = -s;
                    }
                    if (s != 0) {
                        if (k != m) {
                            H[k][k - 1] = -s * x;
                        } else if (l != m) {
                            H[k][k - 1] = -H[k][k - 1];
                        }
                        p = p + s;
                        x = p / s;
                        y = q / s;
                        z = r / s;
                        q = q / p;
                        r = r / p;
553

554
                        // Row modification
555

556 557 558 559 560 561 562 563 564
                        for (int j = k; j < nn; j++) {
                            p = H[k][j] + q * H[k + 1][j];
                            if (notlast) {
                                p = p + r * H[k + 2][j];
                                H[k + 2][j] = H[k + 2][j] - p * z;
                            }
                            H[k][j] = H[k][j] - p * x;
                            H[k + 1][j] = H[k + 1][j] - p * y;
                        }
565

566
                        // Column modification
567

568
                        for (int i = 0; i <= std::min(n1, k + 3); i++) {
569 570 571 572 573 574 575 576
                            p = x * H[i][k] + y * H[i][k + 1];
                            if (notlast) {
                                p = p + z * H[i][k + 2];
                                H[i][k + 2] = H[i][k + 2] - p * r;
                            }
                            H[i][k] = H[i][k] - p;
                            H[i][k + 1] = H[i][k + 1] - p * q;
                        }
577

578
                        // Accumulate transformations
579

580 581 582 583 584 585 586 587 588 589 590 591
                        for (int i = low; i <= high; i++) {
                            p = x * V[i][k] + y * V[i][k + 1];
                            if (notlast) {
                                p = p + z * V[i][k + 2];
                                V[i][k + 2] = V[i][k + 2] - p * r;
                            }
                            V[i][k] = V[i][k] - p;
                            V[i][k + 1] = V[i][k + 1] - p * q;
                        }
                    } // (s != 0)
                } // k loop
            } // check convergence
Andrey Kamaev's avatar
Andrey Kamaev committed
592
        } // while (n1 >= low)
593

594
        // Backsubstitute to find vectors of upper triangular form
595

596 597 598
        if (norm == 0.0) {
            return;
        }
599

Andrey Kamaev's avatar
Andrey Kamaev committed
600 601 602
        for (n1 = nn - 1; n1 >= 0; n1--) {
            p = d[n1];
            q = e[n1];
603

604
            // Real vector
605

606
            if (q == 0) {
Andrey Kamaev's avatar
Andrey Kamaev committed
607 608 609
                int l = n1;
                H[n1][n1] = 1.0;
                for (int i = n1 - 1; i >= 0; i--) {
610 611
                    w = H[i][i] - p;
                    r = 0.0;
Andrey Kamaev's avatar
Andrey Kamaev committed
612 613
                    for (int j = l; j <= n1; j++) {
                        r = r + H[i][j] * H[j][n1];
614 615 616 617 618 619 620 621
                    }
                    if (e[i] < 0.0) {
                        z = w;
                        s = r;
                    } else {
                        l = i;
                        if (e[i] == 0.0) {
                            if (w != 0.0) {
Andrey Kamaev's avatar
Andrey Kamaev committed
622
                                H[i][n1] = -r / w;
623
                            } else {
Andrey Kamaev's avatar
Andrey Kamaev committed
624
                                H[i][n1] = -r / (eps * norm);
625
                            }
626

627
                            // Solve real equations
628

629 630 631 632 633
                        } else {
                            x = H[i][i + 1];
                            y = H[i + 1][i];
                            q = (d[i] - p) * (d[i] - p) + e[i] * e[i];
                            t = (x * s - z * r) / q;
Andrey Kamaev's avatar
Andrey Kamaev committed
634
                            H[i][n1] = t;
635
                            if (std::abs(x) > std::abs(z)) {
Andrey Kamaev's avatar
Andrey Kamaev committed
636
                                H[i + 1][n1] = (-r - w * t) / x;
637
                            } else {
Andrey Kamaev's avatar
Andrey Kamaev committed
638
                                H[i + 1][n1] = (-s - y * t) / z;
639 640
                            }
                        }
641

642
                        // Overflow control
643

Andrey Kamaev's avatar
Andrey Kamaev committed
644
                        t = std::abs(H[i][n1]);
645
                        if ((eps * t) * t > 1) {
Andrey Kamaev's avatar
Andrey Kamaev committed
646 647
                            for (int j = i; j <= n1; j++) {
                                H[j][n1] = H[j][n1] / t;
648 649 650 651 652 653
                            }
                        }
                    }
                }
                // Complex vector
            } else if (q < 0) {
Andrey Kamaev's avatar
Andrey Kamaev committed
654
                int l = n1 - 1;
655

656
                // Last vector component imaginary so matrix is triangular
657

Andrey Kamaev's avatar
Andrey Kamaev committed
658 659 660
                if (std::abs(H[n1][n1 - 1]) > std::abs(H[n1 - 1][n1])) {
                    H[n1 - 1][n1 - 1] = q / H[n1][n1 - 1];
                    H[n1 - 1][n1] = -(H[n1][n1] - p) / H[n1][n1 - 1];
661
                } else {
Andrey Kamaev's avatar
Andrey Kamaev committed
662 663 664
                    cdiv(0.0, -H[n1 - 1][n1], H[n1 - 1][n1 - 1] - p, q);
                    H[n1 - 1][n1 - 1] = cdivr;
                    H[n1 - 1][n1] = cdivi;
665
                }
Andrey Kamaev's avatar
Andrey Kamaev committed
666 667 668
                H[n1][n1 - 1] = 0.0;
                H[n1][n1] = 1.0;
                for (int i = n1 - 2; i >= 0; i--) {
669 670 671
                    double ra, sa, vr, vi;
                    ra = 0.0;
                    sa = 0.0;
Andrey Kamaev's avatar
Andrey Kamaev committed
672 673 674
                    for (int j = l; j <= n1; j++) {
                        ra = ra + H[i][j] * H[j][n1 - 1];
                        sa = sa + H[i][j] * H[j][n1];
675 676
                    }
                    w = H[i][i] - p;
677

678 679 680 681 682 683 684 685
                    if (e[i] < 0.0) {
                        z = w;
                        r = ra;
                        s = sa;
                    } else {
                        l = i;
                        if (e[i] == 0) {
                            cdiv(-ra, -sa, w, q);
Andrey Kamaev's avatar
Andrey Kamaev committed
686 687
                            H[i][n1 - 1] = cdivr;
                            H[i][n1] = cdivi;
688
                        } else {
689

690
                            // Solve complex equations
691

692 693 694 695
                            x = H[i][i + 1];
                            y = H[i + 1][i];
                            vr = (d[i] - p) * (d[i] - p) + e[i] * e[i] - q * q;
                            vi = (d[i] - p) * 2.0 * q;
696
                            if (vr == 0.0 && vi == 0.0) {
697 698 699 700 701
                                vr = eps * norm * (std::abs(w) + std::abs(q) + std::abs(x)
                                                   + std::abs(y) + std::abs(z));
                            }
                            cdiv(x * r - z * ra + q * sa,
                                 x * s - z * sa - q * ra, vr, vi);
Andrey Kamaev's avatar
Andrey Kamaev committed
702 703
                            H[i][n1 - 1] = cdivr;
                            H[i][n1] = cdivi;
704
                            if (std::abs(x) > (std::abs(z) + std::abs(q))) {
Andrey Kamaev's avatar
Andrey Kamaev committed
705 706 707
                                H[i + 1][n1 - 1] = (-ra - w * H[i][n1 - 1] + q
                                                   * H[i][n1]) / x;
                                H[i + 1][n1] = (-sa - w * H[i][n1] - q * H[i][n1
708 709
                                                                            - 1]) / x;
                            } else {
Andrey Kamaev's avatar
Andrey Kamaev committed
710
                                cdiv(-r - y * H[i][n1 - 1], -s - y * H[i][n1], z,
711
                                     q);
Andrey Kamaev's avatar
Andrey Kamaev committed
712 713
                                H[i + 1][n1 - 1] = cdivr;
                                H[i + 1][n1] = cdivi;
714 715
                            }
                        }
716

717
                        // Overflow control
718

719
                        t = std::max(std::abs(H[i][n1 - 1]), std::abs(H[i][n1]));
720
                        if ((eps * t) * t > 1) {
Andrey Kamaev's avatar
Andrey Kamaev committed
721 722 723
                            for (int j = i; j <= n1; j++) {
                                H[j][n1 - 1] = H[j][n1 - 1] / t;
                                H[j][n1] = H[j][n1] / t;
724 725 726 727 728 729
                            }
                        }
                    }
                }
            }
        }
730

731
        // Vectors of isolated roots
732

733
        for (int i = 0; i < nn; i++) {
734
            if (i < low || i > high) {
735 736 737 738 739
                for (int j = i; j < nn; j++) {
                    V[i][j] = H[i][j];
                }
            }
        }
740

741
        // Back transformation to get eigenvectors of original matrix
742

743 744 745
        for (int j = nn - 1; j >= low; j--) {
            for (int i = low; i <= high; i++) {
                z = 0.0;
746
                for (int k = low; k <= std::min(j, high); k++) {
747 748 749 750 751 752
                    z = z + V[i][k] * H[k][j];
                }
                V[i][j] = z;
            }
        }
    }
753

754 755 756 757 758 759 760 761
    // Nonsymmetric reduction to Hessenberg form.
    void orthes() {
        //  This is derived from the Algol procedures orthes and ortran,
        //  by Martin and Wilkinson, Handbook for Auto. Comp.,
        //  Vol.ii-Linear Algebra, and the corresponding
        //  Fortran subroutines in EISPACK.
        int low = 0;
        int high = n - 1;
762

763
        for (int m = low + 1; m <= high - 1; m++) {
764

765
            // Scale column.
766

767 768 769 770 771
            double scale = 0.0;
            for (int i = m; i <= high; i++) {
                scale = scale + std::abs(H[i][m - 1]);
            }
            if (scale != 0.0) {
772

773
                // Compute Householder transformation.
774

775 776 777 778 779
                double h = 0.0;
                for (int i = high; i >= m; i--) {
                    ort[i] = H[i][m - 1] / scale;
                    h += ort[i] * ort[i];
                }
780
                double g = std::sqrt(h);
781 782 783 784 785
                if (ort[m] > 0) {
                    g = -g;
                }
                h = h - ort[m] * g;
                ort[m] = ort[m] - g;
786

787 788
                // Apply Householder similarity transformation
                // H = (I-u*u'/h)*H*(I-u*u')/h)
789

790 791 792 793 794 795 796 797 798 799
                for (int j = m; j < n; j++) {
                    double f = 0.0;
                    for (int i = high; i >= m; i--) {
                        f += ort[i] * H[i][j];
                    }
                    f = f / h;
                    for (int i = m; i <= high; i++) {
                        H[i][j] -= f * ort[i];
                    }
                }
800

801 802 803 804 805 806 807 808 809 810 811 812 813 814
                for (int i = 0; i <= high; i++) {
                    double f = 0.0;
                    for (int j = high; j >= m; j--) {
                        f += ort[j] * H[i][j];
                    }
                    f = f / h;
                    for (int j = m; j <= high; j++) {
                        H[i][j] -= f * ort[j];
                    }
                }
                ort[m] = scale * ort[m];
                H[m][m - 1] = scale * g;
            }
        }
815

816
        // Accumulate transformations (Algol's ortran).
817

818 819 820 821 822
        for (int i = 0; i < n; i++) {
            for (int j = 0; j < n; j++) {
                V[i][j] = (i == j ? 1.0 : 0.0);
            }
        }
823

824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842
        for (int m = high - 1; m >= low + 1; m--) {
            if (H[m][m - 1] != 0.0) {
                for (int i = m + 1; i <= high; i++) {
                    ort[i] = H[i][m - 1];
                }
                for (int j = m; j <= high; j++) {
                    double g = 0.0;
                    for (int i = m; i <= high; i++) {
                        g += ort[i] * V[i][j];
                    }
                    // Double division avoids possible underflow
                    g = (g / ort[m]) / H[m][m - 1];
                    for (int i = m; i <= high; i++) {
                        V[i][j] += g * ort[i];
                    }
                }
            }
        }
    }
843

844 845 846 847 848 849 850 851 852 853 854 855 856
    // Releases all internal working memory.
    void release() {
        // releases the working data
        delete[] d;
        delete[] e;
        delete[] ort;
        for (int i = 0; i < n; i++) {
            delete[] H[i];
            delete[] V[i];
        }
        delete[] H;
        delete[] V;
    }
857

858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881
    // Computes the Eigenvalue Decomposition for a matrix given in H.
    void compute() {
        // Allocate memory for the working data.
        V = alloc_2d<double> (n, n, 0.0);
        d = alloc_1d<double> (n);
        e = alloc_1d<double> (n);
        ort = alloc_1d<double> (n);
        // Reduce to Hessenberg form.
        orthes();
        // Reduce Hessenberg to real Schur form.
        hqr2();
        // Copy eigenvalues to OpenCV Matrix.
        _eigenvalues.create(1, n, CV_64FC1);
        for (int i = 0; i < n; i++) {
            _eigenvalues.at<double> (0, i) = d[i];
        }
        // Copy eigenvectors to OpenCV Matrix.
        _eigenvectors.create(n, n, CV_64FC1);
        for (int i = 0; i < n; i++)
            for (int j = 0; j < n; j++)
                _eigenvectors.at<double> (i, j) = V[i][j];
        // Deallocate the memory by releasing all internal working data.
        release();
    }
882

883 884 885
public:
    EigenvalueDecomposition()
    : n(0) { }
886

887 888 889 890 891 892 893
    // Initializes & computes the Eigenvalue Decomposition for a general matrix
    // given in src. This function is a port of the EigenvalueSolver in JAMA,
    // which has been released to public domain by The MathWorks and the
    // National Institute of Standards and Technology (NIST).
    EigenvalueDecomposition(InputArray src) {
        compute(src);
    }
894

895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925
    // This function computes the Eigenvalue Decomposition for a general matrix
    // given in src. This function is a port of the EigenvalueSolver in JAMA,
    // which has been released to public domain by The MathWorks and the
    // National Institute of Standards and Technology (NIST).
    void compute(InputArray src)
    {
        if(isSymmetric(src)) {
            // Fall back to OpenCV for a symmetric matrix!
            cv::eigen(src, _eigenvalues, _eigenvectors);
        } else {
            Mat tmp;
            // Convert the given input matrix to double. Is there any way to
            // prevent allocating the temporary memory? Only used for copying
            // into working memory and deallocated after.
            src.getMat().convertTo(tmp, CV_64FC1);
            // Get dimension of the matrix.
            this->n = tmp.cols;
            // Allocate the matrix data to work on.
            this->H = alloc_2d<double> (n, n);
            // Now safely copy the data.
            for (int i = 0; i < tmp.rows; i++) {
                for (int j = 0; j < tmp.cols; j++) {
                    this->H[i][j] = tmp.at<double>(i, j);
                }
            }
            // Deallocates the temporary matrix before computing.
            tmp.release();
            // Performs the eigenvalue decomposition of H.
            compute();
        }
    }
926

927
    ~EigenvalueDecomposition() {}
928

929 930 931 932 933 934 935 936 937 938
    // Returns the eigenvalues of the Eigenvalue Decomposition.
    Mat eigenvalues() {    return _eigenvalues; }
    // Returns the eigenvectors of the Eigenvalue Decomposition.
    Mat eigenvectors() { return _eigenvectors; }
};


//------------------------------------------------------------------------------
// Linear Discriminant Analysis implementation
//------------------------------------------------------------------------------
939
void LDA::save(const String& filename) const {
940
    FileStorage fs(filename, FileStorage::WRITE);
941
    if (!fs.isOpened()) {
942
        CV_Error(Error::StsError, "File can't be opened for writing!");
943
    }
944 945 946 947 948
    this->save(fs);
    fs.release();
}

// Deserializes this object from a given filename.
949
void LDA::load(const String& filename) {
950 951
    FileStorage fs(filename, FileStorage::READ);
    if (!fs.isOpened())
952
       CV_Error(Error::StsError, "File can't be opened for writing!");
953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972
    this->load(fs);
    fs.release();
}

// Serializes this object to a given FileStorage.
void LDA::save(FileStorage& fs) const {
    // write matrices
    fs << "num_components" << _num_components;
    fs << "eigenvalues" << _eigenvalues;
    fs << "eigenvectors" << _eigenvectors;
}

// Deserializes this object from a given FileStorage.
void LDA::load(const FileStorage& fs) {
    //read matrices
    fs["num_components"] >> _num_components;
    fs["eigenvalues"] >> _eigenvalues;
    fs["eigenvectors"] >> _eigenvectors;
}

973
void LDA::lda(InputArrayOfArrays _src, InputArray _lbls) {
974 975
    // get data
    Mat src = _src.getMat();
976
    std::vector<int> labels;
977 978 979 980 981 982 983
    // safely copy the labels
    {
        Mat tmp = _lbls.getMat();
        for(unsigned int i = 0; i < tmp.total(); i++) {
            labels.push_back(tmp.at<int>(i));
        }
    }
984 985 986 987 988
    // turn into row sampled matrix
    Mat data;
    // ensure working matrix is double precision
    src.convertTo(data, CV_64FC1);
    // maps the labels, so they're ascending: [0,1,...,C]
989 990 991
    std::vector<int> mapped_labels(labels.size());
    std::vector<int> num2label = remove_dups(labels);
    std::map<int, int> label2num;
992
    for (int i = 0; i < (int)num2label.size(); i++)
993
        label2num[num2label[i]] = i;
994
    for (size_t i = 0; i < labels.size(); i++)
995 996 997 998 999
        mapped_labels[i] = label2num[labels[i]];
    // get sample size, dimension
    int N = data.rows;
    int D = data.cols;
    // number of unique labels
1000
    int C = (int)num2label.size();
1001 1002 1003
    // we can't do a LDA on one class, what do you
    // want to separate from each other then?
    if(C == 1) {
1004
        String error_message = "At least two classes are needed to perform a LDA. Reason: Only one class was given!";
1005
        CV_Error(Error::StsBadArg, error_message);
1006
    }
1007
    // throw error if less labels, than samples
1008
    if (labels.size() != static_cast<size_t>(N)) {
1009
        String error_message = format("The number of samples must equal the number of labels. Given %d labels, %d samples. ", labels.size(), N);
1010
        CV_Error(Error::StsBadArg, error_message);
1011
    }
1012
    // warn if within-classes scatter matrix becomes singular
1013
    if (N < D) {
1014 1015 1016
        std::cout << "Warning: Less observations than feature dimension given!"
                  << "Computation will probably fail."
                  << std::endl;
1017
    }
1018
    // clip number of components to be a valid number
1019
    if ((_num_components <= 0) || (_num_components > (C - 1))) {
1020
        _num_components = (C - 1);
1021
    }
1022 1023 1024
    // holds the mean over all classes
    Mat meanTotal = Mat::zeros(1, D, data.type());
    // holds the mean for each class
1025 1026
    std::vector<Mat> meanClass(C);
    std::vector<int> numClass(C);
1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039
    // initialize
    for (int i = 0; i < C; i++) {
        numClass[i] = 0;
        meanClass[i] = Mat::zeros(1, D, data.type()); //! Dx1 image vector
    }
    // calculate sums
    for (int i = 0; i < N; i++) {
        Mat instance = data.row(i);
        int classIdx = mapped_labels[i];
        add(meanTotal, instance, meanTotal);
        add(meanClass[classIdx], instance, meanClass[classIdx]);
        numClass[classIdx]++;
    }
1040 1041 1042 1043 1044 1045
    // calculate total mean
    meanTotal.convertTo(meanTotal, meanTotal.type(), 1.0 / static_cast<double> (N));
    // calculate class means
    for (int i = 0; i < C; i++) {
        meanClass[i].convertTo(meanClass[i], meanClass[i].type(), 1.0 / static_cast<double> (numClass[i]));
    }
1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073
    // subtract class means
    for (int i = 0; i < N; i++) {
        int classIdx = mapped_labels[i];
        Mat instance = data.row(i);
        subtract(instance, meanClass[classIdx], instance);
    }
    // calculate within-classes scatter
    Mat Sw = Mat::zeros(D, D, data.type());
    mulTransposed(data, Sw, true);
    // calculate between-classes scatter
    Mat Sb = Mat::zeros(D, D, data.type());
    for (int i = 0; i < C; i++) {
        Mat tmp;
        subtract(meanClass[i], meanTotal, tmp);
        mulTransposed(tmp, tmp, true);
        add(Sb, tmp, Sb);
    }
    // invert Sw
    Mat Swi = Sw.inv();
    // M = inv(Sw)*Sb
    Mat M;
    gemm(Swi, Sb, 1.0, Mat(), 0.0, M);
    EigenvalueDecomposition es(M);
    _eigenvalues = es.eigenvalues();
    _eigenvectors = es.eigenvectors();
    // reshape eigenvalues, so they are stored by column
    _eigenvalues = _eigenvalues.reshape(1, 1);
    // get sorted indices descending by their eigenvalue
1074
    std::vector<int> sorted_indices = argsort(_eigenvalues, false);
1075 1076 1077 1078 1079 1080 1081 1082
    // now sort eigenvalues and eigenvectors accordingly
    _eigenvalues = sortMatrixColumnsByIndices(_eigenvalues, sorted_indices);
    _eigenvectors = sortMatrixColumnsByIndices(_eigenvectors, sorted_indices);
    // and now take only the num_components and we're out!
    _eigenvalues = Mat(_eigenvalues, Range::all(), Range(0, _num_components));
    _eigenvectors = Mat(_eigenvectors, Range::all(), Range(0, _num_components));
}

1083
void LDA::compute(InputArrayOfArrays _src, InputArray _lbls) {
1084 1085 1086 1087 1088 1089 1090 1091
    switch(_src.kind()) {
    case _InputArray::STD_VECTOR_MAT:
        lda(asRowMatrix(_src, CV_64FC1), _lbls);
        break;
    case _InputArray::MAT:
        lda(_src.getMat(), _lbls);
        break;
    default:
1092
        String error_message= format("InputArray Datatype %d is not supported.", _src.kind());
1093
        CV_Error(Error::StsBadArg, error_message);
1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106
        break;
    }
}

// Projects samples into the LDA subspace.
Mat LDA::project(InputArray src) {
   return subspaceProject(_eigenvectors, Mat(), _dataAsRow ? src : src.getMat().t());
}

// Reconstructs projections from the LDA subspace.
Mat LDA::reconstruct(InputArray src) {
   return subspaceReconstruct(_eigenvectors, Mat(), _dataAsRow ? src : src.getMat().t());
}
1107

1108
}