/* ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;; File: edges.c ;;; Author: Eero Simoncelli ;;; Description: Boundary handling routines for use with convolve.c ;;; Creation Date: Spring 1987. ;;; MODIFIED, 6/96, to operate on double float arrays. ;;; MODIFIED by dgp, 4/1/97, to support THINK C. ;;; MODIFIED, 8/97: reflect1, reflect2, repeat, extend upgraded to ;;; work properly for non-symmetric filters. Added qreflect2 to handle ;;; even-length QMF's which broke under the reflect2 modification. ;;; ---------------------------------------------------------------- ;;; Object-Based Vision and Image Understanding System (OBVIUS), ;;; Copyright 1988, Vision Science Group, Media Laboratory, ;;; Massachusetts Institute of Technology. ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; */ /* This file contains functions which determine how edges are to be handled when performing convolutions of images with linear filters. Any edge handling function which is local and linear may be defined, except (unfortunately) constants cannot be added. So to treat the edges as if the image is surrounded by a gray field, you must paste it into a gray image, convolve, and crop it out... The main convolution functions are called internal_reduce and internal_expand and are defined in the file convolve.c. The idea is that the convolution function calls the edge handling function which computes a new filter based on the old filter and the distance to the edge of the image. For example, reflection is done by reflecting the filter through the appropriate axis and summing. Currently defined functions are listed below. */ #include #include #include #include "convolve.h" #define sgn(a) ( ((a)>0)?1:(((a)<0)?-1:0) ) #define clip(a,mn,mx) ( ((a)<(mn))?(mn):(((a)>=(mx))?(mx-1):(a)) ) int reflect1(), reflect2(), qreflect2(), repeat(), zero(), Extend(), nocompute(); int ereflect(), predict(); /* Lookup table matching a descriptive string to the edge-handling function */ #if !THINK_C static EDGE_HANDLER edge_foos[] = { { "dont-compute", nocompute }, /* zero output for filter touching edge */ { "zero", zero }, /* zero outside of image */ { "repeat", repeat }, /* repeat edge pixel */ { "reflect1", reflect1 }, /* reflect about edge pixels */ { "reflect2", reflect2 }, /* reflect image, including edge pixels */ { "qreflect2", qreflect2 }, /* reflect image, including edge pixels for even-length QMF decompositions */ { "extend", Extend }, /* extend (reflect & invert) */ { "ereflect", ereflect }, /* orthogonal QMF reflection */ }; #else /* This is really stupid, but THINK C won't allow initialization of static variables in a code resource with string addresses. So we do it this way. The 68K code for a MATLAB 4 MEX file can only be created by THINK C. However, for MATLAB 5, we'll be able to use Metrowerks CodeWarrior for both 68K and PPC, so this cludge can be dropped when we drop support for MATLAB 4. Denis Pelli, 4/1/97. */ static EDGE_HANDLER edge_foos[8]; void InitializeTable(EDGE_HANDLER edge_foos[]) { static int i=0; if(i>0) return; edge_foos[i].name="dont-compute"; edge_foos[i++].func=nocompute; edge_foos[i].name="zero"; edge_foos[i++].func=zero; edge_foos[i].name="repeat"; edge_foos[i++].func=repeat; edge_foos[i].name="reflect1"; edge_foos[i++].func=reflect1; edge_foos[i].name="reflect2"; edge_foos[i++].func=reflect2; edge_foos[i].name="qreflect2"; edge_foos[i++].func=qreflect2; edge_foos[i].name="extend"; edge_foos[i++].func=Extend; edge_foos[i].name="ereflect"; edge_foos[i++].func=ereflect; } #endif /* Function looks up an edge handler id string in the structure above, and returns the associated function */ fptr edge_function(char *edges) { int i; #if THINK_C InitializeTable(edge_foos); #endif for (i = 0; i1) OR (x_pos<-1) OR (y_pos>1) OR (y_pos<-1) ) for (i=0; i0)?(x_pos-1):((x_pos<0)?(x_pos+1):0)); int y_start = x_dim * ((y_pos>0)?(y_pos-1):((y_pos<0)?(y_pos+1):0)); int i; for (i=0; i= 0) AND (y_res < filt_sz)) for (x_filt=y_filt, x_res=x_start; x_filt= 0) AND (x_res < x_dim)) result[y_res+x_res] = filt[x_filt]; return(0); } /* -------------------------------------------------------------------- reflect1() - Reflection through the edge pixels. Continuous, but discontinuous first derivative. This is the right thing to do if you are subsampling by 2, since it maintains parity (even pixels positions remain even, odd ones remain odd). */ int reflect1(filt,x_dim,y_dim,x_pos,y_pos,result,r_or_e) register double *filt, *result; register int x_dim; int y_dim, x_pos, y_pos, r_or_e; { int filt_sz = x_dim*y_dim; register int y_filt,x_filt, y_res, x_res; register int x_base = (x_pos>0)?(x_dim-1):0; register int y_base = x_dim * ((y_pos>0)?(y_dim-1):0); int x_overhang = (x_pos>0)?(x_pos-1):((x_pos<0)?(x_pos+1):0); int y_overhang = x_dim * ((y_pos>0)?(y_pos-1):((y_pos<0)?(y_pos+1):0)); int i; int mx_pos = (x_pos<0)?(x_dim/2):((x_dim-1)/2); int my_pos = x_dim * ((y_pos<0)?(y_dim/2):((y_dim-1)/2)); for (i=0; i y_base-filt_sz; y_filt-=x_dim, y_res-=x_dim) { for (x_res=x_base, x_filt=x_base-x_overhang; x_filt > x_base-x_dim; x_res--, x_filt--) result[ABS(y_res)+ABS(x_res)] += filt[ABS(y_filt)+ABS(x_filt)]; if ((x_overhang ISNT mx_pos) AND (x_pos ISNT 0)) for (x_res=x_base, x_filt=x_base-2*mx_pos+x_overhang; x_filt > x_base-x_dim; x_res--, x_filt--) result[ABS(y_res)+ABS(x_res)] += filt[ABS(y_filt)+ABS(x_filt)]; } if ((y_overhang ISNT my_pos) AND (y_pos ISNT 0)) for (y_res=y_base, y_filt = y_base-2*my_pos+y_overhang; y_filt > y_base-filt_sz; y_filt-=x_dim, y_res-=x_dim) { for (x_res=x_base, x_filt=x_base-x_overhang; x_filt > x_base-x_dim; x_res--, x_filt--) result[ABS(y_res)+ABS(x_res)] += filt[ABS(y_filt)+ABS(x_filt)]; if ((x_overhang ISNT mx_pos) AND (x_pos ISNT 0)) for (x_res=x_base, x_filt=x_base-2*mx_pos+x_overhang; x_filt > x_base-x_dim; x_res--, x_filt--) result[ABS(y_res)+ABS(x_res)] += filt[ABS(y_filt)+ABS(x_filt)]; } } return(0); } /* -------------------------------------------------------------------- reflect2() - Reflect image at boundary. The edge pixel is repeated, then the next pixel, etc. Continuous, but discontinuous first derivative. */ int reflect2(filt,x_dim,y_dim,x_pos,y_pos,result,r_or_e) register double *filt, *result; register int x_dim; int y_dim, x_pos, y_pos, r_or_e; { int filt_sz = x_dim*y_dim; register int y_filt,x_filt, y_res, x_res; register int x_base = (x_pos>0)?(x_dim-1):0; register int y_base = x_dim * ((y_pos>0)?(y_dim-1):0); int x_overhang = (x_pos>0)?(x_pos-1):((x_pos<0)?(x_pos+1):0); int y_overhang = x_dim * ((y_pos>0)?(y_pos-1):((y_pos<0)?(y_pos+1):0)); int i; int mx_pos = (x_pos<0)?(x_dim/2):((x_dim-1)/2); int my_pos = x_dim * ((y_pos<0)?(y_dim/2):((y_dim-1)/2)); for (i=0; i0)?x_dim:0); y_filt0)?1:0); x_filt y_base-filt_sz; y_filt-=x_dim, y_res-=x_dim) { for (x_res=x_base, x_filt=x_base-x_overhang; x_filt > x_base-x_dim; x_res--, x_filt--) result[ABS(y_res)+ABS(x_res)] += filt[ABS(y_filt)+ABS(x_filt)]; if (x_pos ISNT 0) for (x_res=x_base, x_filt=x_base-2*mx_pos+x_overhang-1; x_filt > x_base-x_dim; x_res--, x_filt--) result[ABS(y_res)+ABS(x_res)] += filt[ABS(y_filt)+ABS(x_filt)]; } if (y_pos ISNT 0) for (y_res=y_base, y_filt = y_base-2*my_pos+y_overhang-x_dim; y_filt > y_base-filt_sz; y_filt-=x_dim, y_res-=x_dim) { for (x_res=x_base, x_filt=x_base-x_overhang; x_filt > x_base-x_dim; x_res--, x_filt--) result[ABS(y_res)+ABS(x_res)] += filt[ABS(y_filt)+ABS(x_filt)]; if (x_pos ISNT 0) for (x_res=x_base, x_filt=x_base-2*mx_pos+x_overhang-1; x_filt > x_base-x_dim; x_res--, x_filt--) result[ABS(y_res)+ABS(x_res)] += filt[ABS(y_filt)+ABS(x_filt)]; } } return(0); } /* -------------------------------------------------------------------- qreflect2() - Modified version of reflect2 that works properly for even-length QMF filters. */ int qreflect2(filt,x_dim,y_dim,x_pos,y_pos,result,r_or_e) double *filt, *result; int x_dim, y_dim, x_pos, y_pos, r_or_e; { reflect2(filt,x_dim,y_dim,x_pos,y_pos,result,0); return(0); } /* -------------------------------------------------------------------- repeat() - repeat edge pixel. Continuous, with discontinuous first derivative. */ int repeat(filt,x_dim,y_dim,x_pos,y_pos,result,r_or_e) register double *filt, *result; register int x_dim; int y_dim, x_pos, y_pos, r_or_e; { register int y_filt,x_filt, y_res,x_res, y_tmp, x_tmp; register int x_base = (x_pos>0)?(x_dim-1):0; register int y_base = x_dim * ((y_pos>0)?(y_dim-1):0); int x_overhang = ((x_pos>0)?(x_pos-1):((x_pos<0)?(x_pos+1):0)); int y_overhang = x_dim * ((y_pos>0)?(y_pos-1):((y_pos<0)?(y_pos+1):0)); int filt_sz = x_dim*y_dim; int mx_pos = (x_dim/2); int my_pos = x_dim * (y_dim/2); int i; for (i=0; i=0)?((y_res=0)?((x_res y_base-filt_sz; y_filt-=x_dim, y_res-=x_dim) if ((x_base-x_overhang) ISNT mx_pos) for (x_res=x_base, x_filt=x_base-ABS(x_overhang); x_filt > x_base-x_dim; x_res--, x_filt--) result[ABS(y_res)+ABS(x_res)] += filt[ABS(y_filt)+ABS(x_filt)]; else /* ((x_base-x_overhang) IS mx_pos) */ for (x_res=x_base, x_filt=x_base-ABS(x_overhang); x_filt > x_base-x_dim; x_filt--, x_res--) for(x_tmp=x_filt; x_tmp > x_base-x_dim; x_tmp--) result[ABS(y_res)+ABS(x_res)] += filt[ABS(y_filt)+ABS(x_tmp)]; else /* ((y_base-y_overhang) IS my_pos) */ for (y_res=y_base, y_filt=y_base-ABS(y_overhang); y_filt > y_base-filt_sz; y_filt-=x_dim, y_res-=x_dim) for (y_tmp=y_filt; y_tmp > y_base-filt_sz; y_tmp-=x_dim) if ((x_base-x_overhang) ISNT mx_pos) for (x_res=x_base, x_filt=x_base-ABS(x_overhang); x_filt > x_base-x_dim; x_filt--, x_res--) result[ABS(y_res)+ABS(x_res)] += filt[ABS(y_tmp)+ABS(x_filt)]; else /* ((x_base-x_overhang) IS mx_pos) */ for (x_res=x_base, x_filt=x_base-ABS(x_overhang); x_filt > x_base-x_dim; x_filt--, x_res--) for (x_tmp=x_filt; x_tmp > x_base-x_dim; x_tmp--) result[ABS(y_res)+ABS(x_res)] += filt[ABS(y_tmp)+ABS(x_tmp)]; } /* End, if r_or_e IS EXPAND */ return(0); } /* -------------------------------------------------------------------- extend() - Extend image by reflecting and inverting about edge pixel value. Maintains continuity in intensity AND first derivative (but not higher derivs). */ int Extend(filt,x_dim,y_dim,x_pos,y_pos,result,r_or_e) register double *filt, *result; register int x_dim; int y_dim, x_pos, y_pos, r_or_e; { int filt_sz = x_dim*y_dim; register int y_filt,x_filt, y_res,x_res, y_tmp, x_tmp; register int x_base = (x_pos>0)?(x_dim-1):0; register int y_base = x_dim * ((y_pos>0)?(y_dim-1):0); int x_overhang = (x_pos>0)?(x_pos-1):((x_pos<0)?(x_pos+1):0); int y_overhang = x_dim * ((y_pos>0)?(y_pos-1):((y_pos<0)?(y_pos+1):0)); int mx_pos = (x_pos<0)?(x_dim/2):((x_dim-1)/2); int my_pos = x_dim * ((y_pos<0)?(y_dim/2):((y_dim-1)/2)); int i; for (i=0; i=0) AND (y_res=0) AND (x_res=0) AND (x_res y_base-filt_sz; y_filt-=x_dim, y_res-=x_dim) { for (x_res=x_base, x_filt=x_base-x_overhang; x_filt > x_base-x_dim; x_res--, x_filt--) result[ABS(y_res)+ABS(x_res)] += filt[ABS(y_filt)+ABS(x_filt)]; if (x_pos ISNT 0) if (x_overhang ISNT mx_pos) for (x_res=x_base, x_filt=x_base-2*mx_pos+x_overhang; x_filt > x_base-x_dim; x_res--, x_filt--) result[ABS(y_res)+ABS(x_res)] -= filt[ABS(y_filt)+ABS(x_filt)]; else /* x_overhang IS mx_pos */ for (x_res=x_base, x_filt=x_base-2*mx_pos+x_overhang-1; x_filt > x_base-x_dim; x_res--, x_filt--) for (x_tmp=x_filt; x_tmp > x_base-x_dim; x_tmp--) result[ABS(y_res)+ABS(x_res)] += 2*filt[ABS(y_filt)+ABS(x_tmp)]; } if (y_pos ISNT 0) if (y_overhang ISNT my_pos) for (y_res=y_base, y_filt = y_base-2*my_pos+y_overhang; y_filt > y_base-filt_sz; y_filt-=x_dim, y_res-=x_dim) { for (x_res=x_base, x_filt=x_base-x_overhang; x_filt > x_base-x_dim; x_res--, x_filt--) result[ABS(y_res)+ABS(x_res)] -= filt[ABS(y_filt)+ABS(x_filt)]; if ((x_pos ISNT 0) AND (x_overhang ISNT mx_pos)) for (x_res=x_base, x_filt=x_base-2*mx_pos+x_overhang; x_filt > x_base-x_dim; x_res--, x_filt--) result[ABS(y_res)+ABS(x_res)] -= filt[ABS(y_filt)+ABS(x_filt)]; } else /* y_overhang IS my_pos */ for (y_res=y_base, y_filt = y_base-2*my_pos+y_overhang-x_dim; y_filt > y_base-filt_sz; y_res-=x_dim, y_filt-=x_dim) for (y_tmp=y_filt; y_tmp > y_base-filt_sz; y_tmp-=x_dim) { for (x_res=x_base, x_filt=x_base-x_overhang; x_filt > x_base-x_dim; x_res--, x_filt--) result[ABS(y_res)+ABS(x_res)] += 2*filt[ABS(y_tmp)+ABS(x_filt)]; if ((x_pos ISNT 0) AND (x_overhang IS mx_pos)) for (x_res=x_base, x_filt=x_base-2*mx_pos+x_overhang-1; x_filt > x_base-x_dim; x_res--, x_filt--) for (x_tmp=x_filt; x_tmp > x_base-x_dim; x_tmp--) result[ABS(y_res)+ABS(x_res)] += 2*filt[ABS(y_tmp)+ABS(x_tmp)]; } } /* r_or_e ISNT REDUCE */ return(0); } /* -------------------------------------------------------------------- predict() - Simple prediction. Like zero, but multiplies the result by the reciprocal of the percentage of filter being used. (i.e. if 50% of the filter is hanging over the edge of the image, multiply the taps being used by 2). */ int predict(filt,x_dim,y_dim,x_pos,y_pos,result,r_or_e) register double *filt, *result; register int x_dim; int y_dim, x_pos, y_pos, r_or_e; { register int y_filt,x_filt, y_res,x_res; register double taps_used = 0.0; /* int *** */ register double fraction = 0.0; int filt_sz = x_dim*y_dim; int x_start = ((x_pos>0)?(x_pos-1):((x_pos<0)?(x_pos+1):0)); int y_start = x_dim * ((y_pos>0)?(y_pos-1):((y_pos<0)?(y_pos+1):0)); int i; for (i=0; i= 0) AND (y_res < filt_sz)) for (x_filt=y_filt, x_res=x_start; x_filt= 0) AND (x_res < x_dim)) { result[y_res+x_res] = filt[x_filt]; taps_used += ABS(filt[x_filt]); } if (r_or_e IS REDUCE) { /* fraction = ( (double) filt_sz ) / ( (double) taps_used ); */ for (i=0; i0)?(x_dim-1):0; register int y_base = x_dim * ((y_pos>0)?(y_dim-1):0); int filt_sz = x_dim*y_dim; int x_overhang = (x_pos>1)?(x_pos-x_dim):((x_pos<-1)?(x_pos+1):0); int y_overhang = x_dim * ( (y_pos>1)?(y_pos-y_dim):((y_pos<-1)?(y_pos+1):0) ); int i; double norm,onorm; for (i=0; i