libvmaf/src/feature/integer_ssim.c (195 lines of code) (raw):

/* Copyright 2001-2012 Xiph.Org and contributors. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: - Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include <math.h> #include <string.h> #include "feature_collector.h" #include "feature_extractor.h" #define KERNEL_SHIFT (8) #define KERNEL_WEIGHT (1<<KERNEL_SHIFT) #define KERNEL_ROUND ((1<<KERNEL_SHIFT)>>1) #ifndef M_PI #define M_PI 3.141592653589793238462643 #endif static int gaussian_filter_init(unsigned **_kernel,double _sigma,int _max_len){ unsigned *kernel; double scale; double nhisigma2; double s; double len; unsigned sum; int kernel_len; int kernel_sz; int ci; scale=1/(sqrt(2*M_PI)*_sigma); nhisigma2=-0.5/(_sigma*_sigma); /*Compute the kernel size so that the error in the first truncated coefficient is no larger than 0.5*KERNEL_WEIGHT. There is no point in going beyond this given our working precision.*/ s=sqrt(0.5*M_PI)*_sigma*(1.0/KERNEL_WEIGHT); if(s>=1)len=0; else len=floor(_sigma*sqrt(-2*log(s))); kernel_len=len>=_max_len?_max_len-1:(int)len; kernel_sz=kernel_len<<1|1; kernel=(unsigned *)malloc(kernel_sz*sizeof(*kernel)); sum=0; for(ci=kernel_len;ci>0;ci--){ kernel[kernel_len-ci]=kernel[kernel_len+ci]= (unsigned)(KERNEL_WEIGHT*scale*exp(nhisigma2*ci*ci)+0.5); sum+=kernel[kernel_len-ci]; } kernel[kernel_len]=KERNEL_WEIGHT-(sum<<1); *_kernel=kernel; return kernel_sz; } typedef struct ssim_moments ssim_moments; struct ssim_moments{ int64_t mux; int64_t muy; int64_t x2; int64_t xy; int64_t y2; int64_t w; }; #define SSIM_K1 (0.01*0.01) #define SSIM_K2 (0.03*0.03) static double calc_ssim(const unsigned char *_src,int _systride, const unsigned char *_dst,int _dystride,double _par,int depth,int _w,int _h){ ssim_moments *line_buf; ssim_moments **lines; double ssim; double ssimw; unsigned *hkernel; int hkernel_sz; int hkernel_offs; unsigned *vkernel; int vkernel_sz; int vkernel_offs; int log_line_sz; int line_sz; int line_mask; int x; int y; int samplemax; samplemax = (1 << depth) - 1; vkernel_sz=gaussian_filter_init(&vkernel,1.5,5); vkernel_offs=vkernel_sz>>1; for(line_sz=1,log_line_sz=0;line_sz<vkernel_sz;line_sz<<=1,log_line_sz++); line_mask=line_sz-1; lines=(ssim_moments **)malloc(line_sz*sizeof(*lines)); lines[0]=line_buf=(ssim_moments *)malloc(line_sz*_w*sizeof(*line_buf)); for(y=1;y<line_sz;y++)lines[y]=lines[y-1]+_w; hkernel_sz=gaussian_filter_init(&hkernel,1.5,5); hkernel_offs=hkernel_sz>>1; ssim=0; ssimw=0; for(y=0;y<_h+vkernel_offs;y++){ ssim_moments *buf; int k; int k_min; int k_max; if(y<_h){ buf=lines[y&line_mask]; for(x=0;x<_w;x++){ ssim_moments m; memset(&m,0,sizeof(m)); k_min=hkernel_offs-x<=0?0:hkernel_offs-x; k_max=x+hkernel_offs-_w+1<=0? hkernel_sz:hkernel_sz-(x+hkernel_offs-_w+1); for(k=k_min;k<k_max;k++){ signed s; signed d; signed window; if (depth > 8) { s = _src[(x-hkernel_offs+k)*2] + (_src[(x-hkernel_offs+k)*2 + 1] << 8); d = _dst[(x-hkernel_offs+k)*2] + (_dst[(x-hkernel_offs+k)*2 + 1] << 8); } else { s=_src[(x-hkernel_offs+k)]; d=_dst[(x-hkernel_offs+k)]; } window=hkernel[k]; m.mux+=window*s; m.muy+=window*d; m.x2+=window*s*s; m.xy+=window*s*d; m.y2+=window*d*d; m.w+=window; } *(buf+x)=*&m; } _src+=_systride; _dst+=_dystride; } if(y>=vkernel_offs){ k_min=vkernel_sz-y-1<=0?0:vkernel_sz-y-1; k_max=y+1-_h<=0?vkernel_sz:vkernel_sz-(y+1-_h); for(x=0;x<_w;x++){ ssim_moments m; double c1; double c2; double mx2; double mxy; double my2; double w; memset(&m,0,sizeof(m)); for(k=k_min;k<k_max;k++){ signed window; buf = lines[(y + 1 - vkernel_sz + k) & line_mask] + x; window=vkernel[k]; m.mux+=window*buf->mux; m.muy+=window*buf->muy; m.x2+=window*buf->x2; m.xy+=window*buf->xy; m.y2+=window*buf->y2; m.w+=window*buf->w; } w=m.w; c1=samplemax*samplemax*SSIM_K1*w*w; c2=samplemax*samplemax*SSIM_K2*w*w; mx2=m.mux*(double)m.mux; mxy=m.mux*(double)m.muy; my2=m.muy*(double)m.muy; ssim+=m.w*(2*mxy+c1)*(c2+2*(m.xy*w-mxy))/ ((mx2+my2+c1)*(m.x2*w-mx2+m.y2*w-my2+c2)); ssimw+=m.w; } } } free(line_buf); free(lines); free(vkernel); free(hkernel); return ssim/ssimw; } static int init(VmafFeatureExtractor *fex, enum VmafPixelFormat pix_fmt, unsigned bpc, unsigned w, unsigned h) { return 0; } static int extract(VmafFeatureExtractor *fex, VmafPicture *ref_pic, VmafPicture *ref_pic_90, VmafPicture *dist_pic, VmafPicture *dist_pic_90, unsigned index, VmafFeatureCollector *feature_collector) { (void) ref_pic_90; (void) dist_pic_90; double score = calc_ssim(ref_pic->data[0], ref_pic->stride[0], dist_pic->data[0], dist_pic->stride[0], 1.0, ref_pic->bpc, ref_pic->w[0], ref_pic->h[0]); int err = vmaf_feature_collector_append(feature_collector, "ssim", score, index); if (err) return err; return 0; } static int close(VmafFeatureExtractor *fex) { return 0; } static const char *provided_features[] = { "ssim", NULL }; VmafFeatureExtractor vmaf_fex_ssim = { .name = "ssim", .init = init, .extract = extract, .close = close, .provided_features = provided_features, };