Skip to contentSkip to conversation

tinyjpeg: Better luminance quantization table for Pixart JPEG

PATCH In review Jean-Francois Moine 2 participants 2 messages Mar 23, 2012 – Apr 7, 2012 Media (V4L/DVB)

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 tinyjpeg: Better luminance quantization table for Pixart JPEG

An other luminance quantization table gives a better quality to the Pixart images created by the webcams handled by the gspca drivers pac7302 and pac7311 (pixel format 'PJPG').

Tests have been done with 5 different pac7302 webcams. The marker was always 0x44.

lib/libv4lconvert/tinyjpeg.c +8 -8
index e308f63..687e69c 100644
@@ -206,14 +206,14 @@ static const unsigned char val_ac_chrominance[] = {
 };
 
 const unsigned char pixart_quantization[][64] = { {
-		0x07, 0x07, 0x08, 0x0a, 0x09, 0x07, 0x0d, 0x0b,
-		0x0c, 0x0d, 0x11, 0x10, 0x0f, 0x12, 0x17, 0x27,
-		0x1a, 0x18, 0x16, 0x16, 0x18, 0x31, 0x23, 0x25,
-		0x1d, 0x28, 0x3a, 0x33, 0x3d, 0x3c, 0x39, 0x33,
-		0x38, 0x37, 0x40, 0x48, 0x5c, 0x4e, 0x40, 0x44,
-		0x57, 0x45, 0x37, 0x38, 0x50, 0x6d, 0x51, 0x57,
-		0x5f, 0x62, 0x67, 0x68, 0x67, 0x3e, 0x4d, 0x71,
-		0x79, 0x70, 0x64, 0x78, 0x5c, 0x65, 0x67, 0x63,
+		0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x10, 0x10,
+		0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10,
+		0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10,
+		0x10, 0x10, 0x10, 0x10, 0x20, 0x20, 0x20, 0x20,
+		0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20,
+		0x20, 0x20, 0x20, 0x40, 0x40, 0x40, 0x40, 0x40,
+		0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40,
+		0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40,
 	},
 	{
 		0x11, 0x12, 0x12, 0x18, 0x15, 0x18, 0x2f, 0x1a,
Signature · 2 lines
Ken ar c'hentañ	|	      ** Breizh ha Linux atav! **
Jef		|		http://moinejf.free.fr/
#1 lore raw

Hi,

Thanks for the patch, I've pushed it to v4l-utils master and I will cherry-pick it into the stable-0.8 branch after this mail.

I noticed while testing with a pac7302 camera, that under certain circumstances the jpeg decompression still goes wrong. When you point the camera at a high contrast picture in proper daylight, it switches to a higher compression for certain areas of the picture, and these areas become quite "blockey" when this happens the marker changes to a different value. I think we should use a different quant. table when this happens. It would be great if you can reproduce this and find out a way to make the image less blocky in these cases.

Thanks & Regards,

Hans

Quoted 37 lines · Jean-Francois Moine

An other luminance quantization table gives a better quality to the Pixart images created by the webcams handled by the gspca drivers pac7302 and pac7311 (pixel format 'PJPG').

Tests have been done with 5 different pac7302 webcams. The marker was always 0x44.

lib/libv4lconvert/tinyjpeg.c +8 -8
index e308f63..687e69c 100644
@@ -206,14 +206,14 @@ static const unsigned char val_ac_chrominance[] = {
  };
 
  const unsigned char pixart_quantization[][64] = { {
-		0x07, 0x07, 0x08, 0x0a, 0x09, 0x07, 0x0d, 0x0b,
-		0x0c, 0x0d, 0x11, 0x10, 0x0f, 0x12, 0x17, 0x27,
-		0x1a, 0x18, 0x16, 0x16, 0x18, 0x31, 0x23, 0x25,
-		0x1d, 0x28, 0x3a, 0x33, 0x3d, 0x3c, 0x39, 0x33,
-		0x38, 0x37, 0x40, 0x48, 0x5c, 0x4e, 0x40, 0x44,
-		0x57, 0x45, 0x37, 0x38, 0x50, 0x6d, 0x51, 0x57,
-		0x5f, 0x62, 0x67, 0x68, 0x67, 0x3e, 0x4d, 0x71,
-		0x79, 0x70, 0x64, 0x78, 0x5c, 0x65, 0x67, 0x63,
+		0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x10, 0x10,
+		0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10,
+		0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10,
+		0x10, 0x10, 0x10, 0x10, 0x20, 0x20, 0x20, 0x20,
+		0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20,
+		0x20, 0x20, 0x20, 0x40, 0x40, 0x40, 0x40, 0x40,
+		0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40,
+		0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40,
  	},
  	{
  		0x11, 0x12, 0x12, 0x18, 0x15, 0x18, 0x2f, 0x1a,