Allow for deblocking temporal-denoised signal.

Allow for an option to selectively apply the deblocking loop filter to the denoised
raw block, based on the denoised state (no-filter, filter with zero motion, or filter with non-zero motion)
of the current block and its upper and left denoised block.
This helps to reduce some blocking artifacts from the motion-compensated denoising.

Change-Id: I0ac4e70076df69a98c5391979e739a2681e24ae6

diff --git a/vp8/encoder/denoising.c b/vp8/encoder/denoising.c
index f26741f..5616bda 100644 (file)
--- a/vp8/encoder/denoising.c
+++ b/vp8/encoder/denoising.c
@@ -191,10 +191,12 @@ int vp8_denoiser_filter_c(unsigned char *mc_running_avg_y, int mc_avg_y_stride,
     return FILTER_BLOCK;
 }
 
-int vp8_denoiser_allocate(VP8_DENOISER *denoiser, int width, int height)
+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++)
     {
@@ -222,6 +224,10 @@ int vp8_denoiser_allocate(VP8_DENOISER *denoiser, int width, int height)
 
     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;
 }
 
@@ -243,13 +249,20 @@ void vp8_denoiser_denoise_mb(VP8_DENOISER *denoiser,
                              unsigned int best_sse,
                              unsigned int zero_mv_sse,
                              int recon_yoffset,
-                             int recon_uvoffset)
+                             int recon_uvoffset,
+                             loop_filter_info_n *lfi_n,
+                             int mb_row,
+                             int mb_col,
+                             int block_index)
 {
     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;
 
@@ -362,6 +375,8 @@ void vp8_denoiser_denoise_mb(VP8_DENOISER *denoiser,
                                          running_avg_y, avg_y_stride,
                                          x->thismb, 16, motion_magnitude2,
                                          x->increase_denoising);
+        denoiser->denoise_state[block_index] = motion_magnitude2 > 0 ?
+            kFilterNonZeroMV : kFilterZeroMV;
     }
     if (decision == COPY_BLOCK)
     {
@@ -372,5 +387,59 @@ void vp8_denoiser_denoise_mb(VP8_DENOISER *denoiser,
                 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;
+    }
+    // Option to selectively deblock the denoised signal.
+    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);
+      }
     }
 }

diff --git a/vp8/encoder/denoising.h b/vp8/encoder/denoising.h
index ae744d2..6db0785 100644 (file)
--- a/vp8/encoder/denoising.h
+++ b/vp8/encoder/denoising.h
@@ -12,6 +12,7 @@
 #define VP8_ENCODER_DENOISING_H_
 
 #include "block.h"
+#include "vp8/common/loopfilter.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -27,13 +28,22 @@ enum vp8_denoiser_decision
   FILTER_BLOCK
 };
 
+enum vp8_denoiser_filter_state {
+  kNoFilter,
+  kFilterZeroMV,
+  kFilterNonZeroMV
+};
+
 typedef struct vp8_denoiser
 {
     YV12_BUFFER_CONFIG yv12_running_avg[MAX_REF_FRAMES];
     YV12_BUFFER_CONFIG yv12_mc_running_avg;
+    unsigned char* denoise_state;
+    int num_mb_cols;
 } VP8_DENOISER;
 
-int vp8_denoiser_allocate(VP8_DENOISER *denoiser, int width, int height);
+int vp8_denoiser_allocate(VP8_DENOISER *denoiser, int width, int height,
+                          int num_mb_rows, int num_mb_cols);
 
 void vp8_denoiser_free(VP8_DENOISER *denoiser);
 
@@ -42,7 +52,11 @@ void vp8_denoiser_denoise_mb(VP8_DENOISER *denoiser,
                              unsigned int best_sse,
                              unsigned int zero_mv_sse,
                              int recon_yoffset,
-                             int recon_uvoffset);
+                             int recon_uvoffset,
+                             loop_filter_info_n *lfi_n,
+                             int mb_row,
+                             int mb_col,
+                             int block_index);
 
 #ifdef __cplusplus
 }  // extern "C"

diff --git a/vp8/encoder/encodeframe.c b/vp8/encoder/encodeframe.c
index b550f6b..e6b0f9b 100644 (file)
--- a/vp8/encoder/encodeframe.c
+++ b/vp8/encoder/encodeframe.c
@@ -1246,7 +1246,7 @@ int vp8cx_encode_inter_macroblock
             x->zbin_mode_boost_enabled = 0;
         }
         vp8_rd_pick_inter_mode(cpi, x, recon_yoffset, recon_uvoffset, &rate,
-                               &distortion, &intra_error);
+                               &distortion, &intra_error, mb_row, mb_col);
 
         /* switch back to the regular quantizer for the encode */
         if (cpi->sf.improved_quant)

diff --git a/vp8/encoder/onyx_if.c b/vp8/encoder/onyx_if.c
index 4d16a5f..09854a5 100644 (file)
--- a/vp8/encoder/onyx_if.c
+++ b/vp8/encoder/onyx_if.c
@@ -1751,7 +1751,8 @@ void vp8_change_config(VP8_COMP *cpi, VP8_CONFIG *oxcf)
       {
         int width = (cpi->oxcf.Width + 15) & ~15;
         int height = (cpi->oxcf.Height + 15) & ~15;
-        vp8_denoiser_allocate(&cpi->denoiser, width, height);
+        vp8_denoiser_allocate(&cpi->denoiser, width, height,
+                              cpi->common.mb_rows, cpi->common.mb_cols);
       }
     }
 #endif

diff --git a/vp8/encoder/pickinter.c b/vp8/encoder/pickinter.c
index 817c9ef..3a78ee9 100644 (file)
--- a/vp8/encoder/pickinter.c
+++ b/vp8/encoder/pickinter.c
@@ -1168,6 +1168,7 @@ void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset,
 #if CONFIG_TEMPORAL_DENOISING
     if (cpi->oxcf.noise_sensitivity)
     {
+        int block_index = mb_row * cpi->common.mb_cols + mb_col;
         if (x->best_sse_inter_mode == DC_PRED)
         {
             /* No best MV found. */
@@ -1179,7 +1180,9 @@ void vp8_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset,
         }
         x->increase_denoising = 0;
         vp8_denoiser_denoise_mb(&cpi->denoiser, x, best_sse, zero_mv_sse,
-                                recon_yoffset, recon_uvoffset);
+                                recon_yoffset, recon_uvoffset,
+                                &cpi->common.lf_info, mb_row, mb_col,
+                                block_index);
 
 
         /* Reevaluate ZEROMV after denoising. */

diff --git a/vp8/encoder/rdopt.c b/vp8/encoder/rdopt.c
index f145d09..4465b5e 100644 (file)
--- a/vp8/encoder/rdopt.c
+++ b/vp8/encoder/rdopt.c
@@ -1935,7 +1935,8 @@ static void update_best_mode(BEST_MODE* best_mode, int this_rd,
 
 void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset,
                             int recon_uvoffset, int *returnrate,
-                            int *returndistortion, int *returnintra)
+                            int *returndistortion, int *returnintra,
+                            int mb_row, int mb_col)
 {
     BLOCK *b = &x->block[0];
     BLOCKD *d = &x->e_mbd.block[0];
@@ -2510,6 +2511,7 @@ void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset,
 #if CONFIG_TEMPORAL_DENOISING
     if (cpi->oxcf.noise_sensitivity)
     {
+        int block_index = mb_row * cpi->common.mb_cols + mb_col;
         if (x->best_sse_inter_mode == DC_PRED)
         {
             /* No best MV found. */
@@ -2520,7 +2522,9 @@ void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset,
             best_sse = best_rd_sse;
         }
         vp8_denoiser_denoise_mb(&cpi->denoiser, x, best_sse, zero_mv_sse,
-                                recon_yoffset, recon_uvoffset);
+                                recon_yoffset, recon_uvoffset,
+                                &cpi->common.lf_info, mb_row, mb_col,
+                                block_index);
 
 
         /* Reevaluate ZEROMV after denoising. */

diff --git a/vp8/encoder/rdopt.h b/vp8/encoder/rdopt.h
index fe21b8e..e0da35e 100644 (file)
--- a/vp8/encoder/rdopt.h
+++ b/vp8/encoder/rdopt.h
@@ -70,7 +70,10 @@ static void insertsortsad(int arr[],int idx[], int len)
 }
 
 extern void vp8_initialize_rd_consts(VP8_COMP *cpi, MACROBLOCK *x, int Qvalue);
-extern void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int recon_uvoffset, int *returnrate, int *returndistortion, int *returnintra);
+extern void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x,
+                                   int recon_yoffset, int recon_uvoffset,
+                                   int *returnrate, int *returndistortion,
+                                   int *returnintra, int mb_row, int mb_col);
 extern void vp8_rd_pick_intra_mode(MACROBLOCK *x, int *rate);