Blind quality enhancement method for dynamic point cloud

By employing progressive feature extraction and adaptive feature fusion branches, the problem of handling unknown and multiple distortions in point cloud compression is solved, achieving efficient point cloud quality enhancement.

CN122289407APending Publication Date: 2026-06-26SHANDONG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SHANDONG UNIV
Filing Date
2026-03-11
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Existing point cloud compression methods require prior knowledge of the distortion level, which cannot adapt to unknown distortion in real-world scenarios. Furthermore, processing multiple distortion levels requires significant resources, resulting in low efficiency.

Method used

We employ a progressive feature extraction branch and an adaptive feature fusion branch. The RMC module achieves frame alignment, the TCCA module mines temporal correlations, the NA module enhances local neighborhood modeling, and the QE module predicts distortion levels to guide adaptive feature fusion.

Benefits of technology

It achieves efficient extraction of point cloud features under unknown distortion levels, improving quality enhancement, reducing resource consumption, and increasing processing efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122289407A_ABST
    Figure CN122289407A_ABST
Patent Text Reader

Abstract

This invention relates to a blind quality enhancement method for dynamic point clouds, implemented through a trained blind quality enhancement model. The method includes a progressive feature extraction branch and an adaptive feature fusion branch. In the progressive feature extraction branch, firstly, the input consecutive reconstructed frames are processed by an RMC module to generate virtual reference frames, achieving geometric alignment between the reference frames and the current frame. Subsequently, the aligned frame sequence is input to a TCCA module, where multi-frame feature interaction and fusion are used to fully exploit temporal correlations and generate fused features. Next, the fused features are fed into the progressive feature extraction module to extract feature representations at different distortion levels in a hierarchical manner. In the adaptive feature fusion branch, firstly, the current frame's quality vector is estimated by a QE module. Then, the estimated quality vector is input to the fusion module to guide the adaptive fusion of the extracted hierarchical features. This invention achieves excellent point cloud quality enhancement results.
Need to check novelty before this filing date? Find Prior Art