1 | //---------------------------------------------------------------------- |
---|
2 | // File: ANN.h |
---|
3 | // Programmer: Sunil Arya and David Mount |
---|
4 | // Last modified: 05/03/05 (Release 1.1) |
---|
5 | // Description: Basic include file for approximate nearest |
---|
6 | // neighbor searching. |
---|
7 | //---------------------------------------------------------------------- |
---|
8 | // Copyright (c) 1997-2005 University of Maryland and Sunil Arya and |
---|
9 | // David Mount. All Rights Reserved. |
---|
10 | // |
---|
11 | // This software and related documentation is part of the Approximate |
---|
12 | // Nearest Neighbor Library (ANN). This software is provided under |
---|
13 | // the provisions of the Lesser GNU Public License (LGPL). See the |
---|
14 | // file ../ReadMe.txt for further information. |
---|
15 | // |
---|
16 | // The University of Maryland (U.M.) and the authors make no |
---|
17 | // representations about the suitability or fitness of this software for |
---|
18 | // any purpose. It is provided "as is" without express or implied |
---|
19 | // warranty. |
---|
20 | //---------------------------------------------------------------------- |
---|
21 | // History: |
---|
22 | // Revision 0.1 03/04/98 |
---|
23 | // Initial release |
---|
24 | // Revision 1.0 04/01/05 |
---|
25 | // Added copyright and revision information |
---|
26 | // Added ANNcoordPrec for coordinate precision. |
---|
27 | // Added methods theDim, nPoints, maxPoints, thePoints to ANNpointSet. |
---|
28 | // Cleaned up C++ structure for modern compilers |
---|
29 | // Revision 1.1 05/03/05 |
---|
30 | // Added fixed-radius k-NN searching |
---|
31 | //---------------------------------------------------------------------- |
---|
32 | |
---|
33 | //---------------------------------------------------------------------- |
---|
34 | // ANN - approximate nearest neighbor searching |
---|
35 | // ANN is a library for approximate nearest neighbor searching, |
---|
36 | // based on the use of standard and priority search in kd-trees |
---|
37 | // and balanced box-decomposition (bbd) trees. Here are some |
---|
38 | // references to the main algorithmic techniques used here: |
---|
39 | // |
---|
40 | // kd-trees: |
---|
41 | // Friedman, Bentley, and Finkel, ``An algorithm for finding |
---|
42 | // best matches in logarithmic expected time,'' ACM |
---|
43 | // Transactions on Mathematical Software, 3(3):209-226, 1977. |
---|
44 | // |
---|
45 | // Priority search in kd-trees: |
---|
46 | // Arya and Mount, ``Algorithms for fast vector quantization,'' |
---|
47 | // Proc. of DCC '93: Data Compression Conference, eds. J. A. |
---|
48 | // Storer and M. Cohn, IEEE Press, 1993, 381-390. |
---|
49 | // |
---|
50 | // Approximate nearest neighbor search and bbd-trees: |
---|
51 | // Arya, Mount, Netanyahu, Silverman, and Wu, ``An optimal |
---|
52 | // algorithm for approximate nearest neighbor searching,'' |
---|
53 | // 5th Ann. ACM-SIAM Symposium on Discrete Algorithms, |
---|
54 | // 1994, 573-582. |
---|
55 | //---------------------------------------------------------------------- |
---|
56 | |
---|
57 | #ifndef ANN_H |
---|
58 | #define ANN_H |
---|
59 | |
---|
60 | #ifdef WIN32 |
---|
61 | //---------------------------------------------------------------------- |
---|
62 | // For Microsoft Visual C++, externally accessible symbols must be |
---|
63 | // explicitly indicated with DLL_API, which is somewhat like "extern." |
---|
64 | // |
---|
65 | // The following ifdef block is the standard way of creating macros |
---|
66 | // which make exporting from a DLL simpler. All files within this DLL |
---|
67 | // are compiled with the DLL_EXPORTS preprocessor symbol defined on the |
---|
68 | // command line. In contrast, projects that use (or import) the DLL |
---|
69 | // objects do not define the DLL_EXPORTS symbol. This way any other |
---|
70 | // project whose source files include this file see DLL_API functions as |
---|
71 | // being imported from a DLL, wheras this DLL sees symbols defined with |
---|
72 | // this macro as being exported. |
---|
73 | //---------------------------------------------------------------------- |
---|
74 | #ifdef DLL_EXPORTS |
---|
75 | #define DLL_API __declspec(dllexport) |
---|
76 | #else |
---|
77 | #define DLL_API __declspec(dllimport) |
---|
78 | #endif |
---|
79 | //---------------------------------------------------------------------- |
---|
80 | // DLL_API is ignored for all other systems |
---|
81 | //---------------------------------------------------------------------- |
---|
82 | #else |
---|
83 | #define DLL_API |
---|
84 | #endif |
---|
85 | |
---|
86 | //---------------------------------------------------------------------- |
---|
87 | // basic includes |
---|
88 | //---------------------------------------------------------------------- |
---|
89 | |
---|
90 | #include <cmath> // math includes |
---|
91 | #include <iostream> // I/O streams |
---|
92 | |
---|
93 | //---------------------------------------------------------------------- |
---|
94 | // Limits |
---|
95 | // There are a number of places where we use the maximum double value as |
---|
96 | // default initializers (and others may be used, depending on the |
---|
97 | // data/distance representation). These can usually be found in limits.h |
---|
98 | // (as LONG_MAX, INT_MAX) or in float.h (as DBL_MAX, FLT_MAX). |
---|
99 | // |
---|
100 | // Not all systems have these files. If you are using such a system, |
---|
101 | // you should set the preprocessor symbol ANN_NO_LIMITS_H when |
---|
102 | // compiling, and modify the statements below to generate the |
---|
103 | // appropriate value. For practical purposes, this does not need to be |
---|
104 | // the maximum double value. It is sufficient that it be at least as |
---|
105 | // large than the maximum squared distance between between any two |
---|
106 | // points. |
---|
107 | //---------------------------------------------------------------------- |
---|
108 | #ifdef ANN_NO_LIMITS_H // limits.h unavailable |
---|
109 | #include <cvalues> // replacement for limits.h |
---|
110 | const double ANN_DBL_MAX = MAXDOUBLE; // insert maximum double |
---|
111 | #else |
---|
112 | #include <climits> |
---|
113 | #include <cfloat> |
---|
114 | const double ANN_DBL_MAX = DBL_MAX; |
---|
115 | #endif |
---|
116 | |
---|
117 | #define ANNversion "1.1.1" // ANN version and information |
---|
118 | #define ANNversionCmt "" |
---|
119 | #define ANNcopyright "David M. Mount and Sunil Arya" |
---|
120 | #define ANNlatestRev "Aug 4, 2006" |
---|
121 | |
---|
122 | //---------------------------------------------------------------------- |
---|
123 | // ANNbool |
---|
124 | // This is a simple boolean type. Although ANSI C++ is supposed |
---|
125 | // to support the type bool, some compilers do not have it. |
---|
126 | //---------------------------------------------------------------------- |
---|
127 | |
---|
128 | enum ANNbool {ANNfalse = 0, ANNtrue = 1}; // ANN boolean type (non ANSI C++) |
---|
129 | |
---|
130 | //---------------------------------------------------------------------- |
---|
131 | // ANNcoord, ANNdist |
---|
132 | // ANNcoord and ANNdist are the types used for representing |
---|
133 | // point coordinates and distances. They can be modified by the |
---|
134 | // user, with some care. It is assumed that they are both numeric |
---|
135 | // types, and that ANNdist is generally of an equal or higher type |
---|
136 | // from ANNcoord. A variable of type ANNdist should be large |
---|
137 | // enough to store the sum of squared components of a variable |
---|
138 | // of type ANNcoord for the number of dimensions needed in the |
---|
139 | // application. For example, the following combinations are |
---|
140 | // legal: |
---|
141 | // |
---|
142 | // ANNcoord ANNdist |
---|
143 | // --------- ------------------------------- |
---|
144 | // short short, int, long, float, double |
---|
145 | // int int, long, float, double |
---|
146 | // long long, float, double |
---|
147 | // float float, double |
---|
148 | // double double |
---|
149 | // |
---|
150 | // It is the user's responsibility to make sure that overflow does |
---|
151 | // not occur in distance calculation. |
---|
152 | //---------------------------------------------------------------------- |
---|
153 | |
---|
154 | typedef double ANNcoord; // coordinate data type |
---|
155 | typedef double ANNdist; // distance data type |
---|
156 | |
---|
157 | //---------------------------------------------------------------------- |
---|
158 | // ANNidx |
---|
159 | // ANNidx is a point index. When the data structure is built, the |
---|
160 | // points are given as an array. Nearest neighbor results are |
---|
161 | // returned as an integer index into this array. To make it |
---|
162 | // clearer when this is happening, we define the integer type |
---|
163 | // ANNidx. Indexing starts from 0. |
---|
164 | // |
---|
165 | // For fixed-radius near neighbor searching, it is possible that |
---|
166 | // there are not k nearest neighbors within the search radius. To |
---|
167 | // indicate this, the algorithm returns ANN_NULL_IDX as its result. |
---|
168 | // It should be distinguishable from any valid array index. |
---|
169 | //---------------------------------------------------------------------- |
---|
170 | |
---|
171 | typedef int ANNidx; // point index |
---|
172 | const ANNidx ANN_NULL_IDX = -1; // a NULL point index |
---|
173 | |
---|
174 | //---------------------------------------------------------------------- |
---|
175 | // Infinite distance: |
---|
176 | // The code assumes that there is an "infinite distance" which it |
---|
177 | // uses to initialize distances before performing nearest neighbor |
---|
178 | // searches. It should be as larger or larger than any legitimate |
---|
179 | // nearest neighbor distance. |
---|
180 | // |
---|
181 | // On most systems, these should be found in the standard include |
---|
182 | // file <limits.h> or possibly <float.h>. If you do not have these |
---|
183 | // file, some suggested values are listed below, assuming 64-bit |
---|
184 | // long, 32-bit int and 16-bit short. |
---|
185 | // |
---|
186 | // ANNdist ANN_DIST_INF Values (see <limits.h> or <float.h>) |
---|
187 | // ------- ------------ ------------------------------------ |
---|
188 | // double DBL_MAX 1.79769313486231570e+308 |
---|
189 | // float FLT_MAX 3.40282346638528860e+38 |
---|
190 | // long LONG_MAX 0x7fffffffffffffff |
---|
191 | // int INT_MAX 0x7fffffff |
---|
192 | // short SHRT_MAX 0x7fff |
---|
193 | //---------------------------------------------------------------------- |
---|
194 | |
---|
195 | const ANNdist ANN_DIST_INF = ANN_DBL_MAX; |
---|
196 | |
---|
197 | //---------------------------------------------------------------------- |
---|
198 | // Significant digits for tree dumps: |
---|
199 | // When floating point coordinates are used, the routine that dumps |
---|
200 | // a tree needs to know roughly how many significant digits there |
---|
201 | // are in a ANNcoord, so it can output points to full precision. |
---|
202 | // This is defined to be ANNcoordPrec. On most systems these |
---|
203 | // values can be found in the standard include files <limits.h> or |
---|
204 | // <float.h>. For integer types, the value is essentially ignored. |
---|
205 | // |
---|
206 | // ANNcoord ANNcoordPrec Values (see <limits.h> or <float.h>) |
---|
207 | // -------- ------------ ------------------------------------ |
---|
208 | // double DBL_DIG 15 |
---|
209 | // float FLT_DIG 6 |
---|
210 | // long doesn't matter 19 |
---|
211 | // int doesn't matter 10 |
---|
212 | // short doesn't matter 5 |
---|
213 | //---------------------------------------------------------------------- |
---|
214 | |
---|
215 | #ifdef DBL_DIG // number of sig. bits in ANNcoord |
---|
216 | const int ANNcoordPrec = DBL_DIG; |
---|
217 | #else |
---|
218 | const int ANNcoordPrec = 15; // default precision |
---|
219 | #endif |
---|
220 | |
---|
221 | //---------------------------------------------------------------------- |
---|
222 | // Self match? |
---|
223 | // In some applications, the nearest neighbor of a point is not |
---|
224 | // allowed to be the point itself. This occurs, for example, when |
---|
225 | // computing all nearest neighbors in a set. By setting the |
---|
226 | // parameter ANN_ALLOW_SELF_MATCH to ANNfalse, the nearest neighbor |
---|
227 | // is the closest point whose distance from the query point is |
---|
228 | // strictly positive. |
---|
229 | //---------------------------------------------------------------------- |
---|
230 | |
---|
231 | const ANNbool ANN_ALLOW_SELF_MATCH = ANNtrue; |
---|
232 | |
---|
233 | //---------------------------------------------------------------------- |
---|
234 | // Norms and metrics: |
---|
235 | // ANN supports any Minkowski norm for defining distance. In |
---|
236 | // particular, for any p >= 1, the L_p Minkowski norm defines the |
---|
237 | // length of a d-vector (v0, v1, ..., v(d-1)) to be |
---|
238 | // |
---|
239 | // (|v0|^p + |v1|^p + ... + |v(d-1)|^p)^(1/p), |
---|
240 | // |
---|
241 | // (where ^ denotes exponentiation, and |.| denotes absolute |
---|
242 | // value). The distance between two points is defined to be the |
---|
243 | // norm of the vector joining them. Some common distance metrics |
---|
244 | // include |
---|
245 | // |
---|
246 | // Euclidean metric p = 2 |
---|
247 | // Manhattan metric p = 1 |
---|
248 | // Max metric p = infinity |
---|
249 | // |
---|
250 | // In the case of the max metric, the norm is computed by taking |
---|
251 | // the maxima of the absolute values of the components. ANN is |
---|
252 | // highly "coordinate-based" and does not support general distances |
---|
253 | // functions (e.g. those obeying just the triangle inequality). It |
---|
254 | // also does not support distance functions based on |
---|
255 | // inner-products. |
---|
256 | // |
---|
257 | // For the purpose of computing nearest neighbors, it is not |
---|
258 | // necessary to compute the final power (1/p). Thus the only |
---|
259 | // component that is used by the program is |v(i)|^p. |
---|
260 | // |
---|
261 | // ANN parameterizes the distance computation through the following |
---|
262 | // macros. (Macros are used rather than procedures for |
---|
263 | // efficiency.) Recall that the distance between two points is |
---|
264 | // given by the length of the vector joining them, and the length |
---|
265 | // or norm of a vector v is given by formula: |
---|
266 | // |
---|
267 | // |v| = ROOT(POW(v0) # POW(v1) # ... # POW(v(d-1))) |
---|
268 | // |
---|
269 | // where ROOT, POW are unary functions and # is an associative and |
---|
270 | // commutative binary operator mapping the following types: |
---|
271 | // |
---|
272 | // ** POW: ANNcoord --> ANNdist |
---|
273 | // ** #: ANNdist x ANNdist --> ANNdist |
---|
274 | // ** ROOT: ANNdist (>0) --> double |
---|
275 | // |
---|
276 | // For early termination in distance calculation (partial distance |
---|
277 | // calculation) we assume that POW and # together are monotonically |
---|
278 | // increasing on sequences of arguments, meaning that for all |
---|
279 | // v0..vk and y: |
---|
280 | // |
---|
281 | // POW(v0) #...# POW(vk) <= (POW(v0) #...# POW(vk)) # POW(y). |
---|
282 | // |
---|
283 | // Incremental Distance Calculation: |
---|
284 | // The program uses an optimized method of computing distances for |
---|
285 | // kd-trees and bd-trees, called incremental distance calculation. |
---|
286 | // It is used when distances are to be updated when only a single |
---|
287 | // coordinate of a point has been changed. In order to use this, |
---|
288 | // we assume that there is an incremental update function DIFF(x,y) |
---|
289 | // for #, such that if: |
---|
290 | // |
---|
291 | // s = x0 # ... # xi # ... # xk |
---|
292 | // |
---|
293 | // then if s' is equal to s but with xi replaced by y, that is, |
---|
294 | // |
---|
295 | // s' = x0 # ... # y # ... # xk |
---|
296 | // |
---|
297 | // then the length of s' can be computed by: |
---|
298 | // |
---|
299 | // |s'| = |s| # DIFF(xi,y). |
---|
300 | // |
---|
301 | // Thus, if # is + then DIFF(xi,y) is (yi-x). For the L_infinity |
---|
302 | // norm we make use of the fact that in the program this function |
---|
303 | // is only invoked when y > xi, and hence DIFF(xi,y)=y. |
---|
304 | // |
---|
305 | // Finally, for approximate nearest neighbor queries we assume |
---|
306 | // that POW and ROOT are related such that |
---|
307 | // |
---|
308 | // v*ROOT(x) = ROOT(POW(v)*x) |
---|
309 | // |
---|
310 | // Here are the values for the various Minkowski norms: |
---|
311 | // |
---|
312 | // L_p: p even: p odd: |
---|
313 | // ------------------------- ------------------------ |
---|
314 | // POW(v) = v^p POW(v) = |v|^p |
---|
315 | // ROOT(x) = x^(1/p) ROOT(x) = x^(1/p) |
---|
316 | // # = + # = + |
---|
317 | // DIFF(x,y) = y - x DIFF(x,y) = y - x |
---|
318 | // |
---|
319 | // L_inf: |
---|
320 | // POW(v) = |v| |
---|
321 | // ROOT(x) = x |
---|
322 | // # = max |
---|
323 | // DIFF(x,y) = y |
---|
324 | // |
---|
325 | // By default the Euclidean norm is assumed. To change the norm, |
---|
326 | // uncomment the appropriate set of macros below. |
---|
327 | //---------------------------------------------------------------------- |
---|
328 | |
---|
329 | //---------------------------------------------------------------------- |
---|
330 | // Use the following for the Euclidean norm |
---|
331 | //---------------------------------------------------------------------- |
---|
332 | #define ANN_POW(v) ((v)*(v)) |
---|
333 | #define ANN_ROOT(x) sqrt(x) |
---|
334 | #define ANN_SUM(x,y) ((x) + (y)) |
---|
335 | #define ANN_DIFF(x,y) ((y) - (x)) |
---|
336 | |
---|
337 | //---------------------------------------------------------------------- |
---|
338 | // Use the following for the L_1 (Manhattan) norm |
---|
339 | //---------------------------------------------------------------------- |
---|
340 | // #define ANN_POW(v) fabs(v) |
---|
341 | // #define ANN_ROOT(x) (x) |
---|
342 | // #define ANN_SUM(x,y) ((x) + (y)) |
---|
343 | // #define ANN_DIFF(x,y) ((y) - (x)) |
---|
344 | |
---|
345 | //---------------------------------------------------------------------- |
---|
346 | // Use the following for a general L_p norm |
---|
347 | //---------------------------------------------------------------------- |
---|
348 | // #define ANN_POW(v) pow(fabs(v),p) |
---|
349 | // #define ANN_ROOT(x) pow(fabs(x),1/p) |
---|
350 | // #define ANN_SUM(x,y) ((x) + (y)) |
---|
351 | // #define ANN_DIFF(x,y) ((y) - (x)) |
---|
352 | |
---|
353 | //---------------------------------------------------------------------- |
---|
354 | // Use the following for the L_infinity (Max) norm |
---|
355 | //---------------------------------------------------------------------- |
---|
356 | // #define ANN_POW(v) fabs(v) |
---|
357 | // #define ANN_ROOT(x) (x) |
---|
358 | // #define ANN_SUM(x,y) ((x) > (y) ? (x) : (y)) |
---|
359 | // #define ANN_DIFF(x,y) (y) |
---|
360 | |
---|
361 | //---------------------------------------------------------------------- |
---|
362 | // Array types |
---|
363 | // The following array types are of basic interest. A point is |
---|
364 | // just a dimensionless array of coordinates, a point array is a |
---|
365 | // dimensionless array of points. A distance array is a |
---|
366 | // dimensionless array of distances and an index array is a |
---|
367 | // dimensionless array of point indices. The latter two are used |
---|
368 | // when returning the results of k-nearest neighbor queries. |
---|
369 | //---------------------------------------------------------------------- |
---|
370 | |
---|
371 | typedef ANNcoord* ANNpoint; // a point |
---|
372 | typedef ANNpoint* ANNpointArray; // an array of points |
---|
373 | typedef ANNdist* ANNdistArray; // an array of distances |
---|
374 | typedef ANNidx* ANNidxArray; // an array of point indices |
---|
375 | |
---|
376 | //---------------------------------------------------------------------- |
---|
377 | // Basic point and array utilities: |
---|
378 | // The following procedures are useful supplements to ANN's nearest |
---|
379 | // neighbor capabilities. |
---|
380 | // |
---|
381 | // annDist(): |
---|
382 | // Computes the (squared) distance between a pair of points. |
---|
383 | // Note that this routine is not used internally by ANN for |
---|
384 | // computing distance calculations. For reasons of efficiency |
---|
385 | // this is done using incremental distance calculation. Thus, |
---|
386 | // this routine cannot be modified as a method of changing the |
---|
387 | // metric. |
---|
388 | // |
---|
389 | // Because points (somewhat like strings in C) are stored as |
---|
390 | // pointers. Consequently, creating and destroying copies of |
---|
391 | // points may require storage allocation. These procedures do |
---|
392 | // this. |
---|
393 | // |
---|
394 | // annAllocPt() and annDeallocPt(): |
---|
395 | // Allocate a deallocate storage for a single point, and |
---|
396 | // return a pointer to it. The argument to AllocPt() is |
---|
397 | // used to initialize all components. |
---|
398 | // |
---|
399 | // annAllocPts() and annDeallocPts(): |
---|
400 | // Allocate and deallocate an array of points as well a |
---|
401 | // place to store their coordinates, and initializes the |
---|
402 | // points to point to their respective coordinates. It |
---|
403 | // allocates point storage in a contiguous block large |
---|
404 | // enough to store all the points. It performs no |
---|
405 | // initialization. |
---|
406 | // |
---|
407 | // annCopyPt(): |
---|
408 | // Creates a copy of a given point, allocating space for |
---|
409 | // the new point. It returns a pointer to the newly |
---|
410 | // allocated copy. |
---|
411 | //---------------------------------------------------------------------- |
---|
412 | |
---|
413 | DLL_API ANNdist annDist( |
---|
414 | int dim, // dimension of space |
---|
415 | ANNpoint p, // points |
---|
416 | ANNpoint q); |
---|
417 | |
---|
418 | DLL_API ANNpoint annAllocPt( |
---|
419 | int dim, // dimension |
---|
420 | ANNcoord c = 0); // coordinate value (all equal) |
---|
421 | |
---|
422 | DLL_API ANNpointArray annAllocPts( |
---|
423 | int n, // number of points |
---|
424 | int dim); // dimension |
---|
425 | |
---|
426 | DLL_API void annDeallocPt( |
---|
427 | ANNpoint &p); // deallocate 1 point |
---|
428 | |
---|
429 | DLL_API void annDeallocPts( |
---|
430 | ANNpointArray &pa); // point array |
---|
431 | |
---|
432 | DLL_API ANNpoint annCopyPt( |
---|
433 | int dim, // dimension |
---|
434 | ANNpoint source); // point to copy |
---|
435 | |
---|
436 | //---------------------------------------------------------------------- |
---|
437 | //Overall structure: ANN supports a number of different data structures |
---|
438 | //for approximate and exact nearest neighbor searching. These are: |
---|
439 | // |
---|
440 | // ANNbruteForce A simple brute-force search structure. |
---|
441 | // ANNkd_tree A kd-tree tree search structure. ANNbd_tree |
---|
442 | // A bd-tree tree search structure (a kd-tree with shrink |
---|
443 | // capabilities). |
---|
444 | // |
---|
445 | // At a minimum, each of these data structures support k-nearest |
---|
446 | // neighbor queries. The nearest neighbor query, annkSearch, |
---|
447 | // returns an integer identifier and the distance to the nearest |
---|
448 | // neighbor(s) and annRangeSearch returns the nearest points that |
---|
449 | // lie within a given query ball. |
---|
450 | // |
---|
451 | // Each structure is built by invoking the appropriate constructor |
---|
452 | // and passing it (at a minimum) the array of points, the total |
---|
453 | // number of points and the dimension of the space. Each structure |
---|
454 | // is also assumed to support a destructor and member functions |
---|
455 | // that return basic information about the point set. |
---|
456 | // |
---|
457 | // Note that the array of points is not copied by the data |
---|
458 | // structure (for reasons of space efficiency), and it is assumed |
---|
459 | // to be constant throughout the lifetime of the search structure. |
---|
460 | // |
---|
461 | // The search algorithm, annkSearch, is given the query point (q), |
---|
462 | // and the desired number of nearest neighbors to report (k), and |
---|
463 | // the error bound (eps) (whose default value is 0, implying exact |
---|
464 | // nearest neighbors). It returns two arrays which are assumed to |
---|
465 | // contain at least k elements: one (nn_idx) contains the indices |
---|
466 | // (within the point array) of the nearest neighbors and the other |
---|
467 | // (dd) contains the squared distances to these nearest neighbors. |
---|
468 | // |
---|
469 | // The search algorithm, annkFRSearch, is a fixed-radius kNN |
---|
470 | // search. In addition to a query point, it is given a (squared) |
---|
471 | // radius bound. (This is done for consistency, because the search |
---|
472 | // returns distances as squared quantities.) It does two things. |
---|
473 | // First, it computes the k nearest neighbors within the radius |
---|
474 | // bound, and second, it returns the total number of points lying |
---|
475 | // within the radius bound. It is permitted to set k = 0, in which |
---|
476 | // case it effectively answers a range counting query. If the |
---|
477 | // error bound epsilon is positive, then the search is approximate |
---|
478 | // in the sense that it is free to ignore any point that lies |
---|
479 | // outside a ball of radius r/(1+epsilon), where r is the given |
---|
480 | // (unsquared) radius bound. |
---|
481 | // |
---|
482 | // The generic object from which all the search structures are |
---|
483 | // dervied is given below. It is a virtual object, and is useless |
---|
484 | // by itself. |
---|
485 | //---------------------------------------------------------------------- |
---|
486 | |
---|
487 | class DLL_API ANNpointSet { |
---|
488 | public: |
---|
489 | virtual ~ANNpointSet() {} // virtual distructor |
---|
490 | |
---|
491 | virtual void annkSearch( // approx k near neighbor search |
---|
492 | ANNpoint q, // query point |
---|
493 | int k, // number of near neighbors to return |
---|
494 | ANNidxArray nn_idx, // nearest neighbor array (modified) |
---|
495 | ANNdistArray dd, // dist to near neighbors (modified) |
---|
496 | double eps=0.0 // error bound |
---|
497 | ) = 0; // pure virtual (defined elsewhere) |
---|
498 | |
---|
499 | virtual int annkFRSearch( // approx fixed-radius kNN search |
---|
500 | ANNpoint q, // query point |
---|
501 | ANNdist sqRad, // squared radius |
---|
502 | int k = 0, // number of near neighbors to return |
---|
503 | ANNidxArray nn_idx = NULL, // nearest neighbor array (modified) |
---|
504 | ANNdistArray dd = NULL, // dist to near neighbors (modified) |
---|
505 | double eps=0.0 // error bound |
---|
506 | ) = 0; // pure virtual (defined elsewhere) |
---|
507 | |
---|
508 | virtual int theDim() = 0; // return dimension of space |
---|
509 | virtual int nPoints() = 0; // return number of points |
---|
510 | // return pointer to points |
---|
511 | virtual ANNpointArray thePoints() = 0; |
---|
512 | }; |
---|
513 | |
---|
514 | //---------------------------------------------------------------------- |
---|
515 | // Brute-force nearest neighbor search: |
---|
516 | // The brute-force search structure is very simple but inefficient. |
---|
517 | // It has been provided primarily for the sake of comparison with |
---|
518 | // and validation of the more complex search structures. |
---|
519 | // |
---|
520 | // Query processing is the same as described above, but the value |
---|
521 | // of epsilon is ignored, since all distance calculations are |
---|
522 | // performed exactly. |
---|
523 | // |
---|
524 | // WARNING: This data structure is very slow, and should not be |
---|
525 | // used unless the number of points is very small. |
---|
526 | // |
---|
527 | // Internal information: |
---|
528 | // --------------------- |
---|
529 | // This data structure bascially consists of the array of points |
---|
530 | // (each a pointer to an array of coordinates). The search is |
---|
531 | // performed by a simple linear scan of all the points. |
---|
532 | //---------------------------------------------------------------------- |
---|
533 | |
---|
534 | class DLL_API ANNbruteForce: public ANNpointSet { |
---|
535 | int dim; // dimension |
---|
536 | int n_pts; // number of points |
---|
537 | ANNpointArray pts; // point array |
---|
538 | public: |
---|
539 | ANNbruteForce( // constructor from point array |
---|
540 | ANNpointArray pa, // point array |
---|
541 | int n, // number of points |
---|
542 | int dd); // dimension |
---|
543 | |
---|
544 | ~ANNbruteForce(); // destructor |
---|
545 | |
---|
546 | void annkSearch( // approx k near neighbor search |
---|
547 | ANNpoint q, // query point |
---|
548 | int k, // number of near neighbors to return |
---|
549 | ANNidxArray nn_idx, // nearest neighbor array (modified) |
---|
550 | ANNdistArray dd, // dist to near neighbors (modified) |
---|
551 | double eps=0.0); // error bound |
---|
552 | |
---|
553 | int annkFRSearch( // approx fixed-radius kNN search |
---|
554 | ANNpoint q, // query point |
---|
555 | ANNdist sqRad, // squared radius |
---|
556 | int k = 0, // number of near neighbors to return |
---|
557 | ANNidxArray nn_idx = NULL, // nearest neighbor array (modified) |
---|
558 | ANNdistArray dd = NULL, // dist to near neighbors (modified) |
---|
559 | double eps=0.0); // error bound |
---|
560 | |
---|
561 | int theDim() // return dimension of space |
---|
562 | { return dim; } |
---|
563 | |
---|
564 | int nPoints() // return number of points |
---|
565 | { return n_pts; } |
---|
566 | |
---|
567 | ANNpointArray thePoints() // return pointer to points |
---|
568 | { return pts; } |
---|
569 | }; |
---|
570 | |
---|
571 | //---------------------------------------------------------------------- |
---|
572 | // kd- and bd-tree splitting and shrinking rules |
---|
573 | // kd-trees supports a collection of different splitting rules. |
---|
574 | // In addition to the standard kd-tree splitting rule proposed |
---|
575 | // by Friedman, Bentley, and Finkel, we have introduced a |
---|
576 | // number of other splitting rules, which seem to perform |
---|
577 | // as well or better (for the distributions we have tested). |
---|
578 | // |
---|
579 | // The splitting methods given below allow the user to tailor |
---|
580 | // the data structure to the particular data set. They are |
---|
581 | // are described in greater details in the kd_split.cc source |
---|
582 | // file. The method ANN_KD_SUGGEST is the method chosen (rather |
---|
583 | // subjectively) by the implementors as the one giving the |
---|
584 | // fastest performance, and is the default splitting method. |
---|
585 | // |
---|
586 | // As with splitting rules, there are a number of different |
---|
587 | // shrinking rules. The shrinking rule ANN_BD_NONE does no |
---|
588 | // shrinking (and hence produces a kd-tree tree). The rule |
---|
589 | // ANN_BD_SUGGEST uses the implementors favorite rule. |
---|
590 | //---------------------------------------------------------------------- |
---|
591 | |
---|
592 | enum ANNsplitRule { |
---|
593 | ANN_KD_STD = 0, // the optimized kd-splitting rule |
---|
594 | ANN_KD_MIDPT = 1, // midpoint split |
---|
595 | ANN_KD_FAIR = 2, // fair split |
---|
596 | ANN_KD_SL_MIDPT = 3, // sliding midpoint splitting method |
---|
597 | ANN_KD_SL_FAIR = 4, // sliding fair split method |
---|
598 | ANN_KD_SUGGEST = 5}; // the authors' suggestion for best |
---|
599 | const int ANN_N_SPLIT_RULES = 6; // number of split rules |
---|
600 | |
---|
601 | enum ANNshrinkRule { |
---|
602 | ANN_BD_NONE = 0, // no shrinking at all (just kd-tree) |
---|
603 | ANN_BD_SIMPLE = 1, // simple splitting |
---|
604 | ANN_BD_CENTROID = 2, // centroid splitting |
---|
605 | ANN_BD_SUGGEST = 3}; // the authors' suggested choice |
---|
606 | const int ANN_N_SHRINK_RULES = 4; // number of shrink rules |
---|
607 | |
---|
608 | //---------------------------------------------------------------------- |
---|
609 | // kd-tree: |
---|
610 | // The main search data structure supported by ANN is a kd-tree. |
---|
611 | // The main constructor is given a set of points and a choice of |
---|
612 | // splitting method to use in building the tree. |
---|
613 | // |
---|
614 | // Construction: |
---|
615 | // ------------- |
---|
616 | // The constructor is given the point array, number of points, |
---|
617 | // dimension, bucket size (default = 1), and the splitting rule |
---|
618 | // (default = ANN_KD_SUGGEST). The point array is not copied, and |
---|
619 | // is assumed to be kept constant throughout the lifetime of the |
---|
620 | // search structure. There is also a "load" constructor that |
---|
621 | // builds a tree from a file description that was created by the |
---|
622 | // Dump operation. |
---|
623 | // |
---|
624 | // Search: |
---|
625 | // ------- |
---|
626 | // There are two search methods: |
---|
627 | // |
---|
628 | // Standard search (annkSearch()): |
---|
629 | // Searches nodes in tree-traversal order, always visiting |
---|
630 | // the closer child first. |
---|
631 | // Priority search (annkPriSearch()): |
---|
632 | // Searches nodes in order of increasing distance of the |
---|
633 | // associated cell from the query point. For many |
---|
634 | // distributions the standard search seems to work just |
---|
635 | // fine, but priority search is safer for worst-case |
---|
636 | // performance. |
---|
637 | // |
---|
638 | // Printing: |
---|
639 | // --------- |
---|
640 | // There are two methods provided for printing the tree. Print() |
---|
641 | // is used to produce a "human-readable" display of the tree, with |
---|
642 | // indenation, which is handy for debugging. Dump() produces a |
---|
643 | // format that is suitable reading by another program. There is a |
---|
644 | // "load" constructor, which constructs a tree which is assumed to |
---|
645 | // have been saved by the Dump() procedure. |
---|
646 | // |
---|
647 | // Performance and Structure Statistics: |
---|
648 | // ------------------------------------- |
---|
649 | // The procedure getStats() collects statistics information on the |
---|
650 | // tree (its size, height, etc.) See ANNperf.h for information on |
---|
651 | // the stats structure it returns. |
---|
652 | // |
---|
653 | // Internal information: |
---|
654 | // --------------------- |
---|
655 | // The data structure consists of three major chunks of storage. |
---|
656 | // The first (implicit) storage are the points themselves (pts), |
---|
657 | // which have been provided by the users as an argument to the |
---|
658 | // constructor, or are allocated dynamically if the tree is built |
---|
659 | // using the load constructor). These should not be changed during |
---|
660 | // the lifetime of the search structure. It is the user's |
---|
661 | // responsibility to delete these after the tree is destroyed. |
---|
662 | // |
---|
663 | // The second is the tree itself (which is dynamically allocated in |
---|
664 | // the constructor) and is given as a pointer to its root node |
---|
665 | // (root). These nodes are automatically deallocated when the tree |
---|
666 | // is deleted. See the file src/kd_tree.h for further information |
---|
667 | // on the structure of the tree nodes. |
---|
668 | // |
---|
669 | // Each leaf of the tree does not contain a pointer directly to a |
---|
670 | // point, but rather contains a pointer to a "bucket", which is an |
---|
671 | // array consisting of point indices. The third major chunk of |
---|
672 | // storage is an array (pidx), which is a large array in which all |
---|
673 | // these bucket subarrays reside. (The reason for storing them |
---|
674 | // separately is the buckets are typically small, but of varying |
---|
675 | // sizes. This was done to avoid fragmentation.) This array is |
---|
676 | // also deallocated when the tree is deleted. |
---|
677 | // |
---|
678 | // In addition to this, the tree consists of a number of other |
---|
679 | // pieces of information which are used in searching and for |
---|
680 | // subsequent tree operations. These consist of the following: |
---|
681 | // |
---|
682 | // dim Dimension of space |
---|
683 | // n_pts Number of points currently in the tree |
---|
684 | // n_max Maximum number of points that are allowed |
---|
685 | // in the tree |
---|
686 | // bkt_size Maximum bucket size (no. of points per leaf) |
---|
687 | // bnd_box_lo Bounding box low point |
---|
688 | // bnd_box_hi Bounding box high point |
---|
689 | // splitRule Splitting method used |
---|
690 | // |
---|
691 | //---------------------------------------------------------------------- |
---|
692 | |
---|
693 | //---------------------------------------------------------------------- |
---|
694 | // Some types and objects used by kd-tree functions |
---|
695 | // See src/kd_tree.h and src/kd_tree.cpp for definitions |
---|
696 | //---------------------------------------------------------------------- |
---|
697 | class ANNkdStats; // stats on kd-tree |
---|
698 | class ANNkd_node; // generic node in a kd-tree |
---|
699 | typedef ANNkd_node* ANNkd_ptr; // pointer to a kd-tree node |
---|
700 | |
---|
701 | class DLL_API ANNkd_tree: public ANNpointSet { |
---|
702 | protected: |
---|
703 | int dim; // dimension of space |
---|
704 | int n_pts; // number of points in tree |
---|
705 | int bkt_size; // bucket size |
---|
706 | ANNpointArray pts; // the points |
---|
707 | ANNidxArray pidx; // point indices (to pts array) |
---|
708 | ANNkd_ptr root; // root of kd-tree |
---|
709 | ANNpoint bnd_box_lo; // bounding box low point |
---|
710 | ANNpoint bnd_box_hi; // bounding box high point |
---|
711 | |
---|
712 | void SkeletonTree( // construct skeleton tree |
---|
713 | int n, // number of points |
---|
714 | int dd, // dimension |
---|
715 | int bs, // bucket size |
---|
716 | ANNpointArray pa = NULL, // point array (optional) |
---|
717 | ANNidxArray pi = NULL); // point indices (optional) |
---|
718 | |
---|
719 | public: |
---|
720 | ANNkd_tree( // build skeleton tree |
---|
721 | int n = 0, // number of points |
---|
722 | int dd = 0, // dimension |
---|
723 | int bs = 1); // bucket size |
---|
724 | |
---|
725 | ANNkd_tree( // build from point array |
---|
726 | ANNpointArray pa, // point array |
---|
727 | int n, // number of points |
---|
728 | int dd, // dimension |
---|
729 | int bs = 1, // bucket size |
---|
730 | ANNsplitRule split = ANN_KD_SUGGEST); // splitting method |
---|
731 | |
---|
732 | ANNkd_tree( // build from dump file |
---|
733 | std::istream& in); // input stream for dump file |
---|
734 | |
---|
735 | ~ANNkd_tree(); // tree destructor |
---|
736 | |
---|
737 | void annkSearch( // approx k near neighbor search |
---|
738 | ANNpoint q, // query point |
---|
739 | int k, // number of near neighbors to return |
---|
740 | ANNidxArray nn_idx, // nearest neighbor array (modified) |
---|
741 | ANNdistArray dd, // dist to near neighbors (modified) |
---|
742 | double eps=0.0); // error bound |
---|
743 | |
---|
744 | void annkPriSearch( // priority k near neighbor search |
---|
745 | ANNpoint q, // query point |
---|
746 | int k, // number of near neighbors to return |
---|
747 | ANNidxArray nn_idx, // nearest neighbor array (modified) |
---|
748 | ANNdistArray dd, // dist to near neighbors (modified) |
---|
749 | double eps=0.0); // error bound |
---|
750 | |
---|
751 | int annkFRSearch( // approx fixed-radius kNN search |
---|
752 | ANNpoint q, // the query point |
---|
753 | ANNdist sqRad, // squared radius of query ball |
---|
754 | int k, // number of neighbors to return |
---|
755 | ANNidxArray nn_idx = NULL, // nearest neighbor array (modified) |
---|
756 | ANNdistArray dd = NULL, // dist to near neighbors (modified) |
---|
757 | double eps=0.0); // error bound |
---|
758 | |
---|
759 | int theDim() // return dimension of space |
---|
760 | { return dim; } |
---|
761 | |
---|
762 | int nPoints() // return number of points |
---|
763 | { return n_pts; } |
---|
764 | |
---|
765 | ANNpointArray thePoints() // return pointer to points |
---|
766 | { return pts; } |
---|
767 | |
---|
768 | virtual void Print( // print the tree (for debugging) |
---|
769 | ANNbool with_pts, // print points as well? |
---|
770 | std::ostream& out); // output stream |
---|
771 | |
---|
772 | virtual void Dump( // dump entire tree |
---|
773 | ANNbool with_pts, // print points as well? |
---|
774 | std::ostream& out); // output stream |
---|
775 | |
---|
776 | virtual void getStats( // compute tree statistics |
---|
777 | ANNkdStats& st); // the statistics (modified) |
---|
778 | }; |
---|
779 | |
---|
780 | //---------------------------------------------------------------------- |
---|
781 | // Box decomposition tree (bd-tree) |
---|
782 | // The bd-tree is inherited from a kd-tree. The main difference |
---|
783 | // in the bd-tree and the kd-tree is a new type of internal node |
---|
784 | // called a shrinking node (in the kd-tree there is only one type |
---|
785 | // of internal node, a splitting node). The shrinking node |
---|
786 | // makes it possible to generate balanced trees in which the |
---|
787 | // cells have bounded aspect ratio, by allowing the decomposition |
---|
788 | // to zoom in on regions of dense point concentration. Although |
---|
789 | // this is a nice idea in theory, few point distributions are so |
---|
790 | // densely clustered that this is really needed. |
---|
791 | //---------------------------------------------------------------------- |
---|
792 | |
---|
793 | class DLL_API ANNbd_tree: public ANNkd_tree { |
---|
794 | public: |
---|
795 | ANNbd_tree( // build skeleton tree |
---|
796 | int n, // number of points |
---|
797 | int dd, // dimension |
---|
798 | int bs = 1) // bucket size |
---|
799 | : ANNkd_tree(n, dd, bs) {} // build base kd-tree |
---|
800 | |
---|
801 | ANNbd_tree( // build from point array |
---|
802 | ANNpointArray pa, // point array |
---|
803 | int n, // number of points |
---|
804 | int dd, // dimension |
---|
805 | int bs = 1, // bucket size |
---|
806 | ANNsplitRule split = ANN_KD_SUGGEST, // splitting rule |
---|
807 | ANNshrinkRule shrink = ANN_BD_SUGGEST); // shrinking rule |
---|
808 | |
---|
809 | ANNbd_tree( // build from dump file |
---|
810 | std::istream& in); // input stream for dump file |
---|
811 | }; |
---|
812 | |
---|
813 | //---------------------------------------------------------------------- |
---|
814 | // Other functions |
---|
815 | // annMaxPtsVisit Sets a limit on the maximum number of points |
---|
816 | // to visit in the search. |
---|
817 | // annClose Can be called when all use of ANN is finished. |
---|
818 | // It clears up a minor memory leak. |
---|
819 | //---------------------------------------------------------------------- |
---|
820 | |
---|
821 | DLL_API void annMaxPtsVisit( // max. pts to visit in search |
---|
822 | int maxPts); // the limit |
---|
823 | |
---|
824 | DLL_API void annClose(); // called to end use of ANN |
---|
825 | |
---|
826 | #endif |
---|