hi every body! i am working on a video and image processing project using cyclone II EP2C70. i want to implement real time visual tracking. i have set up the video pipeline through which the real time video in flowing. now i want to implement tracking in it. the problem that i am facing now is "how we can normalize the video frame(image) ? " i want to take FFT of each frame and then IFFT. further i want to ask "is it more feasible to normalize the frame(image) 1st and the take FFT or i should take the frame as a whole and take FFT of it.......?" plz guide me in an efficient way.. thank you Rizwan

It depends how you need to implement it. If you can cope using normalisation factors from the previous frame you minimise any video buffering as you can normallise and run the FFT on the fly, using old rescale co-efficients. But if this will upset the FFT, then you will need some serious video buffering.

--- Quote Start --- It depends how you need to implement it. If you can cope using normalisation factors from the previous frame you minimise any video buffering as you can normallise and run the FFT on the fly, using old rescale co-efficients. But if this will upset the FFT, then you will need some serious video buffering. --- Quote End --- thanks tricky for your interest in this regard, in vhdl how we will decide our normalization factor of a frame? can you plz guide me.. thank you Rizwan

In my experience, you dont want to normaise pre-FFT, but post FFT to get the data in the range 0-255. But for any normalisation in the range 0-1 you use the following formula: X = current pixel (X - min) / (Max - Min) its best not to do the divide on the FPGA.

hi Tricky if we are working in RGB formate(in my case each pixel having 3 colors and each color is represented by 8bits.) you mean each color within a pixel having a particular intensity. 1st of all we will check min and max intensity of each color in a frame. on a particular color of a pixel we will perform this operation (X - min) / (Max - Min) thank you Rizwan

are you just trying to convert RGB to greyscale? Or are you trying to include the colour information in the FFT? Ive only ever done operations on greyscale images.

image normalization | Altera Community

12 Replies

Altera_Forum
Honored Contributor
15 years ago
It depends how you need to implement it.
If you can cope using normalisation factors from the previous frame you minimise any video buffering as you can normallise and run the FFT on the fly, using old rescale co-efficients.

But if this will upset the FFT, then you will need some serious video buffering.
Altera_Forum
Honored Contributor
15 years ago
--- Quote Start ---
It depends how you need to implement it.
If you can cope using normalisation factors from the previous frame you minimise any video buffering as you can normallise and run the FFT on the fly, using old rescale co-efficients.

But if this will upset the FFT, then you will need some serious video buffering.
--- Quote End ---

thanks tricky for your interest in this regard,

in vhdl how we will decide our normalization factor of a frame? can you plz guide me..

thank you
Rizwan
Altera_Forum
Honored Contributor
15 years ago
In my experience, you dont want to normaise pre-FFT, but post FFT to get the data in the range 0-255. But for any normalisation in the range 0-1 you use the following formula:

X = current pixel

(X - min) / (Max - Min)

its best not to do the divide on the FPGA.
Altera_Forum
Honored Contributor
15 years ago
hi Tricky

if we are working in RGB formate(in my case each pixel having 3 colors and each color is represented by 8bits.) you mean each color within a pixel having a particular intensity.
1st of all we will check min and max intensity of each color in a frame.
on a particular color of a pixel we will perform this operation

(X - min) / (Max - Min)

thank you
Rizwan
Altera_Forum
Honored Contributor
15 years ago
are you just trying to convert RGB to greyscale? Or are you trying to include the colour information in the FFT?

Ive only ever done operations on greyscale images.
Altera_Forum
Honored Contributor
15 years ago
hi Tricky

i have the input from the camera to the gaughter-card in NTSC formate, which convert it in YCbCr and input to the FPGA, after that FPGA convert this video to RGB formate(to show on the VGA output port).

i want to implement image normalization and FFT efficiently, but at the same time it must be simple.

so, i think conversion from RGB to greyscale is more feasible to implement this all.(because it will be simple and at the same time save our hardware resources as well)

after implementing image normalization and FFT convert greyscale to RGB again (to show at the VGA output port).

Thank you
Rizwan
Altera_Forum
Honored Contributor
15 years ago
If you want greyscale you can just take the Y part of the YCbCr signal. Then if all you have is a greyscale output from the FFT you just put the output into each of the RGB channels to make the picture display (in black and white)

What is the application of your FFT?
Altera_Forum
Honored Contributor
15 years ago
Hi Tricky!

thanks for your information.
my FFT application is convolution.
i want tracking by cross correlation algorithm. In this algorithm we have a reference small template image(already store in the FPGA ROM), which we will track on the real time video.
then I will take the FFT of both (template image and each frame on the video) followed by the IFFT.

Thank You
Rizwan
Altera_Forum
Honored Contributor
15 years ago
They you will probably want both images in greyscale.
Altera_Forum
Honored Contributor
15 years ago
Hi Tricky!

yes, if it is so, then it will be more effective.
but if take template image (colored) and rest of the video (greyscale) , even then it will done its job correctly???

Thank you
Rizwan

Forum Discussion

image normalization

12 Replies

Recent Discussions

Is Quartus Prime Pro 22.4 Compatible with Stratix 10 NX Series Device?

Timing analysis - long combinational path

QuartusPro 25.3 Crashed after using the Signal Tap Logic Analyzer

Duplicate_hierarchy_depth / duplicate_register

Automatically added negative node for TDS output doesn't work with Agilex 5