Neural Image Processing with Dynamic Range Maps for Saturation Errors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods using neural networks suffer from decreased estimation accuracy due to luminance saturation or blocked-up shadow areas, which result in false edges and loss of object information, leading to inaccurate feature extraction.
Innovation Solution
Incorporating a map outside the dynamic range, indicating luminance saturation or blocked-up shadow areas, as input data for the neural network to identify and mitigate these issues during the learning phase, thereby enhancing the network's ability to accurately process images with such conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a neural network is used for image recognition, then estimation accuracy is improved, but estimation accuracy decreases when luminance saturation or blocked-up shadow areas occur in the image
Solution Approach 1:
The image processing is divided into multiple stages: first generating a map outside the dynamic range to identify problematic areas, then using this map to guide subsequent processing. This segmentation allows the system to handle different image regions with appropriate methods, preventing luminance saturation areas from degrading overall estimation accuracy.
Solution Approach 2:
Before performing the main estimation task, the system preliminarily generates a map outside the dynamic range to identify luminance saturation and blocked-up shadow areas. This preliminary action allows the system to anticipate and compensate for potential accuracy degradation in subsequent processing steps.
2Measurement precision
If the neural network processes images with luminance saturated areas, then false edges appear at boundaries, but generating additional maps increases processing complexity
Solution Approach 1:
A map outside the dynamic range is generated as an intermediary data structure to identify problematic areas. This intermediary map serves as a guide for subsequent processing, enabling the system to handle luminance saturation areas without directly complicating the main neural network processing pipeline.
Solution Approach 2:
The system changes the parameter representation by creating a map based on signal values compared to thresholds. This parameter transformation converts the complex problem of luminance saturation detection into a simpler threshold-based classification, reducing processing complexity while maintaining accuracy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An image processing method comprising steps of obtaining a first map representing a region outside a dynamic range of an input image based on a signal value in the input image and a threshold of the signal value (S201, S202, S402), and inputting input data including the input image and the first map and executing a recognition task or a regression task (S203, S404).