Image Processing Problems

[adiguna]

Hi all,
I have some problem here, hope anybody can explain this to me. thanks before.
Beside the value inside the kernel, is there any other different between edge detection kernel like sobel, roberts, prewitt ? any advantage one from the other ??

thanks.

 

[leekk8]

In my opinion, it's the best if you can test each edge detection technique. It's application dependable. Different application may produce different outcome using a particular technique.

 

[itsthetimetodisco]

Canny edge detection technique has proved to be quite superior compared to roberts, sobel, prewitt etc.

 

[yok]

Friend, in general, edge detection is a highpass filter, those kernel based methods are performing spatial-highpass filtering, the main different between each method as you listed in the message is,

1. the filter coefficients; image spatial-frequency to pick up intensity variation for edge formation
2. the filter boundary discontinuity; ringing on sub-mage spatial-dimensional discontinuity during filtering (convolution)

For detail of each method comparision, I suggest you can search IEEE papers for further understanding.

Added after 15 minutes:

Oh, forgot this,

3. The spatial-dimension of kernal; a 3X3 and 5X5 kernal has different filtering results

Friend, except spatial domain edge detection, you may try frequency domain edge detection method, the Wavlets edge detection mathods I had used for my project like Haar and DB6 filters are very much effective that spatial filtering methods.

Yes, for sure with many papers had been proven Canny edge detection that captures points of sharp variation in an image by calculating the modulus of its gradient vector has algorithm complexity for implementation. By the way Canny operator is not a kernal based method.

 

[nagacnu]

i am not clearly understand your quastion
i am giving some details from my class notes which may be usefull for you
Mainly two different methods, gradient and Laplacian. The gradient method detects the edges by looking for the maximum and minimum in the first derivative of the image. The Laplacian method searches for zero crossings in the second derivative of the image to find edges. laplacian method more sensitive to noise.so you forget about it
all the three matheds you mentioned was gradient methods as we know
Sobel method, Prewitt method ,Roberts method all these will use their respective approximation to the derivative. They returns edges at those points where the gradient of intensity image is maximum.
Sobel filter is less sensitive to image noise then other two and produce smoothy results due to 2 in the mask
Prewitt method it produce noisy then sobel but simpler to implemnt computationally.
roberts method is very old method and it is not symmetric and it can not detect if edges are multiples of 45' but still people are using it due to its speed and is simple to implement.

The most powerful edge-detection method is the Canny method