[FFmpeg-cvslog] utvideo: proper median prediction for interlaced videos

[Kostya Shishkov]

ffmpeg | branch: master | Kostya Shishkov <kostya.shishkov at gmail.com> | Sun Jan  1 17:44:08 2012 +0100| [490dcda6b6051066ad7d83979d81db5506f9b3eb] | committer: Kostya Shishkov

utvideo: proper median prediction for interlaced videos

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=490dcda6b6051066ad7d83979d81db5506f9b3eb
---

 libavcodec/utvideo.c |  101 ++++++++++++++++++++++++++++++++++++++++++++++---
 1 files changed, 94 insertions(+), 7 deletions(-)

diff --git a/libavcodec/utvideo.c b/libavcodec/utvideo.c
index 4c3b2a1..d105d29 100644
--- a/libavcodec/utvideo.c
+++ b/libavcodec/utvideo.c
@@ -282,6 +282,77 @@ static void restore_median(uint8_t *src, int step, int stride,
     }
 }
 
+/* UtVideo interlaced mode treats every two lines as a single one,
+ * so restoring function should take care of possible padding between
+ * two parts of the same "line".
+ */
+static void restore_median_il(uint8_t *src, int step, int stride,
+                              int width, int height, int slices, int rmode)
+{
+    int i, j, slice;
+    int A, B, C;
+    uint8_t *bsrc;
+    int slice_start, slice_height;
+    const int cmask = ~(rmode ? 3 : 1);
+    const int stride2 = stride << 1;
+
+    for (slice = 0; slice < slices; slice++) {
+        slice_start    = ((slice * height) / slices) & cmask;
+        slice_height   = ((((slice + 1) * height) / slices) & cmask) - slice_start;
+        slice_height >>= 1;
+
+        bsrc = src + slice_start * stride;
+
+        // first line - left neighbour prediction
+        bsrc[0] += 0x80;
+        A = bsrc[0];
+        for (i = step; i < width * step; i += step) {
+            bsrc[i] += A;
+            A = bsrc[i];
+        }
+        for (i = 0; i < width * step; i += step) {
+            bsrc[stride + i] += A;
+            A = bsrc[stride + i];
+        }
+        bsrc += stride2;
+        if (slice_height == 1)
+            continue;
+        // second line - first element has top predition, the rest uses median
+        C = bsrc[-stride2];
+        bsrc[0] += C;
+        A = bsrc[0];
+        for (i = step; i < width * step; i += step) {
+            B = bsrc[i - stride2];
+            bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
+            C = B;
+            A = bsrc[i];
+        }
+        for (i = 0; i < width * step; i += step) {
+            B = bsrc[i - stride];
+            bsrc[stride + i] += mid_pred(A, B, (uint8_t)(A + B - C));
+            C = B;
+            A = bsrc[stride + i];
+        }
+        bsrc += stride2;
+        // the rest of lines use continuous median prediction
+        for (j = 2; j < slice_height; j++) {
+            for (i = 0; i < width * step; i += step) {
+                B = bsrc[i - stride2];
+                bsrc[i] += mid_pred(A, B, (uint8_t)(A + B - C));
+                C = B;
+                A = bsrc[i];
+            }
+            for (i = 0; i < width * step; i += step) {
+                B = bsrc[i - stride];
+                bsrc[i + stride] += mid_pred(A, B, (uint8_t)(A + B - C));
+                C = B;
+                A = bsrc[i + stride];
+            }
+            bsrc += stride2;
+        }
+    }
+}
+
 static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPacket *avpkt)
 {
     const uint8_t *buf = avpkt->data;
@@ -381,10 +452,18 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPac
                                c->frame_pred == PRED_LEFT);
             if (ret)
                 return ret;
-            if (c->frame_pred == PRED_MEDIAN)
-                restore_median(c->pic.data[i], 1, c->pic.linesize[i],
-                               avctx->width >> !!i, avctx->height >> !!i,
-                               c->slices, !i);
+            if (c->frame_pred == PRED_MEDIAN) {
+                if (!c->interlaced) {
+                    restore_median(c->pic.data[i], 1, c->pic.linesize[i],
+                                   avctx->width >> !!i, avctx->height >> !!i,
+                                   c->slices, !i);
+                } else {
+                    restore_median_il(c->pic.data[i], 1, c->pic.linesize[i],
+                                      avctx->width  >> !!i,
+                                      avctx->height >> !!i,
+                                      c->slices, !i);
+                }
+            }
         }
         break;
     case PIX_FMT_YUV422P:
@@ -395,9 +474,17 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPac
                                c->frame_pred == PRED_LEFT);
             if (ret)
                 return ret;
-            if (c->frame_pred == PRED_MEDIAN)
-                restore_median(c->pic.data[i], 1, c->pic.linesize[i],
-                               avctx->width >> !!i, avctx->height, c->slices, 0);
+            if (c->frame_pred == PRED_MEDIAN) {
+                if (!c->interlaced) {
+                    restore_median(c->pic.data[i], 1, c->pic.linesize[i],
+                                   avctx->width >> !!i, avctx->height,
+                                   c->slices, 0);
+                } else {
+                    restore_median_il(c->pic.data[i], 1, c->pic.linesize[i],
+                                      avctx->width >> !!i, avctx->height,
+                                      c->slices, 0);
+                }
+            }
         }
         break;
     }