image processing on GPU

[abdalrahman_ehsan]

dear all,

I am going to design a GPU used in image processing applications
I want to know the basic mathematical operations needed in image processing which will help me in my project
I need a recommended references which are simple and comprehensive

thank you in advance

 

[sreevenkjan]

Basic mathematical operations is filtering of the images.This is used in all applications.You need to say which type of applications would you be designing??

 

[andre_luis]

As a universal rule, before effectively process the image, it is almost mandatory for most applications, perform some enhancement at image in order to optimize further calculations. At this step, as sreevenkjan mentioned, basic filtering is done for eliminate unnecessary information, and also highlight points of interests.

I would say that as a practical reference for programming, you could refer to OpenCV library, which contains a set of examples wrote in C language.



+++

 

[futuredreamer]

Hello sreevenkjan,
If we're going to design a general purpose GPU what kind of mathematical operation that must be included in its GPU. And can you recommend any materials that may help us understanding such mathematics.

thanks

 

[sreevenkjan]

Hello sreevenkjan,
If we're going to design a general purpose GPU what kind of mathematical operation that must be included in its GPU. And can you recommend any materials that may help us understanding such mathematics.

thanks

Hi if you want to know which mathematical operations you use in image processing it is basically signal processing.Since the image signal is a 2D signal,you will be using 2 nested for loops one is to process rows and other is to process columns.

if you have an image of size 512x512 then you have a for loop like this

for row in 1 to 512 loop
for column in 1 to 512 loop
...your function...
end loop
end loop

There is computer vision library for nVidia's CUDA programming.so you can see the following link where you can install the library and then design it for your application

https://docs.opencv.org/modules/gpu/doc/introduction.html?highlight=cuda

 

[abdalrahman_ehsan]

thank you for your reply
I am asking about the basic mathematical operations to display an image using GPU (i.e. vertex shading , ..... ), I want to know the mathematics behind these operations and simple image processing used in most GPU (e.g. filtering according to sreevenkjan )
I am asking because I want to know what operations GPUs are optimized to perform such as DSPs are optimized to perform digital signal processing algorithms such as MA (multiply - addition) while division or other mathematical operations not optimized in it
I just want to know which operations GPUs are optimized for ?
also I have another question :
why we don't perform image processing on DSPs ? aren't DSPs optimized for signal processing and image processing is a subset of signal processing ?

I hope that references for further readings are attached with answers
thank you in advanc

 

[sreevenkjan]

thank you for your reply
I am asking about the basic mathematical operations to display an image using GPU (i.e. vertex shading , ..... ), I want to know the mathematics behind these operations and simple image processing used in most GPU (e.g. filtering according to sreevenkjan )
I am asking because I want to know what operations GPUs are optimized to perform such as DSPs are optimized to perform digital signal processing algorithms such as MA (multiply - addition) while division or other mathematical operations not optimized in it
I just want to know which operations GPUs are optimized for ?
also I have another question :
why we don't perform image processing on DSPs ? aren't DSPs optimized for signal processing and image processing is a subset of signal processing ?

I hope that references for further readings are attached with answers
thank you in advanc

I have not worked on a GPU but I believe this might be the reason.

Images are a huge memory signals..images use more memory..GPU's consists of framebuffer which helps you to store frames of images..images are 2D data and using frame buffer you can access such data.DSP's do not have much memory resources to process such a high data..a normal cmos image sensor gives an image of 750KB. so it would be difficult to access such a huge data.some of the image processing operations require original image to be stored.