How do i remove unwanted/false edge pixels ?

I am currently working on detecting edges in an image using fuzzy logic, the resulting image identifies unwanted pixels. How do i go about this? I tried using the bwNoiseReduction function but it wont help. How do i approach this? http://imageshack.us/photo/my-images/833/71582550.jpg/

Answer 1

[Image Analyst]

I never heard of that function. I think you need to change the parameters of your edge detector so you don't get to many spurious edges in the first place.

[Sandesh Puligundla]

I tried changing the parameters,, does not do much.. Shud I try smoothing the edges followed by edge thinning?

[Image Analyst]

Sure, give it a shot.

[Sandesh Puligundla]

How?

[Image Analyst]

Just blur the image before sending it in to your fuzzy logic edge detector. How about

blurredImage = conv2(single(grayImage), ones(15));
edgeImage = myFuzzyLogicEdgeDetector(blurredImage);

Answer 2

[HSO]

Hello, You can use an Median Filter First (Salt and Pepper)with the adquate size to remove any spike noise from the image an then pass by an Gaussian Filter to smooth the image. That depends of course the conditions of the original image.

Regarding the detector, what kind of implementation are you using? Using Sobel? or just an simple Thresh over first derivative? second?, other ? remember that some detector on its implementation of kernel is very sensibility to noise.You can use Canny edge detector based on the search for local directional maxima in the gradient magnitude.

Here are some references that maybe helpful: http://www.mathworks.com/help/toolbox/images/ref/edge.html http://www.mathworks.com/help/toolbox/images/ref/medfilt2.html

Cheers

[Sandesh Puligundla]

Fuzzy rule based edge detection method using standard deviation and gradient values. Apparently the code is supposed to outperform standard techniques by detecting more edges in the image.

[Image Analyst]

Every edge detector has parameters that you can adjust. Like I said, you need to adjust them to get edges that better reflect what you think they should be.