Machine-Learned Image Conversion for Cross-Device Visual Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Images captured by different visual inspection devices have varying tints, textures, and resolutions due to differences in camera and lens configurations, light sources, and angles, making them incompatible for use across multiple devices without modification.
Innovation Solution
An image convertor that includes an image storage unit, training data set generation, image conversion model learning, and conversion units to transform images from one device to another, utilizing machine learning and deformation processing to align orientation and size, enabling cross-device compatibility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If images are captured by different visual inspection devices with different configurations, then each device can perform its own inspection, but the images cannot be used across different devices due to differences in tint, texture, resolution, angle, and lighting
Solution Approach 1:
The patent introduces an image conversion model as an intermediary that transforms images captured by one visual inspection device into images compatible with another device's characteristics. This mediator (conversion model) handles the incompatibility between different device configurations, allowing images to be transferred and used across devices while maintaining inspection reliability.
Solution Approach 2:
The patent applies parameter changes by adjusting image characteristics (tint, texture, resolution, lighting conditions) through the conversion model. The model learns the parameter differences between devices and transforms source images to match the target device's parameters, enabling cross-device compatibility while preserving inspection accuracy.
2Measurement precision
If more images of good and defective products are collected for training, then the inspection program accuracy improves, but the complexity of data collection and management increases when dealing with multiple devices
Solution Approach 1:
The patent creates virtual copies of training images from one device that are adapted to match the characteristics of another device. Instead of physically collecting images from multiple devices, the system generates synthetic training data by converting images from a source device to match the target device's characteristics, significantly reducing data collection complexity while maintaining accuracy.
Solution Approach 2:
The patent performs preliminary image conversion to create device-specific training datasets before the actual inspection process. By pre-converting images to match each device's characteristics, the system prepares training data in advance, avoiding the need for complex real-time data management during inspection operations.
3Adaptability or versatility
If image conversion is performed to make images compatible across devices, then cross-device image usability improves, but additional processing steps and computational resources are required
Solution Approach 1:
The patent replaces complex mechanical/image processing operations with a machine learning-based conversion model. Instead of using traditional image processing techniques that would require multiple manual adjustment steps, the system uses a trained neural network model that automatically performs the conversion, reducing processing complexity while improving image portability.
Data Source
AI summary
An image convertor for visual inspection to convert images captured by two visual inspection devices, each of which comprises a main imaging unit that vertically captures images of an inspected object to be subjected to visual inspection, and a plurality of light sources installed between the main imaging unit and the inspected object comprises an image storage unit that stores a first image group and a second image group, a training data set generation unit that generates a pair of images in the first image group and corresponding images in the second image group as a training data set and an image conversion model learning unit that performs machine learning using the generated training data set to generate an image conversion model that converts an image in the first image group into the corresponding image in the second image group.