Depth Map Filtering for 3D Image Boundary Flicker Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing depth estimation methods using deep learning for generating 3D images from 2D images face limitations in accurately detecting subtle depth changes, leading to degraded 3D effects due to residual images at object boundaries and flicker between frames.
Innovation Solution
An image processing apparatus applies different weights to boundary and non-boundary regions during filtering of depth maps, using first and second filtering techniques to enhance the 3D effect by reducing residual images and reflecting subtle depth changes within objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If deep learning-based single image depth estimation is used to generate depth maps, then the method can generate depth maps using only a single camera image, but the accuracy is lower and subtle depth changes cannot be detected
Solution Approach 1:
The patent segments the depth map into boundary regions and non-boundary regions, applying different filtering strategies to each. Boundary regions use one type of filtering while non-boundary regions use another, allowing precise handling of object edges while maintaining efficiency across the entire image.
Solution Approach 2:
The patent applies local quality by using region-specific filtering weights where boundary regions receive different treatment compared to non-boundary regions. This ensures that subtle depth changes at object boundaries are preserved while maintaining overall depth map quality.
2Productivity
If uniform filtering is applied to the entire depth map, then the processing is simple and fast, but residual images appear at object boundaries and flicker occurs between frames
Solution Approach 1:
The patent divides the depth map into boundary and non-boundary regions, applying different filtering operations to each segment. This segmentation allows the system to maintain fast processing while eliminating residual images at boundaries through region-specific filtering.
Solution Approach 2:
The patent implements local quality by assigning different filtering weights to boundary regions versus non-boundary regions. This local differentiation eliminates flicker and residual images at boundaries while maintaining efficient processing across the entire depth map.
3Adaptability or versatility
If deep learning depth estimation is used, then a depth map can be generated from a single 2D image, but the 3D effect is degraded due to inaccurate depth changes
Solution Approach 1:
The patent applies local quality by using different filtering weights for boundary and non-boundary regions, which preserves subtle depth changes at object boundaries. This improves the overall 3D image quality while maintaining the adaptability of single-image depth estimation.
Data Source
AI summary
An image processing apparatus includes: at least one processor including processing circuitry; and memory including one or more storage media storing one or more instructions, where the at least one processor is configured to, individually or collectively, execute the one or more instructions to cause the image processing apparatus to: obtain an input depth map from a two-dimensional (2D) input image, the input depth map including a boundary region and a non-boundary region of an object, perform first filtering on a first frame and a previous frame of the input depth map to obtain a first filtered depth map, by applying different weights to the boundary region and the non-boundary region, and generate a three-dimensional (3D) image, based on the first filtered depth map and the 2D input image.