An Adaptive Threshold for the Canny Edge with Actor-Critic Algorithm

We propose a method to automatically select proper values of three thresholds in the Canny edge algorithm. Edge detection is widely used for object recognition, detection, and segmentation. Due its good performance, algorithm still among many algorithms. But, it requires manually selecting appropriate given image. Some approaches have been proposed setting they either deal with partial entries or only show their performance limited range variation. In natural scenes, images are acquired under various illumination, pose, weather conditions. This paper proposes that can operate environments. formulate problem by adopting an actor-critic actor critic network solve Also, we suggest reward configuration based on evaluation measure prevent reversal between high low thresholds. The uses original image as input. set negative when reversing occur. adapt unseen environments using without requiring ground truth labels. Experimental results diverse datasets feasibility