Fix for high bitdepth temporal filter

The 8bit temporal filter has changed to use non-local means.
This fix adds the same change into the high bitdepth code.

Change-Id: I3375b13a7d914fc8aa9eb4aac1d2e4d9b74b782f

diff --git a/vp10/encoder/temporal_filter.c b/vp10/encoder/temporal_filter.c
index d5f896aa1bc6cc5a9912e9fa773037f3bd126943..4dc21222549187c2b30060de8b488553161970b7 100644 (file)
--- a/vp10/encoder/temporal_filter.c
+++ b/vp10/encoder/temporal_filter.c
@@ -205,15 +205,38 @@ void vp10_highbd_temporal_filter_apply_c(uint8_t *frame1_8,
 
   for (i = 0, k = 0; i < block_height; i++) {
     for (j = 0; j < block_width; j++, k++) {
-      int src_byte = frame1[byte];
-      int pixel_value = *frame2++;
-
-      modifier   = src_byte - pixel_value;
-      // This is an integer approximation of:
-      // float coeff = (3.0 * modifer * modifier) / pow(2, strength);
-      // modifier =  (int)roundf(coeff > 16 ? 0 : 16-coeff);
-      modifier *= modifier;
+      int pixel_value = *frame2;
+
+      // non-local mean approach
+      int diff_sse[9] = { 0 };
+      int idx, idy, index = 0;
+
+      for (idy = -1; idy <= 1; ++idy) {
+        for (idx = -1; idx <= 1; ++idx) {
+          int row = i + idy;
+          int col = j + idx;
+
+          if (row >= 0 && row < (int)block_height &&
+              col >= 0 && col < (int)block_width) {
+            int diff = frame1[byte + idy * (int)stride + idx] -
+                frame2[idy * (int)block_width + idx];
+            diff_sse[index] = diff * diff;
+            ++index;
+          }
+        }
+      }
+
+      assert(index > 0);
+
+      modifier = 0;
+      for (idx = 0; idx < 9; ++idx)
+        modifier += diff_sse[idx];
+
       modifier *= 3;
+      modifier /= index;
+
+      ++frame2;
+
       modifier += rounding;
       modifier >>= strength;
 
@@ -405,37 +428,37 @@ static void temporal_filter_iterate_c(VP10_COMP *cpi,
             int adj_strength = strength + 2 * (mbd->bd - 8);
             // Apply the filter (YUV)
             vp10_highbd_temporal_filter_apply(f->y_buffer + mb_y_offset,
-                                             f->y_stride,
-                                             predictor, 16, 16, adj_strength,
-                                             filter_weight,
-                                             accumulator, count);
+                                              f->y_stride,
+                                              predictor, 16, 16, adj_strength,
+                                              filter_weight,
+                                              accumulator, count);
             vp10_highbd_temporal_filter_apply(f->u_buffer + mb_uv_offset,
-                                             f->uv_stride, predictor + 256,
-                                             mb_uv_width, mb_uv_height,
-                                             adj_strength,
-                                             filter_weight, accumulator + 256,
-                                             count + 256);
+                                              f->uv_stride, predictor + 256,
+                                              mb_uv_width, mb_uv_height,
+                                              adj_strength,
+                                              filter_weight, accumulator + 256,
+                                              count + 256);
             vp10_highbd_temporal_filter_apply(f->v_buffer + mb_uv_offset,
-                                             f->uv_stride, predictor + 512,
-                                             mb_uv_width, mb_uv_height,
-                                             adj_strength, filter_weight,
-                                             accumulator + 512, count + 512);
+                                              f->uv_stride, predictor + 512,
+                                              mb_uv_width, mb_uv_height,
+                                              adj_strength, filter_weight,
+                                              accumulator + 512, count + 512);
           } else {
             // Apply the filter (YUV)
-            vp10_temporal_filter_apply(f->y_buffer + mb_y_offset, f->y_stride,
-                                      predictor, 16, 16,
-                                      strength, filter_weight,
-                                      accumulator, count);
-            vp10_temporal_filter_apply(f->u_buffer + mb_uv_offset, f->uv_stride,
-                                      predictor + 256,
-                                      mb_uv_width, mb_uv_height, strength,
-                                      filter_weight, accumulator + 256,
-                                      count + 256);
-            vp10_temporal_filter_apply(f->v_buffer + mb_uv_offset, f->uv_stride,
-                                      predictor + 512,
-                                      mb_uv_width, mb_uv_height, strength,
-                                      filter_weight, accumulator + 512,
-                                      count + 512);
+            vp10_temporal_filter_apply_c(f->y_buffer + mb_y_offset, f->y_stride,
+                                         predictor, 16, 16,
+                                         strength, filter_weight,
+                                         accumulator, count);
+            vp10_temporal_filter_apply_c(f->u_buffer + mb_uv_offset,
+                                         f->uv_stride, predictor + 256,
+                                         mb_uv_width, mb_uv_height, strength,
+                                         filter_weight, accumulator + 256,
+                                         count + 256);
+            vp10_temporal_filter_apply_c(f->v_buffer + mb_uv_offset,
+                                         f->uv_stride, predictor + 512,
+                                         mb_uv_width, mb_uv_height, strength,
+                                         filter_weight, accumulator + 512,
+                                         count + 512);
           }
 #else
           // Apply the filter (YUV)