/* * pnmpsnr.c: Compute error (RMSE, PSNR) between images * * * Derived from pnmpnsmr by Ulrich Hafner, part of his fiasco package, * On 2001.03.04. * Copyright (C) 1994-2000 Ullrich Hafner */ #include #include #include #include #include "pm_c_util.h" #include "mallocvar.h" #include "nstring.h" #include "pam.h" #include "shhopt.h" struct CmdlineInfo { /* All the information the user supplied in the command line, in a form easy for the program to use. */ const char * inputFile1Name; /* Name of first input file */ const char * inputFile2Name; /* Name of second input file */ unsigned int rgb; unsigned int machine; unsigned int maxSpec; float max; }; static void parseCommandLine(int argc, const char ** argv, struct CmdlineInfo * const cmdlineP) { /*---------------------------------------------------------------------------- Note that the file spec array we return is stored in the storage that was passed to as as the argv array. -----------------------------------------------------------------------------*/ optEntry * option_def; /* Instructions to pm_optParseOptions3 on how to parse our options. */ optStruct3 opt; unsigned int option_def_index; MALLOCARRAY_NOFAIL(option_def, 100); option_def_index = 0; /* incremented by OPTENT3 */ OPTENT3(0, "rgb", OPT_FLAG, NULL, &cmdlineP->rgb, 0); OPTENT3(0, "machine", OPT_FLAG, NULL, &cmdlineP->machine, 0); OPTENT3(0, "max", OPT_FLOAT, &cmdlineP->max, &cmdlineP->maxSpec, 0); opt.opt_table = option_def; opt.short_allowed = FALSE; /* We have no short (old-fashioned) options */ opt.allowNegNum = FALSE; /* We have no parms that are negative numbers */ pm_optParseOptions3(&argc, (char **)argv, opt, sizeof(opt), 0); /* Uses and sets argc, argv, and some of *cmdlineP and others */ if (argc-1 < 2) pm_error("Takes two arguments: names of the two files to compare"); else { cmdlineP->inputFile1Name = argv[1]; cmdlineP->inputFile2Name = argv[2]; if (argc-1 > 2) pm_error("Too many arguments (%u). The only arguments are " "the names of the two files to compare", argc-1); } free(option_def); } static int udiff(unsigned int const subtrahend, unsigned int const subtractor) { return subtrahend - subtractor; } static double square(double const arg) { return(arg * arg); } static void validateInput(struct pam const pam1, struct pam const pam2) { if (pam1.width != pam2.width) pm_error("images are not the same width, so can't be compared. " "The first is %d columns wide, " "while the second is %d columns wide.", pam1.width, pam2.width); if (pam1.height != pam2.height) pm_error("images are not the same height, so can't be compared. " "The first is %d rows high, " "while the second is %d rows high.", pam1.height, pam2.height); if (pam1.maxval != pam2.maxval) pm_error("images do not have the same maxval. This programs works " "only on like maxvals. " "The first image has maxval %u, " "while the second has %u. Use Pamdepth to change the " "maxval of one of them.", (unsigned int) pam1.maxval, (unsigned int) pam2.maxval); if (!streq(pam1.tuple_type, pam2.tuple_type)) pm_error("images are not of the same type. The tuple types are " "'%s' and '%s', respectively.", pam1.tuple_type, pam2.tuple_type); if (!streq(pam1.tuple_type, PAM_PBM_TUPLETYPE) && !streq(pam1.tuple_type, PAM_PGM_TUPLETYPE) && !streq(pam1.tuple_type, PAM_PPM_TUPLETYPE)) pm_error("Images are not of a PNM type. Tuple type is '%s'", pam1.tuple_type); } enum ColorSpaceId { COLORSPACE_GRAYSCALE, COLORSPACE_YCBCR, COLORSPACE_RGB }; typedef struct { enum ColorSpaceId id; unsigned int componentCt; const char * componentName[3]; /* Only first 'componentCt' elements are valid */ } ColorSpace; struct SqDiff { /*---------------------------------------------------------------------------- The square-differences of the components of two pixels, for some component set. -----------------------------------------------------------------------------*/ double sqDiff[3]; }; static struct SqDiff zeroSqDiff() { struct SqDiff retval; unsigned int i; for (i = 0; i < 3; ++i) retval.sqDiff[i] = 0.0; return retval; } static struct SqDiff sqDiffSum(ColorSpace const colorSpace, struct SqDiff const addend, struct SqDiff const adder) { struct SqDiff retval; unsigned int i; for (i = 0; i < colorSpace.componentCt; ++i) retval.sqDiff[i] = addend.sqDiff[i] + adder.sqDiff[i]; return retval; } #define Y_INDEX 0 #define CB_INDEX 1 #define CR_INDEX 2 static ColorSpace yCbCrColorSpace() { ColorSpace retval; retval.id = COLORSPACE_YCBCR; retval.componentCt = 3; retval.componentName[Y_INDEX] = "Y"; retval.componentName[CR_INDEX] = "CR"; retval.componentName[CB_INDEX] = "CB"; return retval; } static struct SqDiff sqDiffYCbCr(tuple const tuple1, tuple const tuple2) { struct SqDiff retval; double y1, y2, cb1, cb2, cr1, cr2; pnm_YCbCrtuple(tuple1, &y1, &cb1, &cr1); pnm_YCbCrtuple(tuple2, &y2, &cb2, &cr2); retval.sqDiff[Y_INDEX] = square(y1 - y2); retval.sqDiff[CB_INDEX] = square(cb1 - cb2); retval.sqDiff[CR_INDEX] = square(cr1 - cr2); return retval; } #define R_INDEX 0 #define G_INDEX 1 #define B_INDEX 2 static ColorSpace rgbColorSpace() { ColorSpace retval; retval.id = COLORSPACE_RGB; retval.componentCt = 3; retval.componentName[R_INDEX] = "Red"; retval.componentName[G_INDEX] = "Green"; retval.componentName[B_INDEX] = "Blue"; return retval; } static struct SqDiff sqDiffRgb(tuple const tuple1, tuple const tuple2) { struct SqDiff retval; retval.sqDiff[R_INDEX] = square((int)tuple1[PAM_RED_PLANE] - (int)tuple2[PAM_RED_PLANE]); retval.sqDiff[G_INDEX] = square((int)tuple1[PAM_GRN_PLANE] - (int)tuple2[PAM_GRN_PLANE]); retval.sqDiff[B_INDEX] = square((int)tuple1[PAM_BLU_PLANE] - (int)tuple2[PAM_BLU_PLANE]); return retval; } static ColorSpace grayscaleColorSpace() { ColorSpace retval; retval.id = COLORSPACE_GRAYSCALE; retval.componentCt = 1; retval.componentName[Y_INDEX] = "luminance"; return retval; } static struct SqDiff sqDiffGrayscale(tuple const tuple1, tuple const tuple2) { struct SqDiff sqDiff; sqDiff.sqDiff[Y_INDEX] = square(udiff(tuple1[0], tuple2[0])); return sqDiff; } static struct SqDiff sumSqDiffFromRaster(struct pam * const pam1P, struct pam * const pam2P, ColorSpace const colorSpace) { struct SqDiff sumSqDiff; tuple *tuplerow1, *tuplerow2; /* malloc'ed */ unsigned int row; tuplerow1 = pnm_allocpamrow(pam1P); tuplerow2 = pnm_allocpamrow(pam2P); sumSqDiff = zeroSqDiff(); for (row = 0; row < pam1P->height; ++row) { unsigned int col; pnm_readpamrow(pam1P, tuplerow1); pnm_readpamrow(pam2P, tuplerow2); assert(pam1P->width == pam2P->width); for (col = 0; col < pam1P->width; ++col) { struct SqDiff sqDiff; switch (colorSpace.id) { case COLORSPACE_GRAYSCALE: sqDiff = sqDiffGrayscale(tuplerow1[col], tuplerow2[col]); break; case COLORSPACE_YCBCR: sqDiff = sqDiffYCbCr(tuplerow1[col], tuplerow2[col]); break; case COLORSPACE_RGB: sqDiff = sqDiffRgb(tuplerow1[col], tuplerow2[col]); break; } sumSqDiff = sqDiffSum(colorSpace, sumSqDiff, sqDiff); } } pnm_freepamrow(tuplerow1); pnm_freepamrow(tuplerow2); return sumSqDiff; } struct Psnr { /*---------------------------------------------------------------------------- The PSNR of an image, in some unspecified color space. -----------------------------------------------------------------------------*/ double psnr[3]; }; static struct Psnr psnrFromSumSqDiff(struct SqDiff const sumSqDiff, double const maxSumSqDiff, unsigned int const componentCt) { /*---------------------------------------------------------------------------- Compute the PSNR from the sums of the squares of the differences in the pixels 'sumSqDiff' (separated by colorpspace component, where there are 'componentCt' components). 'maxSumSqDiff' is the maximum possible sum square difference, i.e. the sum of the squares of the sample differences between an entirely white image and entirely black image of the given dimensions. Where there is no difference between the images, return infinity. -----------------------------------------------------------------------------*/ struct Psnr retval; unsigned int i; /* The PSNR is the ratio of the maximum possible mean square difference to the actual mean square difference, which is also the ratio of the maximum possible sum square difference to the actual sum square difference. Note that in the important special case that the images are identical, the sum square differences are identically 0.0. No precision error; no rounding error. */ for (i = 0; i < componentCt; ++i) { if (sumSqDiff.sqDiff[i] > 0) retval.psnr[i] = 10 * log10(maxSumSqDiff/sumSqDiff.sqDiff[i]); else retval.psnr[i] = 1.0/0.0; } return retval; } static void reportPsnrHuman(struct Psnr const psnr, ColorSpace const colorSpace, const char * const fileName1, const char * const fileName2) { unsigned int i; pm_message("PSNR between '%s' and '%s':", fileName1, fileName2); for (i = 0; i < colorSpace.componentCt; ++i) { const char * label; pm_asprintf(&label, "%s:", colorSpace.componentName[i]); if (isfinite(psnr.psnr[i])) pm_message(" %-6.6s %.2f dB", label, psnr.psnr[i]); else pm_message(" %-6.6s no difference", label); pm_strfree(label); } } static void reportPsnrMachine(struct Psnr const psnr, unsigned int const componentCt, bool const maxSpec, float const max) { unsigned int i; for (i = 0; i < componentCt; ++i) { double const clipped = maxSpec ? MIN(max, psnr.psnr[i]) : psnr.psnr[i]; if (i > 0) fprintf(stdout, " "); fprintf(stdout, "%.2f", clipped); } fprintf(stdout, "\n"); } int main (int argc, const char **argv) { FILE * if1P; FILE * if2P; struct pam pam1, pam2; ColorSpace colorSpace; struct CmdlineInfo cmdline; pm_proginit(&argc, argv); parseCommandLine(argc, argv, &cmdline); if1P = pm_openr(cmdline.inputFile1Name); if2P = pm_openr(cmdline.inputFile2Name); pnm_readpaminit(if1P, &pam1, PAM_STRUCT_SIZE(tuple_type)); pnm_readpaminit(if2P, &pam2, PAM_STRUCT_SIZE(tuple_type)); validateInput(pam1, pam2); if (streq(pam1.tuple_type, PAM_PPM_TUPLETYPE)) { if (cmdline.rgb) colorSpace = rgbColorSpace(); else colorSpace = yCbCrColorSpace(); } else colorSpace = grayscaleColorSpace(); { struct SqDiff const sumSqDiff = sumSqDiffFromRaster(&pam1, &pam2, colorSpace); double const maxSumSqDiff = square(pam1.maxval) * pam1.width * pam1.height; /* Maximum possible sum square difference, i.e. the sum of the squares of the sample differences between an entirely white image and entirely black image of the given dimensions. */ struct Psnr const psnr = psnrFromSumSqDiff( sumSqDiff, maxSumSqDiff, colorSpace.componentCt); if (cmdline.machine) reportPsnrMachine(psnr, colorSpace.componentCt, cmdline.maxSpec, cmdline.max); else reportPsnrHuman(psnr, colorSpace, cmdline.inputFile1Name, cmdline.inputFile2Name); } pm_close(if2P); pm_close(if1P); return 0; }