An Experimental Comparison of the Trueness of the Blob Detectors
Year: 2026
Type: conference-paper
Abstract: Blob detection is one of the important methods used in the field of image processing to extract meaningful information from images. It is used in many important fields such as medical imaging, object tracking and astronomy. The objective of this study is not to evaluate the key point detection, but to evaluate the actual round blob detection performance of the blob detectors for general purpose computer vision applications. As the first step, the blob detector algorithms that have already been implemented on various platforms are tested using a small but diverse data set to eliminate the worst ones. Then, the remaining contenders are implemented from scratch, in order to adjust their implementation details and parameters, so that, their accuracies score as high as possible on the first dataset. Finally, in order to measure the accuracy and the trueness of these implementations under difficult conditions, they are tested on artificially created datasets that get increasingly more challenging. Blob detectors can show varying accuracies in aspects such as sharpness, contrast, and noise of images. Results show that, if the image quality is high, the simple grayscale blob detector has the highest accuracy. If the image has blur or noise, then the MSER prevails, while the Difference of Gaussians (DoG) has the highest trueness.
Authors: Yağmur Taze, Ibrahim Cem Baykal
Institution: Adana Science and Technology University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7168281925