An edge detection system and method based on multi-granularity attention layered network

Abstract

The invention discloses an edge detection system and method based on a multi-granularity attention hierarchical network, the system includes multiple multi-granularity feature layering modules for capturing high-level features and multiple channel attention for fusing low-level features module and spatial attention module, and a feature map fusion module for merging feature maps; the present invention uses a combination of channel attention module, spatial attention module and multi-granularity feature layering module to solve the problem that the deep neural network predicts that the boundary is too thick It can extract clear edges from natural images without manually extracting edge features, and achieve good visual effects, improving the performance of edge detection evaluation indicators.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an edge detection system and method based on a multi-granularity attention layered network. Background technique [0002] The exponential growth of data is an important feature of the Internet era, and image data is an efficient source for human beings to obtain information from the objective world. Image edge detection is one of the fundamental tasks in image processing and computer vision, especially a research field in feature extraction. Its goal is to obtain a collection of pixels with sharp brightness changes from natural images. Edge is one of the most basic and important basic features of images. It is widely used in digital images such as motion detection, image segmentation, pattern recognition, and face recognition. technology field. The traditional edge detection method focuses on the texture gradient of the image, and the amount of calculation is small, bu...

AI Technical Summary

Problems solved by technology

Although CNN-based methods are good at producing semantic contours without manually extracting edge features, the predicted edges are relatively thick compared to expert annotated maps

Images