Sign in to follow this  
McGrane

Gaussian Blue

Recommended Posts

Hey, Im trying to get a gaussian blur effect to work on  a video source. I currently have it up and running, but i dont think i understand how its ment to work, all refrences i find use alot of greek symbols which i do not understand. But through some examples and source code i am using this..

float dif = 159;

float gaussianMask[5][5];

gaussianMask[0][0] = 2/dif; gaussianMask[0][1] =  4/dif; gaussianMask[0][2] =  5/dif; gaussianMask[0][3] =  4/dif; gaussianMask[0][4] = 2/dif;	
	gaussianMask[1][0] = 4/dif; gaussianMask[1][1] =  9/dif; gaussianMask[1][2] = 12/dif; gaussianMask[1][3] =  9/dif; gaussianMask[1][4] = 4/dif;	
	gaussianMask[2][0] = 5/dif; gaussianMask[2][1] = 12/dif; gaussianMask[2][2] = 15/dif; gaussianMask[2][3] = 12/dif; gaussianMask[2][4] = 2/dif;	
	gaussianMask[3][0] = 4/dif; gaussianMask[3][1] =  9/dif; gaussianMask[3][2] = 12/dif; gaussianMask[3][3] =  9/dif; gaussianMask[3][4] = 4/dif;	
	gaussianMask[4][0] = 2/dif; gaussianMask[4][1] =  4/dif; gaussianMask[4][2] =  5/dif; gaussianMask[4][3] =  4/dif; gaussianMask[4][4] = 2/dif;	

 

This is a 5x5 mask i am using - i then multiply each value to a each corisponding pixel in a 5x5 area, and then add it to the original pixel

And what i end up with is a nice light show :P

Im not asking for a full explanation of all the theory on this, as i am currently reading up on it anyways, but simply want a summary of what needs to be done, so i know im going in the right direction. Or just some clarification

Thanks for any help ;)

Share this post


Link to post
Share on other sites

Your code looks like it should produce a blur -- you're gathering 25 samples, and then weighting them by different (Gaussian-ish) values that all sum up to 1.0, which makes sense...

 

What is "a nice light show"? Perhaps you should post some more code, such as how you gather the 25 samples?

Share this post


Link to post
Share on other sites

This is a 5x5 mask i am using - i then multiply each value to a each corisponding pixel in a 5x5 area, and then add it to the original pixel

 

This sounds very suspicious. You should be writing to a separate buffer (otherwise you will get behavior depending on the order in which you process pixels) and you should be summing up the weighted values and write the sum to a single pixel

Share this post


Link to post
Share on other sites

Yea, this is the part i had seen while reading up on it and was unsure of, i have to get an average, but was unsure what to do with the average. Am i ment to average all of these results for the 5x5 area, and then apply it to each of pixel in the sample area ?

This is how im obtaining the sample: (excuse the messy code ;) )

for( int loop = 0; loop < m_width*(m_height*3); loop+=5 ) {
	
	data[loop+0] += ( data[loop+0] * gaussianMask[0][0] );
	data[loop+1] += ( data[loop+1] * gaussianMask[0][1] );
	data[loop+2] += ( data[loop+2] * gaussianMask[0][2] );
	data[loop+3] += ( data[loop+3] * gaussianMask[0][3] );
	data[loop+4] += ( data[loop+4] * gaussianMask[0][4] );
	
	data[loop+0+offset] += ( data[loop+0+offset] * gaussianMask[1][0] );
	data[loop+1+offset] += ( data[loop+1+offset] * gaussianMask[1][1] );
	data[loop+2+offset] += ( data[loop+2+offset] * gaussianMask[1][2] );
	data[loop+3+offset] += ( data[loop+3+offset] * gaussianMask[1][3] );
	data[loop+4+offset] += ( data[loop+4+offset] * gaussianMask[1][4] );
	
	data[loop+0+(offset*2)] += ( data[loop+0+(offset*2)] * gaussianMask[2][0] );
	data[loop+1+(offset*2)] += ( data[loop+0+(offset*2)] * gaussianMask[2][1] );
	data[loop+2+(offset*2)] += ( data[loop+0+(offset*2)] * gaussianMask[2][2] );
	data[loop+3+(offset*2)] += ( data[loop+0+(offset*2)] * gaussianMask[2][3] );
	data[loop+4+(offset*2)] += ( data[loop+0+(offset*2)] * gaussianMask[2][4] );

	data[loop+0+(offset*3)] += ( data[loop+0+(offset*3)] * gaussianMask[3][0] );
	data[loop+1+(offset*3)] += ( data[loop+0+(offset*3)] * gaussianMask[3][1] );
	data[loop+2+(offset*3)] += ( data[loop+0+(offset*3)] * gaussianMask[3][2] );
	data[loop+3+(offset*3)] += ( data[loop+0+(offset*3)] * gaussianMask[3][3] );
	data[loop+4+(offset*3)] += ( data[loop+0+(offset*3)] * gaussianMask[3][4] );

	data[loop+0+(offset*4)] += ( data[loop+0+(offset*4)] * gaussianMask[4][0] );
	data[loop+1+(offset*4)] += ( data[loop+0+(offset*4)] * gaussianMask[4][1] );
	data[loop+2+(offset*4)] += ( data[loop+0+(offset*4)] * gaussianMask[4][2] );
	data[loop+3+(offset*4)] += ( data[loop+0+(offset*4)] * gaussianMask[4][3] );
	data[loop+4+(offset*4)] += ( data[loop+0+(offset*4)] * gaussianMask[4][4] );
	}

 

Where data is a unsigned char array of the pixel colors. 

I know now that this isint right, and had tought this was causing my crazy colors because its just increasing the color values really.

Share this post


Link to post
Share on other sites

As l0calh05t indicated, you don't want to modify your original pixel after each operation. You want to calculate the weighted average for each pixel, then write the obtained value into the corresponding pixel location in another buffer. Otherwise, your blurred values end up contributing to the blurred value for another pixel, and you get a cascading effect that screws up the result.

Share this post


Link to post
Share on other sites

Thanks for the help lads !

One follow up question. The character array is stored in pixels .. r,g,b,r,g,b,r,g,b

Is the mask applied to each color or each pixel ? as in.. should it be - r * gaussianMask[0][0] g * gaussianMask[0][1] b *gaussianMask[0][2]

or r * gaussianMask[0][0] g * gaussianMask[0][0] b * gaussianMask[0][0].. r * gaussianMask[0][1] g * gaussianMask[0][1] b * gaussianMask[0][1]

Edited by McGrane

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

Sign in to follow this