Dictionary based Approach to Edge Detection

TL;DR

This paper introduces a novel dictionary-based, self-learning edge detection method that adapts to image features, outperforming traditional algorithms in accuracy and detail across diverse image types.

Contribution

The paper presents a new self-learning edge detection technique using eigenfilter dictionaries that adapt to image characteristics, eliminating pre/post processing needs.

Findings

Outperforms traditional edge detection algorithms like Sobel and Canny.

Effectively handles noise, blurriness, and illumination variations.

Successfully applied to diverse image types including microscopic cell images.

Abstract

Edge detection is a very essential part of image processing, as quality and accuracy of detection determines the success of further processing. We have developed a new self learning technique for edge detection using dictionary comprised of eigenfilters constructed using features of the input image. The dictionary based method eliminates the need of pre or post processing of the image and accounts for noise, blurriness, class of image and variation of illumination during the detection process itself. Since, this method depends on the characteristics of the image, the new technique can detect edges more accurately and capture greater detail than existing algorithms such as Sobel, Prewitt Laplacian of Gaussian, Canny method etc which use generic filters and operators. We have demonstrated its application on various classes of images such as text, face, barcodes, traffic and cell images. An application of this technique to cell counting in a microscopic image is also presented.