Abstract
Traditional edge detection methods tend to apply a single threshold over the entire image. However, natural images rarely have uniform illumination throughout, thus just a single threshold across the image is insufficient. This paper explores a method to recursively divide an image into regions and provide each region with an optimal threshold. For each region, we have calculated the threshold automatically using Otsu’s binarisation method. The method’s key goal is to reduce the effect of noise present in images, which leads to the elimination of false edges. It does this while also ensuring that true edges present within the image are not lost. We have proved that asymptotic time complexity of the proposed method is O(MNlogℓ) (where ℓ = min{M, N}). We have compared the performance of our method with the Canny edge detection technique. The Canny edge detector is a well known and widely used edge detection technique which outperforms all the classical edge detection techniques. The results show that our method outperforms the Canny edge detection technique. PSNR values for our method are much higher than that of the Canny edge detection algorithm for almost all the images considered from BSD500 benchmark dataset.
Original language | English |
---|---|
Pages (from-to) | 653-670 |
Number of pages | 18 |
Journal | International Journal of Computational Vision and Robotics |
Volume | 11 |
Issue number | 6 |
DOIs | |
Publication status | Published - 2021 |
All Science Journal Classification (ASJC) codes
- Computer Vision and Pattern Recognition
- Computer Science Applications