how to create median and spatial filter for continuous acquistion

[AnkitG]

Hi all,

I am trying to do some image processing , I want to make an application which will do median filtering and spatial filtering with continous acquisition of images from a gige vision camera. I tried to use nth order vi but I couldnt understand whats the filter x and y should be . images from camera are Mono 8 with height and widht set to 1000. how should i configure the median and spatial filter to do this processing continously with acquisation. I want to have atleast 5-6 frame persec for my filter. attached is my try with median filter  if I change my filter size (x-y) to more then 10 , my labview crashes. 

Please tell me how can I configure the median and spatial filter for this image processing application.

[Klemen]

Hello,

can you post the code for previous version (2013 for example)?

I am guessing you did not increase the border size of the image? This is needed to establish "room" for the convolution kernel...

For example, a 5x5 convolution kernel, requires a border size of 2. For a 9x9 you need a border size of 4. I think the default border value when you create an image is 3, so 7x7 kernel is possible.

Best regards,

K

[AnkitG]

---

@Klemen wrote:

Hello,

can you post the code for previous version (2013 for example)?

I am guessing you did not increase the border size of the image? This is needed to establish "room" for the convolution kernel...

For example, a 5x5 convolution kernel, requires a border size of 2. For a 9x9 you need a border size of 4. I think the default border value when you create an image is 3, so 7x7 kernel is possible.

Best regards,

K

---

Hi Klemen,

Thank you , I saved it for 2013 version. Please take a look

[Klemen]

Hello,

to apply a median filter of n-th order, use the equation:

n = (kernel_size^2-1)/2

So, if your convolution kernel is 9x9 in size, the order should be: n = (81-1)/2 = 40. Also, do not forget to create the proper border size, in the case of 9x9, the border size should be 4.

Also, have you seen IMAQ convolute? The median filter is a non-linear filter, whereas the convolution is a weighted sum of pixels (linear filter).

The median filter is often used to reduce the speckle noise in the images.

What is your goal?

Best regards,

K

[AnkitG]

---

@Klemen wrote:

Hello,

to apply a median filter of n-th order, use the equation:

n = (kernel_size^2-1)/2

So, if your convolution kernel is 9x9 in size, the order should be: n = (81-1)/2 = 40. Also, do not forget to create the proper border size, in the case of 9x9, the border size should be 4.

Also, have you seen IMAQ convolute? The median filter is a non-linear filter, whereas the convolution is a weighted sum of pixels (linear filter).

The median filter is often used to reduce the speckle noise in the images.

What is your goal?

Best regards,

K

---

Hi Klemen,

thank you, the final goal is to do the median/spatial filtering during the acquisition (with 5+ frames per sec). I have Gige vision camera which will do the continuous acquisition and the user will see the unprocessed image and a processed image( filtered, denoised image) without much lag.

basic idea is to present user a continous processed images( noise removed).

ankit