From 5e7d7d3d95ecf2ce2b50ff4b8d5dbea0ae190e32 Mon Sep 17 00:00:00 2001 From: Yaowu Xu Date: Fri, 13 Jan 2012 14:37:17 -0800 Subject: new loop filter functions for macroblock boundaries The commit adds a new set of loop filter for macroblock edge filtering. The new loop filter has a mask to detect so-called "flat" regions. The detection checks 5 pixels of each side of an edge. If the all pixels have value with +/-1 from the edge pixel on the same side, the region is treated as a "flat" region. For such case, a 7 tap filter is used to change 3 pixel values on each side. The 7 taps are: [1, 1, 1, 2, 1, 1, 1]/8 The furthest away pixels used as input are +/-5 away from edge. For non-flat region, we fall back to old filtering. It should be noted here that the thresholds and filter taps may require more optimization for best possible results. Tests on a set of hd clips showed consistent gains: http://www.corp.google.com/~yaowu/no_crawl/mblpf_hd.html (avg psnr: .83% glb psnr: .77% ssim: .82%) Tests on derf set also showed consistent gains: http://www.corp.google.com/~yaowu/no_crawl/mblpf_derf.html (avg psnr: .24% glb psnr: .22% ssim: .48%) Change-Id: I0855b1ff48e79e1175c20b81967137e18b2af352 --- configure | 1 + vp8/common/loopfilter.c | 23 +++++- vp8/common/loopfilter_filters.c | 155 ++++++++++++++++++++++++++++++++++++---- 3 files changed, 161 insertions(+), 18 deletions(-) diff --git a/configure b/configure index 6e96899c9..a955d1466 100755 --- a/configure +++ b/configure @@ -226,6 +226,7 @@ EXPERIMENT_LIST=" qimode uvintra newnear + newlpf " CONFIG_LIST=" external_build diff --git a/vp8/common/loopfilter.c b/vp8/common/loopfilter.c index aa205f10e..052d00e0f 100644 --- a/vp8/common/loopfilter.c +++ b/vp8/common/loopfilter.c @@ -342,18 +342,26 @@ void vp8_loop_filter_frame lfi.hev_thr = lfi_n->hev_thr[hev_index]; if (mb_col > 0) +#if CONFIG_NEWLPF + vp8_loop_filter_mbv_c + (y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi); +#else LF_INVOKE(&cm->rtcd.loopfilter, normal_mb_v) (y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi); - +#endif if (!skip_lf) LF_INVOKE(&cm->rtcd.loopfilter, normal_b_v) (y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi); /* don't apply across umv border */ if (mb_row > 0) +#if CONFIG_NEWLPF + vp8_loop_filter_mbh_c + (y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi); +#else LF_INVOKE(&cm->rtcd.loopfilter, normal_mb_h) (y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi); - +#endif if (!skip_lf) LF_INVOKE(&cm->rtcd.loopfilter, normal_b_h) (y_ptr, u_ptr, v_ptr, post->y_stride, post->uv_stride, &lfi); @@ -454,8 +462,13 @@ void vp8_loop_filter_frame_yonly lfi.hev_thr = lfi_n->hev_thr[hev_index]; if (mb_col > 0) +#if CONFIG_NEWLPF + vp8_loop_filter_mbv_c + (y_ptr, 0, 0, post->y_stride, 0, &lfi); +#else LF_INVOKE(&cm->rtcd.loopfilter, normal_mb_v) (y_ptr, 0, 0, post->y_stride, 0, &lfi); +#endif if (!skip_lf) LF_INVOKE(&cm->rtcd.loopfilter, normal_b_v) @@ -463,9 +476,13 @@ void vp8_loop_filter_frame_yonly /* don't apply across umv border */ if (mb_row > 0) +#if CONFIG_NEWLPF + vp8_loop_filter_mbh_c + (y_ptr, 0, 0, post->y_stride, 0, &lfi); +#else LF_INVOKE(&cm->rtcd.loopfilter, normal_mb_h) (y_ptr, 0, 0, post->y_stride, 0, &lfi); - +#endif if (!skip_lf) LF_INVOKE(&cm->rtcd.loopfilter, normal_b_h) (y_ptr, 0, 0, post->y_stride, 0, &lfi); diff --git a/vp8/common/loopfilter_filters.c b/vp8/common/loopfilter_filters.c index 10228ae09..72ba31eaa 100644 --- a/vp8/common/loopfilter_filters.c +++ b/vp8/common/loopfilter_filters.c @@ -148,7 +148,8 @@ void vp8_loop_filter_vertical_edge_c do { mask = vp8_filter_mask(limit[0], blimit[0], - s[-4], s[-3], s[-2], s[-1], s[0], s[1], s[2], s[3]); + s[-4], s[-3], s[-2], s[-1], + s[0], s[1], s[2], s[3]); hev = vp8_hevmask(thresh[0], s[-2], s[-1], s[0], s[1]); @@ -158,9 +159,100 @@ void vp8_loop_filter_vertical_edge_c } while (++i < count * 8); } +#if CONFIG_NEWLPF +static __inline signed char vp8_flatmask(uc thresh, + uc p4, uc p3, uc p2, uc p1, uc p0, + uc q0, uc q1, uc q2, uc q3, uc q4) +{ + signed char flat = 0; + flat |= (abs(p1 - p0) > 1) * -1; + flat |= (abs(q1 - q0) > 1) * -1; + flat |= (abs(p0 - p2) > 1) * -1; + flat |= (abs(q0 - q2) > 1) * -1; + flat |= (abs(p3 - p0) > 1) * -1; + flat |= (abs(q3 - q0) > 1) * -1; + flat |= (abs(p4 - p0) > 1) * -1; + flat |= (abs(q4 - q0) > 1) * -1; + flat = ~flat; + return flat; +} +static __inline void vp8_mbfilter(signed char mask, uc hev, uc flat, + uc *op4, uc *op3, uc *op2, uc *op1, uc *op0, + uc *oq0, uc *oq1, uc *oq2, uc *oq3, uc *oq4) +{ + /* use a 7 tap filter [1, 1, 1, 2, 1, 1, 1] for flat line */ + if(flat && mask) + { + unsigned char p0, q0; + unsigned char p1, q1; + unsigned char p2, q2; + unsigned char p3, q3; + unsigned char p4, q4; + + p4 = *op4; + p3 = *op3; + p2 = *op2; + p1 = *op1; + p0 = *op0; + q0 = *oq0; + q1 = *oq1; + q2 = *oq2; + q3 = *oq3; + q4 = *oq4; + + *op2 = ( p4 + p4 + p3 + p2 + p2 + p1 + p0 + q0 + 4)>>3; + *op1 = ( p4 + p3 + p2 + p1 + p1 + p0 + q0 + q1 + 4)>>3; + *op0 = ( p3 + p2 + p1 + p0 + p0 + q0 + q1 + q2 + 4)>>3; + *oq0 = ( p2 + p1 + p0 + q0 + q0 + q1 + q2 + q3 + 4)>>3; + *oq1 = ( p1 + p0 + q0 + q1 + q1 + q2 + q3 + q4 + 4)>>3; + *oq2 = ( p0 + q0 + q1 + q2 + q2 + q3 + q4 + q4 + 4)>>3; + } + else + { + signed char ps0, qs0; + signed char ps1, qs1; + signed char vp8_filter, Filter1, Filter2; + signed char u; + + ps1 = (signed char) * op1 ^ 0x80; + ps0 = (signed char) * op0 ^ 0x80; + qs0 = (signed char) * oq0 ^ 0x80; + qs1 = (signed char) * oq1 ^ 0x80; + + /* add outer taps if we have high edge variance */ + vp8_filter = vp8_signed_char_clamp(ps1 - qs1); + vp8_filter &= hev; + + /* inner taps */ + vp8_filter = vp8_signed_char_clamp(vp8_filter + 3 * (qs0 - ps0)); + vp8_filter &= mask; + + Filter1 = vp8_signed_char_clamp(vp8_filter + 4); + Filter2 = vp8_signed_char_clamp(vp8_filter + 3); + Filter1 >>= 3; + Filter2 >>= 3; + u = vp8_signed_char_clamp(qs0 - Filter1); + *oq0 = u ^ 0x80; + u = vp8_signed_char_clamp(ps0 + Filter2); + *op0 = u ^ 0x80; + vp8_filter = Filter1; + + /* outer tap adjustments */ + vp8_filter += 1; + vp8_filter >>= 1; + vp8_filter &= ~hev; + + u = vp8_signed_char_clamp(qs1 - vp8_filter); + *oq1 = u ^ 0x80; + u = vp8_signed_char_clamp(ps1 + vp8_filter); + *op1 = u ^ 0x80; + } +} +#else static __inline void vp8_mbfilter(signed char mask, uc hev, - uc *op2, uc *op1, uc *op0, uc *oq0, uc *oq1, uc *oq2) + uc *op2, uc *op1, uc *op0, + uc *oq0, uc *oq1, uc *oq2) { signed char s, u; signed char vp8_filter, Filter1, Filter2; @@ -213,7 +305,7 @@ static __inline void vp8_mbfilter(signed char mask, uc hev, s = vp8_signed_char_clamp(ps2 + u); *op2 = s ^ 0x80; } - +#endif void vp8_mbloop_filter_horizontal_edge_c ( unsigned char *s, @@ -226,6 +318,9 @@ void vp8_mbloop_filter_horizontal_edge_c { signed char hev = 0; /* high edge variance */ signed char mask = 0; +#if CONFIG_NEWLPF + signed char flat = 0; +#endif int i = 0; /* loop filter designed to work using chars so that we can make maximum use @@ -236,12 +331,21 @@ void vp8_mbloop_filter_horizontal_edge_c mask = vp8_filter_mask(limit[0], blimit[0], s[-4*p], s[-3*p], s[-2*p], s[-1*p], - s[0*p], s[1*p], s[2*p], s[3*p]); + s[ 0*p], s[ 1*p], s[ 2*p], s[ 3*p]); hev = vp8_hevmask(thresh[0], s[-2*p], s[-1*p], s[0*p], s[1*p]); - - vp8_mbfilter(mask, hev, s - 3 * p, s - 2 * p, s - 1 * p, s, s + 1 * p, s + 2 * p); - +#if CONFIG_NEWLPF + flat = vp8_flatmask(thresh[0], + s[-5*p], s[-4*p], s[-3*p], s[-2*p], s[-1*p], + s[ 0*p], s[ 1*p], s[ 2*p], s[ 3*p], s[ 4*p]); + vp8_mbfilter(mask, hev, flat, + s - 5*p, s - 4*p, s- 3*p, s - 2*p, s - 1*p, + s, s + 1*p, s+ 2*p, s + 3*p, s + 4*p ); +#else + vp8_mbfilter(mask, hev, + s - 3*p, s - 2*p, s - 1*p, + s, s + 1*p, s + 2*p); +#endif ++s; } while (++i < count * 8); @@ -261,18 +365,31 @@ void vp8_mbloop_filter_vertical_edge_c { signed char hev = 0; /* high edge variance */ signed char mask = 0; +#if CONFIG_NEWLPF + signed char flat = 0; +#endif int i = 0; do { mask = vp8_filter_mask(limit[0], blimit[0], - s[-4], s[-3], s[-2], s[-1], s[0], s[1], s[2], s[3]); + s[-4], s[-3], s[-2], s[-1], + s[0], s[1], s[2], s[3]); hev = vp8_hevmask(thresh[0], s[-2], s[-1], s[0], s[1]); - - vp8_mbfilter(mask, hev, s - 3, s - 2, s - 1, s, s + 1, s + 2); - +#if CONFIG_NEWLPF + flat = vp8_flatmask(thresh[0], + s[-5],s[-4],s[-3],s[-2],s[-1], + s[ 0],s[ 1],s[ 2],s[ 3],s[ 4]); + vp8_mbfilter(mask, hev, flat, + s - 5, s - 4, s - 3, s - 2, s - 1, + s, s + 1, s + 2, s + 3, s + 4); +#else + vp8_mbfilter(mask, hev, + s - 3, s - 2, s - 1, + s, s + 1, s + 2); +#endif s += p; } while (++i < count * 8); @@ -280,7 +397,9 @@ void vp8_mbloop_filter_vertical_edge_c } /* should we apply any filter at all ( 11111111 yes, 00000000 no) */ -static __inline signed char vp8_simple_filter_mask(uc blimit, uc p1, uc p0, uc q0, uc q1) +static __inline signed char vp8_simple_filter_mask(uc blimit, + uc p1, uc p0, + uc q0, uc q1) { /* Why does this cause problems for win32? * error C2143: syntax error : missing ';' before 'type' @@ -290,7 +409,9 @@ static __inline signed char vp8_simple_filter_mask(uc blimit, uc p1, uc p0, uc q return mask; } -static __inline void vp8_simple_filter(signed char mask, uc *op1, uc *op0, uc *oq0, uc *oq1) +static __inline void vp8_simple_filter(signed char mask, + uc *op1, uc *op0, + uc *oq0, uc *oq1) { signed char vp8_filter, Filter1, Filter2; signed char p1 = (signed char) * op1 ^ 0x80; @@ -327,8 +448,12 @@ void vp8_loop_filter_simple_horizontal_edge_c do { - mask = vp8_simple_filter_mask(blimit[0], s[-2*p], s[-1*p], s[0*p], s[1*p]); - vp8_simple_filter(mask, s - 2 * p, s - 1 * p, s, s + 1 * p); + mask = vp8_simple_filter_mask(blimit[0], + s[-2*p], s[-1*p], + s[0*p], s[1*p]); + vp8_simple_filter(mask, + s - 2 * p, s - 1 * p, + s, s + 1 * p); ++s; } while (++i < 16); -- cgit v1.2.3