queso-0.53.0
kd_pr_search.cpp
Go to the documentation of this file.
1 //----------------------------------------------------------------------
2 // File: kd_pr_search.cpp
3 // Programmer: Sunil Arya and David Mount
4 // Description: Priority search for kd-trees
5 // Last modified: 01/04/05 (Version 1.0)
6 //----------------------------------------------------------------------
7 // Copyright (c) 1997-2005 University of Maryland and Sunil Arya and
8 // David Mount. All Rights Reserved.
9 //
10 // This software and related documentation is part of the Approximate
11 // Nearest Neighbor Library (ANN). This software is provided under
12 // the provisions of the Lesser GNU Public License (LGPL). See the
13 // file ../ReadMe.txt for further information.
14 //
15 // The University of Maryland (U.M.) and the authors make no
16 // representations about the suitability or fitness of this software for
17 // any purpose. It is provided "as is" without express or implied
18 // warranty.
19 //----------------------------------------------------------------------
20 // History:
21 // Revision 0.1 03/04/98
22 // Initial release
23 //----------------------------------------------------------------------
24 
25 #include "kd_pr_search.h" // kd priority search declarations
26 
27 //----------------------------------------------------------------------
28 // Approximate nearest neighbor searching by priority search.
29 // The kd-tree is searched for an approximate nearest neighbor.
30 // The point is returned through one of the arguments, and the
31 // distance returned is the SQUARED distance to this point.
32 //
33 // The method used for searching the kd-tree is called priority
34 // search. (It is described in Arya and Mount, ``Algorithms for
35 // fast vector quantization,'' Proc. of DCC '93: Data Compression
36 // Conference}, eds. J. A. Storer and M. Cohn, IEEE Press, 1993,
37 // 381--390.)
38 //
39 // The cell of the kd-tree containing the query point is located,
40 // and cells are visited in increasing order of distance from the
41 // query point. This is done by placing each subtree which has
42 // NOT been visited in a priority queue, according to the closest
43 // distance of the corresponding enclosing rectangle from the
44 // query point. The search stops when the distance to the nearest
45 // remaining rectangle exceeds the distance to the nearest point
46 // seen by a factor of more than 1/(1+eps). (Implying that any
47 // point found subsequently in the search cannot be closer by more
48 // than this factor.)
49 //
50 // The main entry point is annkPriSearch() which sets things up and
51 // then call the recursive routine ann_pri_search(). This is a
52 // recursive routine which performs the processing for one node in
53 // the kd-tree. There are two versions of this virtual procedure,
54 // one for splitting nodes and one for leaves. When a splitting node
55 // is visited, we determine which child to continue the search on
56 // (the closer one), and insert the other child into the priority
57 // queue. When a leaf is visited, we compute the distances to the
58 // points in the buckets, and update information on the closest
59 // points.
60 //
61 // Some trickery is used to incrementally update the distance from
62 // a kd-tree rectangle to the query point. This comes about from
63 // the fact that which each successive split, only one component
64 // (along the dimension that is split) of the squared distance to
65 // the child rectangle is different from the squared distance to
66 // the parent rectangle.
67 //----------------------------------------------------------------------
68 
69 //----------------------------------------------------------------------
70 // To keep argument lists short, a number of global variables
71 // are maintained which are common to all the recursive calls.
72 // These are given below.
73 //----------------------------------------------------------------------
74 
75 double ANNprEps; // the error bound
76 int ANNprDim; // dimension of space
77 ANNpoint ANNprQ; // query point
78 double ANNprMaxErr; // max tolerable squared error
79 ANNpointArray ANNprPts; // the points
80 ANNpr_queue *ANNprBoxPQ; // priority queue for boxes
81 ANNmin_k *ANNprPointMK; // set of k closest points
82 
83 //----------------------------------------------------------------------
84 // annkPriSearch - priority search for k nearest neighbors
85 //----------------------------------------------------------------------
86 
88  ANNpoint q, // query point
89  int k, // number of near neighbors to return
90  ANNidxArray nn_idx, // nearest neighbor indices (returned)
91  ANNdistArray dd, // dist to near neighbors (returned)
92  double eps) // error bound (ignored)
93 {
94  // max tolerable squared error
95  ANNprMaxErr = ANN_POW(1.0 + eps);
96  ANN_FLOP(2) // increment floating ops
97 
98  ANNprDim = dim; // copy arguments to static equivs
99  ANNprQ = q;
100  ANNprPts = pts;
101  ANNptsVisited = 0; // initialize count of points visited
102 
103  ANNprPointMK = new ANNmin_k(k); // create set for closest k points
104 
105  // distance to root box
106  ANNdist box_dist = annBoxDistance(q,
108 
109  ANNprBoxPQ = new ANNpr_queue(n_pts);// create priority queue for boxes
110  ANNprBoxPQ->insert(box_dist, root); // insert root in priority queue
111 
112  while (ANNprBoxPQ->non_empty() &&
114  ANNkd_ptr np; // next box from prior queue
115 
116  // extract closest box from queue
117  ANNprBoxPQ->extr_min(box_dist, (void *&) np);
118 
119  ANN_FLOP(2) // increment floating ops
120  if (box_dist*ANNprMaxErr >= ANNprPointMK->max_key())
121  break;
122 
123  np->ann_pri_search(box_dist); // search this subtree.
124  }
125 
126  for (int i = 0; i < k; i++) { // extract the k-th closest points
127  dd[i] = ANNprPointMK->ith_smallest_key(i);
128  nn_idx[i] = ANNprPointMK->ith_smallest_info(i);
129  }
130 
131  delete ANNprPointMK; // deallocate closest point set
132  delete ANNprBoxPQ; // deallocate priority queue
133 }
134 
135 //----------------------------------------------------------------------
136 // kd_split::ann_pri_search - search a splitting node
137 //----------------------------------------------------------------------
138 
140 {
141  ANNdist new_dist; // distance to child visited later
142  // distance to cutting plane
143  ANNcoord cut_diff = ANNprQ[cut_dim] - cut_val;
144 
145  if (cut_diff < 0) { // left of cutting plane
146  ANNcoord box_diff = cd_bnds[ANN_LO] - ANNprQ[cut_dim];
147  if (box_diff < 0) // within bounds - ignore
148  box_diff = 0;
149  // distance to further box
150  new_dist = (ANNdist) ANN_SUM(box_dist,
151  ANN_DIFF(ANN_POW(box_diff), ANN_POW(cut_diff)));
152 
153  if (child[ANN_HI] != KD_TRIVIAL)// enqueue if not trivial
154  ANNprBoxPQ->insert(new_dist, child[ANN_HI]);
155  // continue with closer child
156  child[ANN_LO]->ann_pri_search(box_dist);
157  }
158  else { // right of cutting plane
159  ANNcoord box_diff = ANNprQ[cut_dim] - cd_bnds[ANN_HI];
160  if (box_diff < 0) // within bounds - ignore
161  box_diff = 0;
162  // distance to further box
163  new_dist = (ANNdist) ANN_SUM(box_dist,
164  ANN_DIFF(ANN_POW(box_diff), ANN_POW(cut_diff)));
165 
166  if (child[ANN_LO] != KD_TRIVIAL)// enqueue if not trivial
167  ANNprBoxPQ->insert(new_dist, child[ANN_LO]);
168  // continue with closer child
169  child[ANN_HI]->ann_pri_search(box_dist);
170  }
171  ANN_SPL(1) // one more splitting node visited
172  ANN_FLOP(8) // increment floating ops
173 }
174 
175 //----------------------------------------------------------------------
176 // kd_leaf::ann_pri_search - search points in a leaf node
177 //
178 // This is virtually identical to the ann_search for standard search.
179 //----------------------------------------------------------------------
180 
182 {
183  register ANNdist dist; // distance to data point
184  register ANNcoord* pp; // data coordinate pointer
185  register ANNcoord* qq; // query coordinate pointer
186  register ANNdist min_dist; // distance to k-th closest point
187  register ANNcoord t;
188  register int d;
189 
190  min_dist = ANNprPointMK->max_key(); // k-th smallest distance so far
191 
192  for (int i = 0; i < n_pts; i++) { // check points in bucket
193 
194  pp = ANNprPts[bkt[i]]; // first coord of next data point
195  qq = ANNprQ; // first coord of query point
196  dist = 0;
197 
198  for(d = 0; d < ANNprDim; d++) {
199  ANN_COORD(1) // one more coordinate hit
200  ANN_FLOP(4) // increment floating ops
201 
202  t = *(qq++) - *(pp++); // compute length and adv coordinate
203  // exceeds dist to k-th smallest?
204  if( (dist = ANN_SUM(dist, ANN_POW(t))) > min_dist) {
205  break;
206  }
207  }
208 
209  if (d >= ANNprDim && // among the k best?
210  (ANN_ALLOW_SELF_MATCH || dist!=0)) { // and no self-match problem
211  // add it to the list
212  ANNprPointMK->insert(dist, bkt[i]);
213  min_dist = ANNprPointMK->max_key();
214  }
215  }
216  ANN_LEAF(1) // one more leaf node visited
217  ANN_PTS(n_pts) // increment points visited
218  ANNptsVisited += n_pts; // increment number of points visited
219 }
#define ANN_LEAF(n)
Definition: ANNperf.h:132
#define ANN_PTS(n)
Definition: ANNperf.h:135
ANNpoint ANNprQ
void insert(PQkey kv, PQinfo inf)
Definition: pr_queue.h:84
ANNpr_queue * ANNprBoxPQ
ANNcoord cd_bnds[2]
Definition: kd_tree.h:146
PQKkey max_key()
Definition: pr_queue_k.h:90
#define ANN_SPL(n)
Definition: ANNperf.h:133
ANNmin_k * ANNprPointMK
double ANNcoord
Definition: ANN.h:158
int ANNprDim
double eps
Definition: ann_sample.cpp:55
const ANNbool ANN_ALLOW_SELF_MATCH
Definition: ANN.h:235
#define ANN_SUM(x, y)
Definition: ANN.h:362
int ANNptsVisited
Definition: ANN.cpp:191
#define ANN_DIFF(x, y)
Definition: ANN.h:363
double ANNprMaxErr
int n_pts
Definition: kd_tree.h:93
ANNpoint bnd_box_lo
Definition: ANN.h:713
Definition: ANNx.h:45
ANNdist * ANNdistArray
Definition: ANN.h:377
ANNkd_ptr root
Definition: ANN.h:712
void annkPriSearch(ANNpoint q, int k, ANNidxArray nn_idx, ANNdistArray dd, double eps=0.0)
ANNkd_leaf * KD_TRIVIAL
Definition: kd_tree.cpp:50
double ANNprEps
virtual void ann_pri_search(ANNdist)
int dim
Definition: ANN.h:707
ANNidxArray bkt
Definition: kd_tree.h:94
ANNkd_ptr child[2]
Definition: kd_tree.h:148
ANNpointArray ANNprPts
int n_pts
Definition: ANN.h:708
ANNpoint bnd_box_hi
Definition: ANN.h:714
ANNpoint * ANNpointArray
Definition: ANN.h:376
#define ANN_FLOP(n)
Definition: ANNperf.h:131
ANNbool non_empty()
Definition: pr_queue.h:78
int cut_dim
Definition: kd_tree.h:144
virtual void ann_pri_search(ANNdist)=0
Definition: ANNx.h:45
void insert(PQKkey kv, PQKinfo inf)
Definition: pr_queue_k.h:99
virtual void ann_pri_search(ANNdist)
#define ANN_COORD(n)
Definition: ANNperf.h:136
ANNcoord * ANNpoint
Definition: ANN.h:375
double ANNdist
Definition: ANN.h:159
PQKinfo ith_smallest_info(int i)
Definition: pr_queue_k.h:96
ANNdist annBoxDistance(const ANNpoint q, const ANNpoint lo, const ANNpoint hi, int dim)
Definition: kd_util.cpp:124
ANNpointArray pts
Definition: ANN.h:710
PQKkey ith_smallest_key(int i)
Definition: pr_queue_k.h:93
#define ANN_POW(v)
Definition: ANN.h:360
void extr_min(PQkey &kv, PQinfo &inf)
Definition: pr_queue.h:102
int k
Definition: ann_sample.cpp:53
ANNidx * ANNidxArray
Definition: ANN.h:378
int ANNmaxPtsVisited
Definition: ANN.cpp:190
ANNcoord cut_val
Definition: kd_tree.h:145

Generated on Thu Jun 11 2015 13:52:31 for queso-0.53.0 by  doxygen 1.8.5