aom/vp8/encoder/denoising.c

636 строки
24 KiB
C

/*
* Copyright (c) 2012 The WebM project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "denoising.h"
#include "vp8/common/reconinter.h"
#include "vpx/vpx_integer.h"
#include "vpx_mem/vpx_mem.h"
#include "vp8_rtcd.h"
static const unsigned int NOISE_MOTION_THRESHOLD = 25 * 25;
/* SSE_DIFF_THRESHOLD is selected as ~95% confidence assuming
* var(noise) ~= 100.
*/
static const unsigned int SSE_DIFF_THRESHOLD = 16 * 16 * 20;
static const unsigned int SSE_THRESHOLD = 16 * 16 * 40;
static const unsigned int SSE_THRESHOLD_HIGH = 16 * 16 * 60;
/*
* The filter function was modified to reduce the computational complexity.
* Step 1:
* Instead of applying tap coefficients for each pixel, we calculated the
* pixel adjustments vs. pixel diff value ahead of time.
* adjustment = filtered_value - current_raw
* = (filter_coefficient * diff + 128) >> 8
* where
* filter_coefficient = (255 << 8) / (256 + ((absdiff * 330) >> 3));
* filter_coefficient += filter_coefficient /
* (3 + motion_magnitude_adjustment);
* filter_coefficient is clamped to 0 ~ 255.
*
* Step 2:
* The adjustment vs. diff curve becomes flat very quick when diff increases.
* This allowed us to use only several levels to approximate the curve without
* changing the filtering algorithm too much.
* The adjustments were further corrected by checking the motion magnitude.
* The levels used are:
* diff adjustment w/o motion correction adjustment w/ motion correction
* [-255, -16] -6 -7
* [-15, -8] -4 -5
* [-7, -4] -3 -4
* [-3, 3] diff diff
* [4, 7] 3 4
* [8, 15] 4 5
* [16, 255] 6 7
*/
int vp8_denoiser_filter_c(unsigned char *mc_running_avg_y, int mc_avg_y_stride,
unsigned char *running_avg_y, int avg_y_stride,
unsigned char *sig, int sig_stride,
unsigned int motion_magnitude,
int increase_denoising)
{
unsigned char *running_avg_y_start = running_avg_y;
unsigned char *sig_start = sig;
int sum_diff_thresh;
int r, c;
int sum_diff = 0;
int adj_val[3] = {3, 4, 6};
int shift_inc1 = 0;
int shift_inc2 = 1;
/* If motion_magnitude is small, making the denoiser more aggressive by
* increasing the adjustment for each level. Add another increment for
* blocks that are labeled for increase denoising. */
if (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD)
{
if (increase_denoising) {
shift_inc1 = 1;
shift_inc2 = 2;
}
adj_val[0] += shift_inc2;
adj_val[1] += shift_inc2;
adj_val[2] += shift_inc2;
}
for (r = 0; r < 16; ++r)
{
for (c = 0; c < 16; ++c)
{
int diff = 0;
int adjustment = 0;
int absdiff = 0;
diff = mc_running_avg_y[c] - sig[c];
absdiff = abs(diff);
// When |diff| <= |3 + shift_inc1|, use pixel value from
// last denoised raw.
if (absdiff <= 3 + shift_inc1)
{
running_avg_y[c] = mc_running_avg_y[c];
sum_diff += diff;
}
else
{
if (absdiff >= 4 && absdiff <= 7)
adjustment = adj_val[0];
else if (absdiff >= 8 && absdiff <= 15)
adjustment = adj_val[1];
else
adjustment = adj_val[2];
if (diff > 0)
{
if ((sig[c] + adjustment) > 255)
running_avg_y[c] = 255;
else
running_avg_y[c] = sig[c] + adjustment;
sum_diff += adjustment;
}
else
{
if ((sig[c] - adjustment) < 0)
running_avg_y[c] = 0;
else
running_avg_y[c] = sig[c] - adjustment;
sum_diff -= adjustment;
}
}
}
/* Update pointers for next iteration. */
sig += sig_stride;
mc_running_avg_y += mc_avg_y_stride;
running_avg_y += avg_y_stride;
}
sum_diff_thresh= SUM_DIFF_THRESHOLD;
if (increase_denoising) sum_diff_thresh = SUM_DIFF_THRESHOLD_HIGH;
if (abs(sum_diff) > sum_diff_thresh) {
// Before returning to copy the block (i.e., apply no denoising), check
// if we can still apply some (weaker) temporal filtering to this block,
// that would otherwise not be denoised at all. Simplest is to apply
// an additional adjustment to running_avg_y to bring it closer to sig.
// The adjustment is capped by a maximum delta, and chosen such that
// in most cases the resulting sum_diff will be within the
// accceptable range given by sum_diff_thresh.
// The delta is set by the excess of absolute pixel diff over threshold.
int delta = ((abs(sum_diff) - sum_diff_thresh) >> 8) + 1;
// Only apply the adjustment for max delta up to 3.
if (delta < 4) {
sig -= sig_stride * 16;
mc_running_avg_y -= mc_avg_y_stride * 16;
running_avg_y -= avg_y_stride * 16;
for (r = 0; r < 16; ++r) {
for (c = 0; c < 16; ++c) {
int diff = mc_running_avg_y[c] - sig[c];
int adjustment = abs(diff);
if (adjustment > delta)
adjustment = delta;
if (diff > 0) {
// Bring denoised signal down.
if (running_avg_y[c] - adjustment < 0)
running_avg_y[c] = 0;
else
running_avg_y[c] = running_avg_y[c] - adjustment;
sum_diff -= adjustment;
} else if (diff < 0) {
// Bring denoised signal up.
if (running_avg_y[c] + adjustment > 255)
running_avg_y[c] = 255;
else
running_avg_y[c] = running_avg_y[c] + adjustment;
sum_diff += adjustment;
}
}
// TODO(marpan): Check here if abs(sum_diff) has gone below the
// threshold sum_diff_thresh, and if so, we can exit the row loop.
sig += sig_stride;
mc_running_avg_y += mc_avg_y_stride;
running_avg_y += avg_y_stride;
}
if (abs(sum_diff) > sum_diff_thresh)
return COPY_BLOCK;
} else {
return COPY_BLOCK;
}
}
vp8_copy_mem16x16(running_avg_y_start, avg_y_stride, sig_start, sig_stride);
return FILTER_BLOCK;
}
int vp8_denoiser_filter_uv_c(unsigned char *mc_running_avg_uv,
int mc_avg_uv_stride,
unsigned char *running_avg_uv,
int avg_uv_stride,
unsigned char *sig,
int sig_stride,
unsigned int motion_magnitude,
int increase_denoising) {
unsigned char *running_avg_uv_start = running_avg_uv;
unsigned char *sig_start = sig;
int sum_diff_thresh;
int r, c;
int sum_diff = 0;
int sum_block = 0;
int adj_val[3] = {3, 4, 6};
int shift_inc1 = 0;
int shift_inc2 = 1;
/* If motion_magnitude is small, making the denoiser more aggressive by
* increasing the adjustment for each level. Add another increment for
* blocks that are labeled for increase denoising. */
if (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD_UV) {
if (increase_denoising) {
shift_inc1 = 1;
shift_inc2 = 2;
}
adj_val[0] += shift_inc2;
adj_val[1] += shift_inc2;
adj_val[2] += shift_inc2;
}
// Avoid denoising color signal if its close to average level.
for (r = 0; r < 8; ++r) {
for (c = 0; c < 8; ++c) {
sum_block += sig[c];
}
sig += sig_stride;
}
if (abs(sum_block - (128 * 8 * 8)) < SUM_DIFF_FROM_AVG_THRESH_UV) {
return COPY_BLOCK;
}
sig -= sig_stride * 8;
for (r = 0; r < 8; ++r) {
for (c = 0; c < 8; ++c) {
int diff = 0;
int adjustment = 0;
int absdiff = 0;
diff = mc_running_avg_uv[c] - sig[c];
absdiff = abs(diff);
// When |diff| <= |3 + shift_inc1|, use pixel value from
// last denoised raw.
if (absdiff <= 3 + shift_inc1) {
running_avg_uv[c] = mc_running_avg_uv[c];
sum_diff += diff;
} else {
if (absdiff >= 4 && absdiff <= 7)
adjustment = adj_val[0];
else if (absdiff >= 8 && absdiff <= 15)
adjustment = adj_val[1];
else
adjustment = adj_val[2];
if (diff > 0) {
if ((sig[c] + adjustment) > 255)
running_avg_uv[c] = 255;
else
running_avg_uv[c] = sig[c] + adjustment;
sum_diff += adjustment;
} else {
if ((sig[c] - adjustment) < 0)
running_avg_uv[c] = 0;
else
running_avg_uv[c] = sig[c] - adjustment;
sum_diff -= adjustment;
}
}
}
/* Update pointers for next iteration. */
sig += sig_stride;
mc_running_avg_uv += mc_avg_uv_stride;
running_avg_uv += avg_uv_stride;
}
sum_diff_thresh= SUM_DIFF_THRESHOLD_UV;
if (increase_denoising) sum_diff_thresh = SUM_DIFF_THRESHOLD_HIGH_UV;
if (abs(sum_diff) > sum_diff_thresh) {
// Before returning to copy the block (i.e., apply no denoising), check
// if we can still apply some (weaker) temporal filtering to this block,
// that would otherwise not be denoised at all. Simplest is to apply
// an additional adjustment to running_avg_y to bring it closer to sig.
// The adjustment is capped by a maximum delta, and chosen such that
// in most cases the resulting sum_diff will be within the
// accceptable range given by sum_diff_thresh.
// The delta is set by the excess of absolute pixel diff over threshold.
int delta = ((abs(sum_diff) - sum_diff_thresh) >> 8) + 1;
// Only apply the adjustment for max delta up to 3.
if (delta < 4) {
sig -= sig_stride * 8;
mc_running_avg_uv -= mc_avg_uv_stride * 8;
running_avg_uv -= avg_uv_stride * 8;
for (r = 0; r < 8; ++r) {
for (c = 0; c < 8; ++c) {
int diff = mc_running_avg_uv[c] - sig[c];
int adjustment = abs(diff);
if (adjustment > delta)
adjustment = delta;
if (diff > 0) {
// Bring denoised signal down.
if (running_avg_uv[c] - adjustment < 0)
running_avg_uv[c] = 0;
else
running_avg_uv[c] = running_avg_uv[c] - adjustment;
sum_diff -= adjustment;
} else if (diff < 0) {
// Bring denoised signal up.
if (running_avg_uv[c] + adjustment > 255)
running_avg_uv[c] = 255;
else
running_avg_uv[c] = running_avg_uv[c] + adjustment;
sum_diff += adjustment;
}
}
// TODO(marpan): Check here if abs(sum_diff) has gone below the
// threshold sum_diff_thresh, and if so, we can exit the row loop.
sig += sig_stride;
mc_running_avg_uv += mc_avg_uv_stride;
running_avg_uv += avg_uv_stride;
}
if (abs(sum_diff) > sum_diff_thresh)
return COPY_BLOCK;
} else {
return COPY_BLOCK;
}
}
vp8_copy_mem8x8(running_avg_uv_start, avg_uv_stride, sig_start,
sig_stride);
return FILTER_BLOCK;
}
int vp8_denoiser_allocate(VP8_DENOISER *denoiser, int width, int height,
int num_mb_rows, int num_mb_cols)
{
int i;
assert(denoiser);
denoiser->num_mb_cols = num_mb_cols;
for (i = 0; i < MAX_REF_FRAMES; i++)
{
denoiser->yv12_running_avg[i].flags = 0;
if (vp8_yv12_alloc_frame_buffer(&(denoiser->yv12_running_avg[i]), width,
height, VP8BORDERINPIXELS)
< 0)
{
vp8_denoiser_free(denoiser);
return 1;
}
vpx_memset(denoiser->yv12_running_avg[i].buffer_alloc, 0,
denoiser->yv12_running_avg[i].frame_size);
}
denoiser->yv12_mc_running_avg.flags = 0;
if (vp8_yv12_alloc_frame_buffer(&(denoiser->yv12_mc_running_avg), width,
height, VP8BORDERINPIXELS) < 0)
{
vp8_denoiser_free(denoiser);
return 1;
}
vpx_memset(denoiser->yv12_mc_running_avg.buffer_alloc, 0,
denoiser->yv12_mc_running_avg.frame_size);
denoiser->denoise_state = vpx_calloc((num_mb_rows * num_mb_cols), 1);
vpx_memset(denoiser->denoise_state, 0, (num_mb_rows * num_mb_cols));
return 0;
}
void vp8_denoiser_free(VP8_DENOISER *denoiser)
{
int i;
assert(denoiser);
for (i = 0; i < MAX_REF_FRAMES ; i++)
{
vp8_yv12_de_alloc_frame_buffer(&denoiser->yv12_running_avg[i]);
}
vp8_yv12_de_alloc_frame_buffer(&denoiser->yv12_mc_running_avg);
vpx_free(denoiser->denoise_state);
}
void vp8_denoiser_denoise_mb(VP8_DENOISER *denoiser,
MACROBLOCK *x,
unsigned int best_sse,
unsigned int zero_mv_sse,
int recon_yoffset,
int recon_uvoffset,
loop_filter_info_n *lfi_n,
int mb_row,
int mb_col,
int block_index,
int uv_denoise)
{
int mv_row;
int mv_col;
unsigned int motion_magnitude2;
unsigned int sse_thresh;
int sse_diff_thresh = 0;
// Spatial loop filter: only applied selectively based on
// temporal filter state of block relative to top/left neighbors.
int apply_spatial_loop_filter = 1;
MV_REFERENCE_FRAME frame = x->best_reference_frame;
MV_REFERENCE_FRAME zero_frame = x->best_zeromv_reference_frame;
enum vp8_denoiser_decision decision = FILTER_BLOCK;
enum vp8_denoiser_decision decision_u = COPY_BLOCK;
enum vp8_denoiser_decision decision_v = COPY_BLOCK;
if (zero_frame)
{
YV12_BUFFER_CONFIG *src = &denoiser->yv12_running_avg[frame];
YV12_BUFFER_CONFIG *dst = &denoiser->yv12_mc_running_avg;
YV12_BUFFER_CONFIG saved_pre,saved_dst;
MB_MODE_INFO saved_mbmi;
MACROBLOCKD *filter_xd = &x->e_mbd;
MB_MODE_INFO *mbmi = &filter_xd->mode_info_context->mbmi;
int sse_diff = 0;
// Bias on zero motion vector sse.
int zero_bias = 95;
zero_mv_sse = (unsigned int)((int64_t)zero_mv_sse * zero_bias / 100);
sse_diff = zero_mv_sse - best_sse;
saved_mbmi = *mbmi;
/* Use the best MV for the compensation. */
mbmi->ref_frame = x->best_reference_frame;
mbmi->mode = x->best_sse_inter_mode;
mbmi->mv = x->best_sse_mv;
mbmi->need_to_clamp_mvs = x->need_to_clamp_best_mvs;
mv_col = x->best_sse_mv.as_mv.col;
mv_row = x->best_sse_mv.as_mv.row;
// Bias to zero_mv if small amount of motion.
// Note sse_diff_thresh is intialized to zero, so this ensures
// we will always choose zero_mv for denoising if
// zero_mv_see <= best_sse (i.e., sse_diff <= 0).
if ((unsigned int)(mv_row * mv_row + mv_col * mv_col)
<= NOISE_MOTION_THRESHOLD)
sse_diff_thresh = (int)SSE_DIFF_THRESHOLD;
if (frame == INTRA_FRAME ||
sse_diff <= sse_diff_thresh)
{
/*
* Handle intra blocks as referring to last frame with zero motion
* and let the absolute pixel difference affect the filter factor.
* Also consider small amount of motion as being random walk due
* to noise, if it doesn't mean that we get a much bigger error.
* Note that any changes to the mode info only affects the
* denoising.
*/
mbmi->ref_frame =
x->best_zeromv_reference_frame;
src = &denoiser->yv12_running_avg[zero_frame];
mbmi->mode = ZEROMV;
mbmi->mv.as_int = 0;
x->best_sse_inter_mode = ZEROMV;
x->best_sse_mv.as_int = 0;
best_sse = zero_mv_sse;
}
saved_pre = filter_xd->pre;
saved_dst = filter_xd->dst;
/* Compensate the running average. */
filter_xd->pre.y_buffer = src->y_buffer + recon_yoffset;
filter_xd->pre.u_buffer = src->u_buffer + recon_uvoffset;
filter_xd->pre.v_buffer = src->v_buffer + recon_uvoffset;
/* Write the compensated running average to the destination buffer. */
filter_xd->dst.y_buffer = dst->y_buffer + recon_yoffset;
filter_xd->dst.u_buffer = dst->u_buffer + recon_uvoffset;
filter_xd->dst.v_buffer = dst->v_buffer + recon_uvoffset;
if (!x->skip)
{
vp8_build_inter_predictors_mb(filter_xd);
}
else
{
vp8_build_inter16x16_predictors_mb(filter_xd,
filter_xd->dst.y_buffer,
filter_xd->dst.u_buffer,
filter_xd->dst.v_buffer,
filter_xd->dst.y_stride,
filter_xd->dst.uv_stride);
}
filter_xd->pre = saved_pre;
filter_xd->dst = saved_dst;
*mbmi = saved_mbmi;
}
mv_row = x->best_sse_mv.as_mv.row;
mv_col = x->best_sse_mv.as_mv.col;
motion_magnitude2 = mv_row * mv_row + mv_col * mv_col;
sse_thresh = SSE_THRESHOLD;
if (x->increase_denoising) sse_thresh = SSE_THRESHOLD_HIGH;
if (best_sse > sse_thresh || motion_magnitude2
> 8 * NOISE_MOTION_THRESHOLD)
{
decision = COPY_BLOCK;
}
if (decision == FILTER_BLOCK)
{
unsigned char *mc_running_avg_y =
denoiser->yv12_mc_running_avg.y_buffer + recon_yoffset;
int mc_avg_y_stride = denoiser->yv12_mc_running_avg.y_stride;
unsigned char *running_avg_y =
denoiser->yv12_running_avg[INTRA_FRAME].y_buffer + recon_yoffset;
int avg_y_stride = denoiser->yv12_running_avg[INTRA_FRAME].y_stride;
/* Filter. */
decision = vp8_denoiser_filter(mc_running_avg_y, mc_avg_y_stride,
running_avg_y, avg_y_stride,
x->thismb, 16, motion_magnitude2,
x->increase_denoising);
denoiser->denoise_state[block_index] = motion_magnitude2 > 0 ?
kFilterNonZeroMV : kFilterZeroMV;
// Only denoise UV for zero motion, and if y channel was denoised.
if (uv_denoise &&
motion_magnitude2 == 0 &&
decision == FILTER_BLOCK) {
unsigned char *mc_running_avg_u =
denoiser->yv12_mc_running_avg.u_buffer + recon_uvoffset;
unsigned char *running_avg_u =
denoiser->yv12_running_avg[INTRA_FRAME].u_buffer + recon_uvoffset;
unsigned char *mc_running_avg_v =
denoiser->yv12_mc_running_avg.v_buffer + recon_uvoffset;
unsigned char *running_avg_v =
denoiser->yv12_running_avg[INTRA_FRAME].v_buffer + recon_uvoffset;
int mc_avg_uv_stride = denoiser->yv12_mc_running_avg.uv_stride;
int avg_uv_stride = denoiser->yv12_running_avg[INTRA_FRAME].uv_stride;
int signal_stride = x->block[16].src_stride;
decision_u =
vp8_denoiser_filter_uv(mc_running_avg_u, mc_avg_uv_stride,
running_avg_u, avg_uv_stride,
x->block[16].src + *x->block[16].base_src,
signal_stride, motion_magnitude2, 0);
decision_v =
vp8_denoiser_filter_uv(mc_running_avg_v, mc_avg_uv_stride,
running_avg_v, avg_uv_stride,
x->block[20].src + *x->block[20].base_src,
signal_stride, motion_magnitude2, 0);
}
}
if (decision == COPY_BLOCK)
{
/* No filtering of this block; it differs too much from the predictor,
* or the motion vector magnitude is considered too big.
*/
vp8_copy_mem16x16(
x->thismb, 16,
denoiser->yv12_running_avg[INTRA_FRAME].y_buffer + recon_yoffset,
denoiser->yv12_running_avg[INTRA_FRAME].y_stride);
denoiser->denoise_state[block_index] = kNoFilter;
}
if (uv_denoise) {
if (decision_u == COPY_BLOCK) {
vp8_copy_mem8x8(
x->block[16].src + *x->block[16].base_src, x->block[16].src_stride,
denoiser->yv12_running_avg[INTRA_FRAME].u_buffer + recon_uvoffset,
denoiser->yv12_running_avg[INTRA_FRAME].uv_stride);
}
if (decision_v == COPY_BLOCK) {
vp8_copy_mem8x8(
x->block[20].src + *x->block[20].base_src, x->block[16].src_stride,
denoiser->yv12_running_avg[INTRA_FRAME].v_buffer + recon_uvoffset,
denoiser->yv12_running_avg[INTRA_FRAME].uv_stride);
}
}
// Option to selectively deblock the denoised signal, for y channel only.
if (apply_spatial_loop_filter) {
loop_filter_info lfi;
int apply_filter_col = 0;
int apply_filter_row = 0;
int apply_filter = 0;
int y_stride = denoiser->yv12_running_avg[INTRA_FRAME].y_stride;
int uv_stride =denoiser->yv12_running_avg[INTRA_FRAME].uv_stride;
// Fix filter level to some nominal value for now.
int filter_level = 32;
int hev_index = lfi_n->hev_thr_lut[INTER_FRAME][filter_level];
lfi.mblim = lfi_n->mblim[filter_level];
lfi.blim = lfi_n->blim[filter_level];
lfi.lim = lfi_n->lim[filter_level];
lfi.hev_thr = lfi_n->hev_thr[hev_index];
// Apply filter if there is a difference in the denoiser filter state
// between the current and left/top block, or if non-zero motion vector
// is used for the motion-compensated filtering.
if (mb_col > 0) {
apply_filter_col = !((denoiser->denoise_state[block_index] ==
denoiser->denoise_state[block_index - 1]) &&
denoiser->denoise_state[block_index] != kFilterNonZeroMV);
if (apply_filter_col) {
// Filter left vertical edge.
apply_filter = 1;
vp8_loop_filter_mbv(
denoiser->yv12_running_avg[INTRA_FRAME].y_buffer + recon_yoffset,
NULL, NULL, y_stride, uv_stride, &lfi);
}
}
if (mb_row > 0) {
apply_filter_row = !((denoiser->denoise_state[block_index] ==
denoiser->denoise_state[block_index - denoiser->num_mb_cols]) &&
denoiser->denoise_state[block_index] != kFilterNonZeroMV);
if (apply_filter_row) {
// Filter top horizontal edge.
apply_filter = 1;
vp8_loop_filter_mbh(
denoiser->yv12_running_avg[INTRA_FRAME].y_buffer + recon_yoffset,
NULL, NULL, y_stride, uv_stride, &lfi);
}
}
if (apply_filter) {
// Update the signal block |x|. Pixel changes are only to top and/or
// left boundary pixels: can we avoid full block copy here.
vp8_copy_mem16x16(
denoiser->yv12_running_avg[INTRA_FRAME].y_buffer + recon_yoffset,
y_stride, x->thismb, 16);
}
}
}