diff --git a/drivers/media/video/cx18/cx18-av-core.c b/drivers/media/video/cx18/cx18-av-core.c
index d98010e..33ec269 100644
--- a/drivers/media/video/cx18/cx18-av-core.c
+++ b/drivers/media/video/cx18/cx18-av-core.c
@@ -299,13 +299,13 @@
 	if (std & V4L2_STD_625_50) {
 		/*
 		 * The following relationships of half line counts should hold:
-		 * 625 = vblank656 + vactive + postvactive
+		 * 625 = vblank656 + vactive
 		 * 10 = vblank656 - vblank = vsync pulses + equalization pulses
 		 *
 		 * vblank656: half lines after line 625/mid-313 of blanked video
 		 * vblank:    half lines, after line 5/317, of blanked video
-		 * vactive:   half lines of active video
-		 * postvactive: 5 half lines after the end of active video
+		 * vactive:   half lines of active video +
+		 * 		5 half lines after the end of active video
 		 *
 		 * As far as I can tell:
 		 * vblank656 starts counting from the falling edge of the first
@@ -316,10 +316,21 @@
 		 * For 625 line systems the driver will extract VBI information
 		 * from lines 6-23 and lines 318-335 (but the slicer can only
 		 * handle 17 lines, not the 18 in the vblank region).
+		 * In addition, we need vblank656 and vblank to be one whole
+		 * line longer, to cover line 24 and 336, so the SAV/EAV RP
+		 * codes get generated such that the encoder can actually
+		 * extract line 23 & 335 (WSS).  We'll lose 1 line in each field
+		 * at the top of the screen.
+		 *
+		 * It appears the 5 half lines that happen after active
+		 * video must be included in vactive (579 instead of 574),
+		 * otherwise the colors get badly displayed in various regions
+		 * of the screen.  I guess the chroma comb filter gets confused
+		 * without them (at least when a PVR-350 is the PAL source).
 		 */
-		vblank656 = 46; /* lines  1 -  23  &  313 - 335 */
-		vblank = 36;    /* lines  6 -  23  &  318 - 335 */
-		vactive = 574;  /* lines 24 - 310  &  336 - 622 */
+		vblank656 = 48; /* lines  1 -  24  &  313 - 336 */
+		vblank = 38;    /* lines  6 -  24  &  318 - 336 */
+		vactive = 579;  /* lines 24 - 313  &  337 - 626 */
 
 		/*
 		 * For a 13.5 Mpps clock and 15,625 Hz line rate, a line is
@@ -989,9 +1000,9 @@
 		 * cx18_av_std_setup(), above standard values:
 		 *
 		 * 480 + 1 for 60 Hz systems
-		 * 576 - 2 for 50 Hz systems
+		 * 576 + 3 for 50 Hz systems
 		 */
-		Vlines = pix->height + (is_50Hz ? -2 : 1);
+		Vlines = pix->height + (is_50Hz ? 3 : 1);
 
 		/*
 		 * Invalid height and width scaling requests are:
diff --git a/drivers/media/video/cx18/cx18-av-vbi.c b/drivers/media/video/cx18/cx18-av-vbi.c
index 6401214..a51732b 100644
--- a/drivers/media/video/cx18/cx18-av-vbi.c
+++ b/drivers/media/video/cx18/cx18-av-vbi.c
@@ -256,7 +256,7 @@
 
 	cx18_av_write(cx, 0x43c, 0x16);
 	/* Should match vblank set in cx18_av_std_setup() */
-	cx18_av_write(cx, 0x474, is_pal ? 36 : 26);
+	cx18_av_write(cx, 0x474, is_pal ? 38 : 26);
 	return 0;
 }
 
diff --git a/drivers/media/video/cx18/cx18-streams.c b/drivers/media/video/cx18/cx18-streams.c
index a080039..54d248e 100644
--- a/drivers/media/video/cx18/cx18-streams.c
+++ b/drivers/media/video/cx18/cx18-streams.c
@@ -374,12 +374,11 @@
 		 * For 625/50 systems, according to the VIP 2 & BT.656 std:
 		 * The EAV RP code's Field bit toggles on line 1, a few lines
 		 * after the Vertcal Blank bit has already toggled.
-		 * Tell the encoder to capture 23-1+1=23 lines per field,
-		 * since we want lines 6 through 23.
-		 *
-		 * FIXME - revisit for 625/50 systems
+		 * (We've actually set the digitizer so that the Field bit
+		 * toggles on line 2.) Tell the encoder to capture 23-2+1=22
+		 * lines per field, since we want lines 6 through 23.
 		 */
-		lines = cx->is_60hz ? (21 - 4 + 1) * 2 : (23 - 1 + 1) * 2;
+		lines = cx->is_60hz ? (21 - 4 + 1) * 2 : (23 - 2 + 1) * 2;
 	}
 
 	data[0] = s->handle;