Image centerline detection method, computer equipment and storage medium

Abstract

The present application relates to an image centerline detection method, computer equipment and a storage medium. The method includes: using multiple layers of the backbone network in the preset midline detection model to perform feature extraction on the image to be detected to obtain multiple feature maps; based on the resolution of each feature map, perform feature fusion in sequence from high to low to obtain each The first fusion feature map corresponding to each feature map; based on the resolution of each feature map, feature fusion is performed in sequence from low to high, and the second fusion feature map corresponding to each feature map is obtained; according to each first fusion feature map and each The second fused feature map determines the midline of the image. Adopting this method can improve the robustness of detection.

Description

technical field [0001] The present application relates to the technical field of image processing, in particular to an image centerline detection method, computer equipment and storage media. Background technique [0002] Symmetry or near-symmetry phenomena are ubiquitous in the world around us and play an important role in pattern perception in humans and animals. Similarly, accurate symmetry detection methods can help many computer vision methods to better perform tasks such as pattern perception, object recognition, and scene understanding. For example, determining the image midline before image segmentation is beneficial to the accuracy of image segmentation. In two-dimensional Euclidean geometric space, there are four common symmetries of objects, which are reflection, rotation, translation and sliding reflection, that is, the symmetry axes are all straight lines, and the detection methods used to detect the above four symmetries have also been It is widely used in fac...

AI Technical Summary

Problems solved by technology

Faced with this situation, many traditional methods based on local feature detection often have poor robustness

Moreover, the existing commonly used symmetry detection methods are mainly traditional methods based on artificially designed features. This method also needs to pre-process the input image to obtain a binary image that can be processed. However, due to the characteristics of the local gray value of the image, the detection The results are easily affected by noise, and it is also difficult to guarantee the robustness of the detection

Images