How can we can we perform accurate edge detection for grayscale images using adaptive edge detection schemes?

In spite of using conventional edge thresholding schemes to check the performance on edges on images with lot of details.

1. How it can perform thresholding to removes noise as well.
2. How it can adaptively threshold for 0-255 at each pixel without converting into binary image?

Answer 1

[Image Analyst]

I don't know what "conventional edge thresholding schemes" you used. I don't know what "performance on edges" means. I assume you did some kind of edge detection and you disagree with the edges that were found. Why?

To answer your specific questions:

1. I don't know what this means. Maybe you mean that the noise is so far away from the image data that you can threshold to find it, like "Salt and Pepper noise". If so, see my attached demos.
2. You can scan the image and get a local window around the current pixel. Then apply whatever algorithm you want to that local window. Use nlfilter() for this. You don't have to create a binary image, though thresholding usually does. You can create whatever kind of image you want with your custom algorithm.

[Ashfaq Ahmed]

I used canny for edge detection and threshold value bit the problem is threshold value is between 0.5--0.7. However I have the gray scale image taken after gradient and the pixel value is 0--1064.This thresholding show that after 0.5 all are edges. Similarly graythresh also gives 0.6 threshold value. My objective is to get threshold value adaptively on the full image. I am using cubic interpolation above edges and bilinear interpolation below edges. I am thinking that if I am able to get adaptive thresholding it's value should be between 0-255. But still trying to get thresholding adaptively. The image is attached with the mail.