3D Model Color Smoothing via Anisotropic Bilateral Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current 3D scanning technologies produce low-resolution images with noise due to real-time spatial alignment of color and depth information, leading to smudges in 3D printed models, and existing noise reduction techniques are either highly manual or compromise details like edges.
Innovation Solution
Applying an anisotropic box kernel variant of a bilateral filter to each vertex of a 3D mesh model, conditionally adjusting colors based on neighboring vertices, while preserving edges by rejecting color changes in areas with significant normal differences or high-frequency details.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If real-time spatial alignment of color and depth information is performed to enable 3D scanning, then 3D models can be generated quickly, but image resolution becomes low and noise increases
Solution Approach 1:
The patent applies bilateral filtering as a post-processing step after the 3D model is generated from real-time scanning. This preliminary action during post-processing compensates for the low resolution and noise introduced by real-time processing, allowing the system to maintain fast scanning speeds while improving final image quality.
2Measurement precision
If noise reduction techniques are applied to remove noise from image data, then image quality improves, but edge details and high-frequency information are compromised
Solution Approach 1:
The bilateral filter applies different smoothing strengths to different regions of the image based on local characteristics. In homogeneous regions, strong smoothing removes noise effectively, while in edge regions, the filter detects the discontinuity and applies minimal smoothing to preserve edge details. This local adaptation resolves the contradiction between noise reduction and edge preservation.
Solution Approach 2:
The filter dynamically adjusts its parameters (spatial kernel size and color similarity threshold) based on local image characteristics. In smooth regions, larger kernels and higher thresholds enable aggressive noise reduction, while near edges, smaller kernels and lower thresholds preserve detail. This parameter adaptation allows the system to achieve noise reduction without losing edge information.
3Loss of information
If manual techniques are used to remove noise from image data, then edge details are preserved, but the process becomes highly manual and time-consuming
Solution Approach 1:
The bilateral filter is an automated algorithm that autonomously analyzes image characteristics and applies appropriate smoothing without human intervention. It self-adjusts its behavior based on local image properties, performing the work that would otherwise require manual editing. This automation eliminates the time-consuming manual process while maintaining edge detail preservation through its intelligent local adaptation.
Data Source
AI summary
Techniques are disclosed for noise reduction and color smoothing of scanned 3D models for 3D printing. An anisotropic box kernel variant of a bilateral filter is applied to each vertex of a 3D mesh model, where the vertices contain color information. With bilateral filtering, the color at a given vertex is, under certain conditions, replaced by a Gaussian weighted average of colors of neighboring vertices. The weighting is a function of a geometric distance between vertices. The direction of traversal is limited towards neighboring vertices where a color mismatch is a color difference larger than the threshold value. The color of a neighboring vertex is rejected if the vertex normals differ from by a large angle. The weighted average of colors of the neighboring vertices is also rejected if the given vertex lies in a region where the color match/mismatch ratio is low at the end of the traversal.