Panoramic image segmentation method and apparatus, electronic device, and computer readable medium

By combining multi-scale feature extraction and feature enhancement with candidate target detection and background semantic segmentation, the problems of target scale loss and poor robustness of background segmentation in panoramic image segmentation are solved, achieving clearer panoramic image segmentation results and small target detection capabilities.

CN122244827APending Publication Date: 2026-06-19BEIHANG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
BEIHANG UNIV
Filing Date
2026-03-13
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

Existing panoramic image segmentation methods rely on large-scale backbone networks and complex decoding structures, which leads to the loss of details when the target scale is uniformly pooled to a fixed resolution. Furthermore, the segmentation robustness is poor due to changes in the color and texture of the background region, resulting in problems such as blurred boundaries, loss of foreground, and object clipping.

Method used

Multi-scale feature extraction and feature enhancement are employed to generate a multi-scale feature tensor set. Through candidate object detection and background semantic segmentation, target mask fusion is performed by combining conditional convolution kernels to directly generate a mask on the high-resolution feature map, avoiding detection box pooling errors and improving the robustness of background segmentation.

Benefits of technology

It improves the segmentation effect of panoramic images, ensures clear boundaries and good consistency between foreground and background, reduces boundary blurring and foreground loss, enhances the detection capability of small targets, and reduces the false detection rate.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122244827A_ABST
    Figure CN122244827A_ABST
Patent Text Reader

Abstract

This disclosure presents embodiments of a panoramic image segmentation method, apparatus, electronic device, and computer-readable medium. One specific implementation of the method includes: receiving an image to be processed; performing image preprocessing on the image to be processed to obtain a preprocessed image tensor; performing multi-scale feature extraction on the preprocessed image tensor to obtain a feature tensor set; performing feature enhancement processing on the feature tensor set to obtain an enhanced feature tensor set; generating a multi-scale feature tensor set; performing candidate target detection processing on the multi-scale feature tensor set to obtain at least one candidate target information; generating at least one target mask; performing background semantic segmentation prediction fusion processing on the multi-scale feature tensor set to obtain at least one background semantic prediction result; performing mask fusion processing on the at least one background semantic prediction result and the at least one target mask to obtain a panoramic image segmentation result; and generating a panoramic segmented image. This implementation improves the segmentation effect of panoramic images.
Need to check novelty before this filing date? Find Prior Art