Selection of an Automated Morphological Gradient Threshold for Image Segmentation. Application to Vision-Based Path Planning

摘要

Segmentation is an essential part of practically any automated image recognition system, since it is necessary for further processing such as feature extraction or object recognition. There exist a variety of techniques for threshold selection, as it is a fast and robust method. Threshold value will have considerable effects on the boundary position and overall size of the extracted objects. In this work, we propose an automated thresholding selection, which considers the local properties of a pixel. To do this, the algorithm calculates the morphological gradient and Laplacian and, afterwards, chooses a suitable threshold after estimating the lowest distance between the ideal segmentation and the morphological gradient thresholding segmentation. As an application of this segmentation process, we have implemented a path planning algorithm for a vision-based system. The experiments show that our path planning algorithm is able to find good solution paths after a training process.