Adaptive Bilateral Filter for Depth Map Edge Clarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems face issues with determining pixel depths in stereoscopic images, particularly due to occlusion and exceeding search range boundaries, leading to compromised operations like view synthesis and gesture control, and previous methods introduce blurred edges between objects.
Innovation Solution
A method is introduced to generate a second depth map by assigning weights to selected pixels based on value similarity and spatial proximity, using an adaptive bilateral filter to replace invalid pixels with valid ones, ensuring clear edges and accurate depth values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If bilinear interpolation is used to fill invalid depth values, then the depth map can be completed, but edge clarity deteriorates and blurred edges appear between objects
Solution Approach 1:
The patent applies different processing strategies to different regions of the depth map based on local characteristics. Valid depth pixels are preserved with their original values, while invalid pixels are selectively filled only when they fall within homogeneous regions. Edge detection mechanisms prevent filling operations across object boundaries, ensuring that each region is processed according to its local quality requirements.
Solution Approach 2:
The patent changes the parameter of depth value interpolation by using adaptive bilateral filtering instead of standard bilinear interpolation. This filtering method adjusts its parameters dynamically based on local image characteristics, using spatial and range weights to preserve edges while filling invalid regions. The filter adapts its behavior to maintain edge sharpness while completing the depth map.
2Quantity of substance
If depth values are determined for all pixels in stereoscopic images, then complete depth information is achieved, but measurement precision deteriorates due to occlusion and search range boundary issues
Solution Approach 1:
The patent performs preliminary classification of pixels into valid and invalid categories before attempting to fill missing values. By identifying invalid pixels through depth map analysis and edge detection in advance, the system prepares a mask that guides subsequent filling operations. This preliminary action prevents inaccurate depth values from being assigned to pixels that cannot be reliably measured.
Solution Approach 2:
The patent introduces an adaptive bilateral filter as an intermediary mechanism between the invalid pixels and the filling operation. This filter acts as a mediator that selectively transfers depth information from valid neighboring pixels to invalid pixels, using spatial and range weights to ensure that only appropriate depth values are propagated. The intermediary prevents direct copying of depth values that might cross edge boundaries.
3Productivity
If invalid pixels are replaced using simple interpolation methods, then processing speed is maintained, but depth map quality deteriorates with blurred edges
Solution Approach 1:
The patent implements a dynamic filtering approach where the bilateral filter parameters are adjusted based on local image characteristics. The spatial and range weights are computed dynamically for each pixel based on its neighborhood, allowing the processing to adapt to local features such as edges and homogeneous regions. This dynamic behavior maintains processing efficiency while significantly improving depth map quality compared to static interpolation methods.
Data Source
AI summary
A first depth map is generated in response to a stereoscopic image from a camera. The first depth map includes first pixels having valid depths and second pixels having invalid depths. In response to the first depth map, a second depth map is generated for replacing at least some of the second pixels with respective third pixels having valid depths. For generating the second depth map, a particular one of the third pixels is generated for replacing a particular one of the second pixels. For generating the particular third pixel, respective weight(s) is/are assigned to a selected one or more of the first pixels in response to value similarity and spatial proximity between the selected first pixel(s) and the particular second pixel. The particular third pixel is computed in response to the selected first pixel(s) and the weight(s).