Loading [a11y]/accessibility-menu.js
An improved method for reasonable segmentation based on Post-DAE | IEEE Conference Publication | IEEE Xplore

An improved method for reasonable segmentation based on Post-DAE


Abstract:

The shape prior can be added to refine the segmentation edge when the shape of object is regular. Post-DAE based on denoising autoencoder has provided a post-processing m...Show More

Abstract:

The shape prior can be added to refine the segmentation edge when the shape of object is regular. Post-DAE based on denoising autoencoder has provided a post-processing method that uses shape prior. To make the segmentation result more consistent with the real label in the case of small objects, discontinuity in the same object and other adverse factors, this paper adds morphological processing on the basis of Post-DAE. Specifically, we first perform close operation and dilation to enlarge small objects, then input them into Post-DAE, finally we carry out erosion to restore the size of objects. We performed experiments in hole segmentation of renal endoscope images, and compared the prediction results using quantitative evaluation indicators. The mean Dice and mean IOU of the improved method with morphological processing were 0.8412 and 0.7678, which significantly improved the segmentation performance.
Date of Conference: 24-26 March 2023
Date Added to IEEE Xplore: 08 May 2023
ISBN Information:
Conference Location: Wuhan, China

Funding Agency:


References

References is not available for this document.