Image Recognition Reliability Analysis for Underexposed Segments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition techniques, such as Semantic Segmentation, face challenges in determining the category of each field in an image with high accuracy due to unsuitable imaging parameters, particularly in fields prone to errors like underexposure.
Innovation Solution
An apparatus that acquires images with different parameters, segments them, calculates feature quantities, determines reliability, and re-captures images with optimized parameters for low-reliability fields to improve category determination accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Illumination intensity
If auto exposure (AE) and auto white balance (AWB) functions are used to generate images with suitable image quality, then image quality is improved, but recognition accuracy is not primarily improved
Solution Approach 1:
The patent applies local quality by performing category recognition for each segmented field in the image with field-specific processing. Different fields (e.g., sky, building, person) receive tailored image processing and noise reduction suitable for their specific recognition requirements, rather than applying uniform processing to the entire image. This enables optimization of recognition accuracy for each local region while maintaining overall image quality.
Solution Approach 2:
The patent divides the image into multiple fields through segmentation before performing category recognition. By segmenting the image into distinct regions (sky, building, person, etc.), the system can apply different imaging parameters and recognition techniques to each field, thereby improving overall recognition accuracy while maintaining image quality.
2Measurement precision
If exposure is readjusted in dark conditions to improve scene determination accuracy, then scene determination accuracy is improved, but suitable imaging parameters are not set for category determination in each field
Solution Approach 1:
The patent applies local quality by performing category recognition for each segmented field in the image with field-specific processing. Different fields (e.g., sky, building, person) receive tailored image processing and noise reduction suitable for their specific recognition requirements, rather than applying uniform processing to the entire image. This enables optimization of recognition accuracy for each local region while maintaining overall image quality.
Solution Approach 2:
The patent divides the image into multiple fields through segmentation before performing category recognition. By segmenting the image into distinct regions (sky, building, person, etc.), the system can apply different imaging parameters and recognition techniques to each field, thereby improving overall recognition accuracy while maintaining image quality.
Data Source
AI summary
An apparatus includes an acquisition unit that acquires a first image based on a first parameter, and a second image based on a second parameter, a segmentation unit that segments each of the first and second images into a plurality of segments, an acquisition unit that acquires feature quantities from each of the plurality of segments formed by segmenting the first and second images, respectively, a calculation unit that calculates a reliability of each of the plurality of segments of the first image based on the feature quantities acquired from the first image, a classification unit that classifies the plurality of segments of the first image into a first field having a relatively high reliability and a second field having a relatively low reliability, and a determination unit that determines categories for the first and second fields based on the feature quantities acquired from the first and second images.