dm6437 YUV to RGB, noise

zhiqiang li

Hello, everyone,

I am now doing video processing using DM6437.

Now I just want to convert from YUV ( 4:2:2 ) to RGB, but unfortunately , I encountered a problem, the converted result is bellow:

Obviously, there exists a lot of noise.

The code I used are as follows( marked in red ):

First,I extract the YUV components, and put them in the arrays below,

unsigned char YArray[720*480] = {0};
unsigned char CrArray[720*240] = {0};
unsigned char CbArray[720*240] = {0};

then, I invoke the function below,

PIXEL_COLOR *YUV2RGB(unsigned char *Y,unsigned char *Cr,unsigned char *Cb,PIXEL_COLOR *RGB,int width,int height)
{
int i = 0;
int j = 0;
int k = 0;
for( i = 0; i< height;i++)
{
  k = i*width;
  for(j = 0;j < width;j++)
  {

   RGB[k+j].r = (298*(Y[k+j]-16) + 409*(Cr[ (k+j) / 2] - 128) + 128) >> 8;
   RGB[k+j].g = (298*(Y[k+j]-16) - 208*(Cr[(k+j) / 2] - 128) - 100*(Cb[(k+j) / 2] - 128) + 128) >> 8;
   RGB[k+j].b = (298*(Y[k+j]-16) + 516*(Cb[(k+j) / 2] - 128) + 128) >> 8;

   if(RGB[k+j].r > 255)
   {
    RGB[k+j].r = 255;
   }
   else if(RGB[k+j].r < 0)
   {
    RGB[k+j].r = 0;
   }
   if(RGB[k+j].g > 255)
   {
    RGB[k+j].g = 255;
   }
   else if(RGB[k+j].g < 0)
   {
    RGB[k+j].g = 0;
   }
   if(RGB[k+j].b > 255)
   {
    RGB[k+j].b = 255;
   }
   else if(RGB[k+j].b < 0)
   {
    RGB[k+j].b = 0;
   }
  }
}
return RGB;
}

finally, I show the image using ccs, the detailed steps are as follows:

View -> Graph -> Image...

Could anyone help me?

Thank you!

Zhiqiang

over 14 years ago

0 Paul.Yin over 14 years ago

TI__Genius 14405 points

Hi Zhiqiang,

I am not sure where you got the coefficient

   RGB[k+j].r = (298*(Y[k+j]-16) + 409*(Cr[ (k+j) / 2] - 128) + 128) >> 8;
   RGB[k+j].g = (298*(Y[k+j]-16) - 208*(Cr[(k+j) / 2] - 128) - 100*(Cb[(k+j) / 2] - 128) + 128) >> 8;
   RGB[k+j].b = (298*(Y[k+j]-16) + 516*(Cb[(k+j) / 2] - 128) + 128) >> 8;

Would you be able to try

.
R = (1.00000 ´ Y) + (0.00000 ´ (Cb – 128)) + (1.40200 ´ (Cr – 128))
G = (1.00000 ´ Y) – (0.34414 ´ (Cb – 128)) – (0.71444 ´ (Cr – 128))
B = (1.00000 ´ Y) + (1.72200 ´ (Cb – 128)) + (0.00000 ´ (Cr – 128))

R = (1.00000 ´ Y) + (0.00000 ´ (Cb – 128)) + (1.37110 ´ (Cr – 128))
G = (1.00000 ´ Y) – (0.33690 ´ (Cb – 128)) – (0.69820 ´ (Cr – 128))
B = (1.00000 ´ Y) + (1.73240 ´ (Cb – 128)) + (0.00000 ´ (Cr – 128))

0 Steve Clynes over 14 years ago

TI__Mastermind 43875 points

In addition to Paul's comments...

This looks like you have overflow occurring somewhere.

What are the structure types for "PIXEL_COLOR" ?If these are not int then you will get overflow and the comparison against ">255" will never be true.

Additionally I would look at the precedence of the calculation step. Sometimes there are implicit type conversions which may be causing issues.

What you are seeing is not actually noise, but roll-over on some of the color channels. This type of 'noise' appears when you have shaded regions which have color components which are close to either maximum or minimum values and you have not correctly handled roll over and calculation ranges.

What are the intermediate values of "RGB[k+j].r" just after the calculation?

Try putting a break point inside the range checking code to make sure that it does get there at some time.

   if(RGB[k+j].r > 255)
   {
    RGB[k+j].r = 255; //Set break point here
   }
   else if(RGB[k+j].r < 0)
   {
    RGB[k+j].r = 0; //Set break point here
   }

BR,

Steve

0 zhiqiang li over 14 years ago in reply to Steve Clynes

Intellectual 270 points

Hi Steve and Paul.Yin

I am sure that the roll-over has happened.

Now I have solved the problem.

Thank you!

Zhiqiang

Processors

Processors forum

dm6437 YUV to RGB, noise