vpx/vp8/encoder/denoising.c
Yunqing Wang 64075c9b01 Encoder denoiser performance improvement
The denoiser function was modified to reduce the computational
complexity.

1. The denoiser c function modification:
The original implementation calculated pixel's filter_coefficient
based on the pixel value difference between current raw frame and last
denoised raw frame, and stored them in lookup tables. For each pixel c,
find its coefficient using
    filter_coefficient[c] = LUT[abs_diff[c]];
and then apply filtering operation for the pixel.

The denoising filter costed about 12% of encoding time when it was
turned on, and half of the time was spent on finding coefficients in
lookup tables. In order to simplify the process, a short cut was taken.
The pixel adjustments vs. pixel diff value were calculated ahead of time.
    adjustment = filtered_value - current_raw
               = (filter_coefficient * diff + 128) >> 8

The adjustment vs. diff curve becomes flat very quick when diff increases.
This allowed us to use only several levels to get a close approximation
of the curve. Following the denoiser algorithm, the adjustments are
further modified according to how big the motion magnitude is.

2. The sse2 function was rewritten.

This change made denoiser filter function 3x faster, and improved the
encoder performance by 7% ~ 10% with the denoiser on.

Change-Id: I93a4308963b8e80c7307f96ffa8b8c667425bf50
2012-08-31 13:48:13 -07:00

310 lines
11 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 "vpx_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;
/*
* 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(YV12_BUFFER_CONFIG *mc_running_avg,
YV12_BUFFER_CONFIG *running_avg, MACROBLOCK *signal,
unsigned int motion_magnitude, int y_offset,
int uv_offset)
{
unsigned char *sig = signal->thismb;
int sig_stride = 16;
unsigned char *mc_running_avg_y = mc_running_avg->y_buffer + y_offset;
int mc_avg_y_stride = mc_running_avg->y_stride;
unsigned char *running_avg_y = running_avg->y_buffer + y_offset;
int avg_y_stride = running_avg->y_stride;
int r, c, i;
int sum_diff = 0;
int adj_val[3] = {3, 4, 6};
/* If motion_magnitude is small, making the denoiser more aggressive by
* increasing the adjustment for each level. */
if (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD)
{
for (i = 0; i < 3; i++)
adj_val[i] += 1;
}
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| < 4, use pixel value from last denoised raw. */
if (absdiff <= 3)
{
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;
}
if (abs(sum_diff) > SUM_DIFF_THRESHOLD)
return COPY_BLOCK;
vp8_copy_mem16x16(running_avg->y_buffer + y_offset, avg_y_stride,
signal->thismb, sig_stride);
return FILTER_BLOCK;
}
int vp8_denoiser_allocate(VP8_DENOISER *denoiser, int width, int height)
{
int i;
assert(denoiser);
/* don't need one for intra start at 1 */
for (i = 1; 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);
return 0;
}
void vp8_denoiser_free(VP8_DENOISER *denoiser)
{
int i;
assert(denoiser);
/* we don't have one for intra ref frame */
for (i = 1; 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);
}
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)
{
int mv_row;
int mv_col;
unsigned int motion_magnitude2;
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;
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 mv_col;
int mv_row;
int 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;
if (frame == INTRA_FRAME ||
((unsigned int)(mv_row *mv_row + mv_col *mv_col)
<= NOISE_MOTION_THRESHOLD &&
sse_diff < (int)SSE_DIFF_THRESHOLD))
{
/*
* 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;
if (best_sse > SSE_THRESHOLD || motion_magnitude2
> 8 * NOISE_MOTION_THRESHOLD)
{
decision = COPY_BLOCK;
}
if (decision == FILTER_BLOCK)
{
/* Filter. */
decision = vp8_denoiser_filter(&denoiser->yv12_mc_running_avg,
&denoiser->yv12_running_avg[LAST_FRAME],
x,
motion_magnitude2,
recon_yoffset, recon_uvoffset);
}
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[LAST_FRAME].y_buffer + recon_yoffset,
denoiser->yv12_running_avg[LAST_FRAME].y_stride);
}
}