Find local maximum and minimum of each edge in an edge map (row-wise)

Hi everyone!

I'll start off by saying I'm new to MATLAB, and this is the first time I'm trying an application related to image processing.

What I'm trying to achieve is building an application that measures the quality of an image. I'm following the method proposed by Marziliano et al in their paper "A no-reference perceptual blur metric".

Here, they first convert the original image to a gray-scale image, then use an edge detector to generate the edge map (I'm using the Canny edge detector). The resultant binary map is traversed row-wise, looking for a pixel corresponding to an edge. If found, the local min and max are searched for. Then the edge width is calculated (localMax - localMin) and numOfEdges is incremented by one. Then the search continues to the next pixel, until the final pixel is reached. At this point the totalEdgeWidth is divided by the total numOfEdges. This is the quality measurement.

I have, so far got the edge map using the Canny edge detector. I need to know; 1. How to traverse the binary image pixel-wise and row-wise looking for an edge. 2. Find the local min and max for the edge found.

Appreciate any help. Thanks in advance :)

Answer 1

[Image Analyst]

I'd think you just multiply your edge map by what you get from rangefilt().

rangeImage = rangefilt(grayImage);
out = cannyEdgeMap .* rangeImage;

In out, at each pixel, you'll have the (localMax - localMin). If you need max and min separately, you'll have to do the same thing but with imdilate() and imerode() instead of rangefilt().

[Lahiru]

Thanks for the reply :) Can you please explain how this works?

[Image Analyst]

rangefilt give you the max minus the min, like you wanted. But it gives it everywhere whereas you want it just at the edge locations. So I multiplied by the edge map to zero out everywhere except at the edges.

[Lahiru]

Forgive me if I sound stupid, but what I'm hoping to achieve is finding the local min and max of each and every edge in the image. I don't see how the rangefilt() function can help me achieve this.

Also, the edge map is a binary image right? Will there be such a thing as a spread of the edge (max - min > 0)?

[Image Analyst]

Let's say your image around some particular pixel is this

2 3 8
4 8 9
2 8 9

Let's say you ran rangefilt on that. The min is 2 and the max is 9, so the range is 7. So the center pixel in the output image would be 7. But since you ran rangefilt over the whole image, the other output pixels will also have values, I just showed you how ONE of them was arrived at. So let's say rangefilt output at that location looked like

6 8 9
4 7 3
4 9 8

Notice the 7 at the center, which is the one I showed you how it was calculated. The others were computed with a similar process, just with the window centered on those other pixels and considering other pixels not shown since I was just showing a small chunk of the image.

Now let's say that you have an edge image and at that location it looked like

0 0 1
0 1 0
0 1 0

So to get the range (max-min) at only the edge pixels (the pixels that are 1) you want to multiply

6 8 9         0 0 1        0 0 9
4 7 3    *    0 1 0    =   0 7 0
4 9 8         0 1 0        0 9 0

So now you're getting the range values only where the edges are and nowhere else.

[Kushal Virupakshappa]

I dont think he is asking for the range of local value rather the pixel location where max and min are