• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * principal component analysis (PCA)
3  * Copyright (c) 2004 Michael Niedermayer <michaelni@gmx.at>
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 /**
23  * @file
24  * principal component analysis (PCA)
25  */
26 
27 #include "common.h"
28 #include "pca.h"
29 
30 typedef struct PCA{
31     int count;
32     int n;
33     double *covariance;
34     double *mean;
35     double *z;
36 }PCA;
37 
ff_pca_init(int n)38 PCA *ff_pca_init(int n){
39     PCA *pca;
40     if(n<=0)
41         return NULL;
42 
43     pca= av_mallocz(sizeof(*pca));
44     if (!pca)
45         return NULL;
46 
47     pca->n= n;
48     pca->z = av_malloc_array(n, sizeof(*pca->z));
49     pca->count=0;
50     pca->covariance= av_calloc(n*n, sizeof(double));
51     pca->mean= av_calloc(n, sizeof(double));
52 
53     if (!pca->z || !pca->covariance || !pca->mean) {
54         ff_pca_free(pca);
55         return NULL;
56     }
57 
58     return pca;
59 }
60 
ff_pca_free(PCA * pca)61 void ff_pca_free(PCA *pca){
62     av_freep(&pca->covariance);
63     av_freep(&pca->mean);
64     av_freep(&pca->z);
65     av_free(pca);
66 }
67 
ff_pca_add(PCA * pca,const double * v)68 void ff_pca_add(PCA *pca, const double *v){
69     int i, j;
70     const int n= pca->n;
71 
72     for(i=0; i<n; i++){
73         pca->mean[i] += v[i];
74         for(j=i; j<n; j++)
75             pca->covariance[j + i*n] += v[i]*v[j];
76     }
77     pca->count++;
78 }
79 
ff_pca(PCA * pca,double * eigenvector,double * eigenvalue)80 int ff_pca(PCA *pca, double *eigenvector, double *eigenvalue){
81     int i, j, pass;
82     int k=0;
83     const int n= pca->n;
84     double *z = pca->z;
85 
86     memset(eigenvector, 0, sizeof(double)*n*n);
87 
88     for(j=0; j<n; j++){
89         pca->mean[j] /= pca->count;
90         eigenvector[j + j*n] = 1.0;
91         for(i=0; i<=j; i++){
92             pca->covariance[j + i*n] /= pca->count;
93             pca->covariance[j + i*n] -= pca->mean[i] * pca->mean[j];
94             pca->covariance[i + j*n] = pca->covariance[j + i*n];
95         }
96         eigenvalue[j]= pca->covariance[j + j*n];
97         z[j]= 0;
98     }
99 
100     for(pass=0; pass < 50; pass++){
101         double sum=0;
102 
103         for(i=0; i<n; i++)
104             for(j=i+1; j<n; j++)
105                 sum += fabs(pca->covariance[j + i*n]);
106 
107         if(sum == 0){
108             for(i=0; i<n; i++){
109                 double maxvalue= -1;
110                 for(j=i; j<n; j++){
111                     if(eigenvalue[j] > maxvalue){
112                         maxvalue= eigenvalue[j];
113                         k= j;
114                     }
115                 }
116                 eigenvalue[k]= eigenvalue[i];
117                 eigenvalue[i]= maxvalue;
118                 for(j=0; j<n; j++){
119                     double tmp= eigenvector[k + j*n];
120                     eigenvector[k + j*n]= eigenvector[i + j*n];
121                     eigenvector[i + j*n]= tmp;
122                 }
123             }
124             return pass;
125         }
126 
127         for(i=0; i<n; i++){
128             for(j=i+1; j<n; j++){
129                 double covar= pca->covariance[j + i*n];
130                 double t,c,s,tau,theta, h;
131 
132                 if(pass < 3 && fabs(covar) < sum / (5*n*n)) //FIXME why pass < 3
133                     continue;
134                 if(fabs(covar) == 0.0) //FIXME should not be needed
135                     continue;
136                 if(pass >=3 && fabs((eigenvalue[j]+z[j])/covar) > (1LL<<32) && fabs((eigenvalue[i]+z[i])/covar) > (1LL<<32)){
137                     pca->covariance[j + i*n]=0.0;
138                     continue;
139                 }
140 
141                 h= (eigenvalue[j]+z[j]) - (eigenvalue[i]+z[i]);
142                 theta=0.5*h/covar;
143                 t=1.0/(fabs(theta)+sqrt(1.0+theta*theta));
144                 if(theta < 0.0) t = -t;
145 
146                 c=1.0/sqrt(1+t*t);
147                 s=t*c;
148                 tau=s/(1.0+c);
149                 z[i] -= t*covar;
150                 z[j] += t*covar;
151 
152 #define ROTATE(a,i,j,k,l) {\
153     double g=a[j + i*n];\
154     double h=a[l + k*n];\
155     a[j + i*n]=g-s*(h+g*tau);\
156     a[l + k*n]=h+s*(g-h*tau); }
157                 for(k=0; k<n; k++) {
158                     if(k!=i && k!=j){
159                         ROTATE(pca->covariance,FFMIN(k,i),FFMAX(k,i),FFMIN(k,j),FFMAX(k,j))
160                     }
161                     ROTATE(eigenvector,k,i,k,j)
162                 }
163                 pca->covariance[j + i*n]=0.0;
164             }
165         }
166         for (i=0; i<n; i++) {
167             eigenvalue[i] += z[i];
168             z[i]=0.0;
169         }
170     }
171 
172     return -1;
173 }
174