tiny_ssim.c 6.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216
  1. /*
  2. * Copyright (c) 2003-2013 Loren Merritt
  3. *
  4. * This program is free software; you can redistribute it and/or modify
  5. * it under the terms of the GNU General Public License as published by
  6. * the Free Software Foundation; either version 2 of the License, or
  7. * (at your option) any later version.
  8. *
  9. * This program is distributed in the hope that it will be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. * GNU General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU General Public License
  15. * along with this program; if not, write to the Free Software
  16. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110 USA
  17. */
  18. /*
  19. * tiny_ssim.c
  20. * Computes the Structural Similarity Metric between two rawYV12 video files.
  21. * original algorithm:
  22. * Z. Wang, A. C. Bovik, H. R. Sheikh and E. P. Simoncelli,
  23. * "Image quality assessment: From error visibility to structural similarity,"
  24. * IEEE Transactions on Image Processing, vol. 13, no. 4, pp. 600-612, Apr. 2004.
  25. *
  26. * To improve speed, this implementation uses the standard approximation of
  27. * overlapped 8x8 block sums, rather than the original gaussian weights.
  28. */
  29. #include <inttypes.h>
  30. #include <math.h>
  31. #include <stdio.h>
  32. #include <stdlib.h>
  33. #define FFSWAP(type,a,b) do{type SWAP_tmp= b; b= a; a= SWAP_tmp;}while(0)
  34. #define FFMIN(a,b) ((a) > (b) ? (b) : (a))
  35. #define BIT_DEPTH 8
  36. #define PIXEL_MAX ((1 << BIT_DEPTH)-1)
  37. typedef uint8_t pixel;
  38. /****************************************************************************
  39. * structural similarity metric
  40. ****************************************************************************/
  41. static void ssim_4x4x2_core( const pixel *pix1, intptr_t stride1,
  42. const pixel *pix2, intptr_t stride2,
  43. int sums[2][4] )
  44. {
  45. int x,y,z;
  46. for( z = 0; z < 2; z++ )
  47. {
  48. uint32_t s1 = 0, s2 = 0, ss = 0, s12 = 0;
  49. for( y = 0; y < 4; y++ )
  50. for( x = 0; x < 4; x++ )
  51. {
  52. int a = pix1[x+y*stride1];
  53. int b = pix2[x+y*stride2];
  54. s1 += a;
  55. s2 += b;
  56. ss += a*a;
  57. ss += b*b;
  58. s12 += a*b;
  59. }
  60. sums[z][0] = s1;
  61. sums[z][1] = s2;
  62. sums[z][2] = ss;
  63. sums[z][3] = s12;
  64. pix1 += 4;
  65. pix2 += 4;
  66. }
  67. }
  68. static float ssim_end1( int s1, int s2, int ss, int s12 )
  69. {
  70. /* Maximum value for 10-bit is: ss*64 = (2^10-1)^2*16*4*64 = 4286582784, which will overflow in some cases.
  71. * s1*s1, s2*s2, and s1*s2 also obtain this value for edge cases: ((2^10-1)*16*4)^2 = 4286582784.
  72. * Maximum value for 9-bit is: ss*64 = (2^9-1)^2*16*4*64 = 1069551616, which will not overflow. */
  73. #if BIT_DEPTH > 9
  74. #define type float
  75. static const float ssim_c1 = .01*.01*PIXEL_MAX*PIXEL_MAX*64;
  76. static const float ssim_c2 = .03*.03*PIXEL_MAX*PIXEL_MAX*64*63;
  77. #else
  78. #define type int
  79. static const int ssim_c1 = (int)(.01*.01*PIXEL_MAX*PIXEL_MAX*64 + .5);
  80. static const int ssim_c2 = (int)(.03*.03*PIXEL_MAX*PIXEL_MAX*64*63 + .5);
  81. #endif
  82. type fs1 = s1;
  83. type fs2 = s2;
  84. type fss = ss;
  85. type fs12 = s12;
  86. type vars = fss*64 - fs1*fs1 - fs2*fs2;
  87. type covar = fs12*64 - fs1*fs2;
  88. return (float)(2*fs1*fs2 + ssim_c1) * (float)(2*covar + ssim_c2)
  89. / ((float)(fs1*fs1 + fs2*fs2 + ssim_c1) * (float)(vars + ssim_c2));
  90. #undef type
  91. }
  92. static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
  93. {
  94. float ssim = 0.0;
  95. int i;
  96. for( i = 0; i < width; i++ )
  97. ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
  98. sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
  99. sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
  100. sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
  101. return ssim;
  102. }
  103. float ssim_plane(
  104. pixel *pix1, intptr_t stride1,
  105. pixel *pix2, intptr_t stride2,
  106. int width, int height, void *buf, int *cnt )
  107. {
  108. int z = 0;
  109. int x, y;
  110. float ssim = 0.0;
  111. int (*sum0)[4] = buf;
  112. int (*sum1)[4] = sum0 + (width >> 2) + 3;
  113. width >>= 2;
  114. height >>= 2;
  115. for( y = 1; y < height; y++ )
  116. {
  117. for( ; z <= y; z++ )
  118. {
  119. FFSWAP( void*, sum0, sum1 );
  120. for( x = 0; x < width; x+=2 )
  121. ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
  122. }
  123. for( x = 0; x < width-1; x += 4 )
  124. ssim += ssim_end4( sum0+x, sum1+x, FFMIN(4,width-x-1) );
  125. }
  126. // *cnt = (height-1) * (width-1);
  127. return ssim / ((height-1) * (width-1));
  128. }
  129. uint64_t ssd_plane( const uint8_t *pix1, const uint8_t *pix2, int size )
  130. {
  131. uint64_t ssd = 0;
  132. int i;
  133. for( i=0; i<size; i++ )
  134. {
  135. int d = pix1[i] - pix2[i];
  136. ssd += d*d;
  137. }
  138. return ssd;
  139. }
  140. double ssd_to_psnr( uint64_t ssd, uint64_t denom )
  141. {
  142. return -10*log((double)ssd/(denom*255*255))/log(10);
  143. }
  144. int main(int argc, char* argv[])
  145. {
  146. FILE *f[2];
  147. uint8_t *buf[2], *plane[2][3];
  148. int *temp;
  149. uint64_t ssd[3] = {0,0,0};
  150. double ssim[3] = {0,0,0};
  151. int frame_size, w, h;
  152. int frames, seek;
  153. int i;
  154. if( argc<4 || 2 != sscanf(argv[3], "%dx%d", &w, &h) )
  155. {
  156. printf("tiny_ssim <file1.yuv> <file2.yuv> <width>x<height> [<seek>]\n");
  157. return -1;
  158. }
  159. f[0] = fopen(argv[1], "rb");
  160. f[1] = fopen(argv[2], "rb");
  161. sscanf(argv[3], "%dx%d", &w, &h);
  162. frame_size = w*h*3/2;
  163. for( i=0; i<2; i++ )
  164. {
  165. buf[i] = malloc(frame_size);
  166. plane[i][0] = buf[i];
  167. plane[i][1] = plane[i][0] + w*h;
  168. plane[i][2] = plane[i][1] + w*h/4;
  169. }
  170. temp = malloc((2*w+12)*sizeof(*temp));
  171. seek = argc<5 ? 0 : atoi(argv[4]);
  172. fseek(f[seek<0], seek < 0 ? -seek : seek, SEEK_SET);
  173. for( frames=0;; frames++ )
  174. {
  175. if( fread(buf[0], frame_size, 1, f[0]) != 1) break;
  176. if( fread(buf[1], frame_size, 1, f[1]) != 1) break;
  177. for( i=0; i<3; i++ )
  178. {
  179. ssd[i] += ssd_plane ( plane[0][i], plane[1][i], w*h>>2*!!i );
  180. ssim[i] += ssim_plane( plane[0][i], w>>!!i,
  181. plane[1][i], w>>!!i,
  182. w>>!!i, h>>!!i, temp, NULL );
  183. }
  184. }
  185. if( !frames ) return 0;
  186. printf( "PSNR Y:%.3f U:%.3f V:%.3f All:%.3f\n",
  187. ssd_to_psnr( ssd[0], (uint64_t)frames*w*h ),
  188. ssd_to_psnr( ssd[1], (uint64_t)frames*w*h/4 ),
  189. ssd_to_psnr( ssd[2], (uint64_t)frames*w*h/4 ),
  190. ssd_to_psnr( ssd[0] + ssd[1] + ssd[2], (uint64_t)frames*w*h*3/2 ) );
  191. printf( "SSIM Y:%.5f U:%.5f V:%.5f All:%.5f\n",
  192. ssim[0] / frames,
  193. ssim[1] / frames,
  194. ssim[2] / frames,
  195. (ssim[0]*4 + ssim[1] + ssim[2]) / (frames*6) );
  196. return 0;
  197. }