An Improved Edge Detection Method Using Adaptive Threshold

Abstract

Edge detection is an important step for extracting interesting feature information in image processing and computer vision. Although ant colony optimization (ACO) has been improved by using distributed adaptive threshold strategy (DATS), the artificial ants of this approach still easily ignore weak edges with lower edge gradient which results in detecting discontinuous edges of interesting features. To detect more continuous edges of features by using ACO in color and gray scale images, this work proposes an image pre-processing for ACO with DATS. The result of image pre-processing, which is the image after binarization processing by using adaptive threshold generated form Otsu’s method, is taken as input for ACO. The purpose of image pre-processing is to provide salient changes of image gradient that original images couldn’t provide for artificial ants. By quantitative analysis and subjective comparison of images in different sizes and types used as benchmarks for edge detection, our method extracts more continuous edges and provides more accuracy of interesting feature information than original ACO with DATS does. What’s more, our approach detects all positive edge points of ground truth in our experiments.