Image Tone Matching for CG-Trained Target Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition systems face challenges in achieving robust recognition results when using computer-generated (CG) images as learning data, as they differ significantly from actual images, leading to reduced recognition accuracy.
Innovation Solution
An image processing apparatus applies predetermined image processing to actual images to reduce differences in tone characteristics with CG images, enabling robust recognition even when CG images are used as learning data, by converting actual images to have characteristics similar to CG images through CG production processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If CG images are used as learning data to supplement insufficient actual images, then the quantity of learning data is improved, but the difference in image characteristics between learning data and actual captured images worsens recognition accuracy
Solution Approach 1:
The patent applies image processing that changes tone parameters of actual captured images to match CG image characteristics. Specifically, it adjusts brightness, contrast, and color balance parameters to transform actual images into tones similar to CG images, enabling the recognition model trained on CG images to accurately process actual captured images despite their originally different characteristics
Solution Approach 2:
The patent introduces an intermediary image processing step that acts as a bridge between CG images and actual captured images. By applying tone-matching processing to actual images before input to the recognition model, this intermediary step enables compatibility between the two different image types, allowing the model trained on CG images to recognize targets in actual images with high accuracy
2Reliability
If pre-processing such as monochromatic conversion and contrast adjustment is applied to achieve robust recognition, then recognition robustness is improved, but the case where CG images are used as learning data is not considered, limiting adaptability
Solution Approach 1:
The patent creates a universal image processing approach that works for both CG images and actual captured images. The tone-matching processing is designed to be applicable regardless of the image source, making the system adaptable to different image types. The processing unit can apply the same tone adjustment parameters to actual images to make them compatible with CG-trained models, achieving both robustness and adaptability
Data Source
AI summary
An image processing apparatus applies predetermined image processing to an actual image captured by an image capturing device and recognizes a target within the captured image with use of the image which the predetermined image processing has been applied to. The image processing apparatus applies, to tones of the actual image, the predetermined image processing for reducing a difference from tones of a CG image which shows the same scene and which is presented through computer graphics.