you need to quantify what you mean by "best" mathematically (or using ground truth), and you can pick a filter accordingly

You can see this link

https://in.mathworks.com/help/images/ref/edge.html

Thank you for your replies dear all. I am working with OpenCv/C++,.I mean by the "best" the filter that will lead to good results comparing with ground truth and that has low complexity

It certainly depends on your further objectives (beyond mere edge detection), but pre-filtering either with a Gauss or a median filter (5 × 5 is often sufficient) should sufficiently remove noise and artefacts of no interest. They are not the least-effort filters, but yield good results.

Regards

Dear Aasma, I have introduced multi-directional Sobel filter for edge detection, giving better performance than conventional Sobel filter. You may  refer to the following paper for the same

Article A Novel Color Image Segmentation Approach Based On K-Means C...

Linear filters are often faster than non-linear filters. Bilinear filter is an example of a nonlinear filter. The best filter will depend on the structures you want to highlight. For example if you want to highlight only horizontal or vertical structures Sobel filter responds well because it eliminates other structures. If you need continuous edges, the Canny filter will respond well and fast. Finally, if you have structures well-defined by geometric figures Haar wavelets (accompanied by Integral Images) is very fast and robust. Best Regards

I think that Canny edge detector works good

The pre-processing (such as noise reduction, contrast enhancement and PCA) can increase the accuracy of final detection. Then you can use the conventional algorithms such as Canny and ...

Thank you all for your replies. Actually, I am working on a new method for edge detection, I want it to be in real time. That is why I am looking for a low complexity filter as a preprocessing operation.

Using Shock filter might be a good choice!

For instance see the following link:

https://github.com/ClaireXie/edgeGuidedSDSP/blob/master/funcs/shock.m

or

http://www.codeforge.com/read/159801/shock.m__html

or

http://www.pudn.com/downloads76/sourcecode/graph/texture_mapping/detail287312.html

Regards

Thank you, dear Omid for your reply, is this filter a predefined one in Matlab?

You're welcome Assma, I don't think so, but you can use the following links:

http://www.pudn.com/downloads76/sourcecode/graph/texture_mapping/detail287312.html

you can take a look at demo files.

AND

https://www.mathworks.com/matlabcentral/fileexchange/29593-shock-filtering-in-matlab

Good luck

Many thanks Omid, I have just implemented it using opencv/c++, it work but it is computationally expensive

I do not understand how the shock filter can be used for edge detection. In my opinion, it seems like a blur filter only. Maybe you can use it to pre-processing. For example, I use Bilateral filter to enhancement or sharpening the image. See in the image the application of the filters on the original image. How to use the Shock Filter to Edge Detection?

That's good, if the result is satisfactory, parallel programming would solve your problem.

You are right Andre, Shock filter is indeed used as a pre-process step and I suggested it as Assma asked a pre-processing method to enhance the edges!

You can try bilateral filtering using permutohedral lattice. There are open implementations even for gpu. Good luck 

http://graphics.stanford.edu/papers/permutohedral/

Thank you very much, dear Cristian Marti, I can not use the parallel programming because it will take more time than sequential programming. As you know the time required to copy data from host to device and from device to host, in my case, is greater than what I get by using sequential program.

hi Assma, sequential implementation of the permutohedral is still very very fast, good luck

Thanks Cristian, then, I will try it on GPU

Dear

As per my research work and projects so far I am satisfied with HMR Technique as

a best Pre-Processing Technique. Please find more about this technique in the following and my research work

https://www.researchgate.net/publication/311454427_An_Improved_Model_for_Analysis_of_Diabetic_Retinopathy_Related_Imagery 

Article An Improved Model for Analysis of Diabetic Retinopathy Related Imagery

Edge Detection For Color Images involves choosing a proper color space for the color arrangement. We have found that RGB color space is not suitable for this. Also, gradient based methods show better performance in terms of less computation complexity. We have recently proposed an approach for color edge detection. The link for the paper is attached herewith.

Article A Novel Approach for Color Image Edge Detection Using Multid...

Review on edge detection :

https://www.omicsonline.org/open-access/methods-of-image-edge-detection-a-review-2332-0796-1000136.pdf