Depth-Verified Image Object Labeling for Targeted Enhancement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image enhancement methods suffer from erroneous object classification, leading to unnecessary enhancement of non-relevant image portions and requiring extensive computational resources for post-processing.
Innovation Solution
An image object labeling method using object classification and depth estimation to assign weights to regions, distinguishing between matching and non-matching pixels, thereby reducing erroneous processing and optimizing post-processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional object classification is used for image segmentation, then image enhancement can be performed on segmented objects, but erroneous classification occurs leading to enhancement of regions that should not be enhanced
Solution Approach 1:
The patent introduces depth information as an intermediary verification layer between object classification and image enhancement. The depth estimation module generates depth maps that serve as a mediator to validate whether pixels in segmented regions should truly be enhanced, preventing erroneous classification from causing harmful effects.
Solution Approach 2:
The patent implements a feedback mechanism where depth information is used to verify and correct object classification results. The system compares depth consistency within segmented regions and provides feedback to identify and correct misclassified pixels before enhancement is applied, ensuring higher reliability.
2Manufacturing precision
If post-processing is applied to all segmented regions, then image quality can be enhanced, but computational resources and processing time are excessively consumed
Solution Approach 1:
The patent applies local quality by performing image enhancement only on specific pixels that pass the depth verification test, rather than uniformly processing all segmented regions. This selective approach maintains image quality improvement while significantly reducing unnecessary computational waste on already-processed or misclassified areas.
Solution Approach 2:
The patent uses partial action by applying post-processing only to the extent necessary - specifically to pixels that are both correctly classified and fail the depth consistency test. This avoids excessive processing while ensuring quality improvement where needed.
3Measurement precision
If depth estimation is performed for all pixels, then classification accuracy can be improved, but computational load increases significantly
Solution Approach 1:
The patent segments the image into regions first through object classification, then performs depth estimation only within these segmented regions rather than on the entire image. This hierarchical segmentation approach maintains depth measurement accuracy for classification verification while reducing the overall computational energy required.
Solution Approach 2:
The patent applies partial action by performing depth estimation only on pixels within segmented regions that require verification, rather than uniformly processing all pixels in the image. This selective depth estimation maintains measurement precision where needed while minimizing unnecessary computational energy consumption.
Data Source
AI summary
A method for labeling an image object and a circuit system are provided. In the method, an object classification method is used to segment an image into one or more regions. Each of the regions can be classified into one classification assigned with a classification label. A depth estimation method is used to estimate a depth of each pixel of the image. Whether or not the depth of the pixel matches the classification of the region to which the pixel belongs is determined. When the depth of the pixel matches the classification of the region, a post-processing process is performed on the image based on weights assigned to the regions according to the classification label of the region. Conversely, when the depth of the pixel does not match the classification of the region, the pixel is regarded as noise that does not require the post-processing process.