Feature-Map Image Processing for Fast Accurate Object Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods using machine learning for segmentation and identification are computationally expensive and require high-resolution images, making them unsuitable for devices with low real-time processing capabilities, especially in applications like microscopic image analysis where accurate segmentation and identification of cells or bacteria is challenging.
Innovation Solution
An image processing device and method that generates and processes images based on feature extraction maps to selectively extract relevant features, reducing computational load while maintaining high accuracy through controlled image generation and feature extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning is used for segmentation and identification, then identification accuracy is improved, but processing cost increases
Solution Approach 1:
The image processing is segmented into multiple stages: low-resolution initial processing to identify candidate regions, followed by high-resolution processing only for those candidate regions. This segmentation allows machine learning to be applied selectively rather than to the entire image, reducing overall processing cost while maintaining identification accuracy for critical objects.
Solution Approach 2:
Different processing qualities are applied to different regions of the image. Low-resolution processing is applied to the entire image for initial screening, while high-resolution processing is applied locally only to candidate regions identified in the first stage. This local quality approach reduces computational cost while preserving necessary detail where needed.
2Measurement precision
If high resolution image is used for proper segmentation and identification, then identification accuracy is improved, but processing cost increases
Solution Approach 1:
The processing is divided into two resolution stages. The high-resolution image is processed only for candidate regions identified from the low-resolution image, rather than processing the entire high-resolution image. This segmentation of processing by region and resolution reduces computational cost while maintaining segmentation accuracy for objects of interest.
3Productivity
If low resolution image is used for segmentation, then processing cost is reduced, but identification accuracy deteriorates
Solution Approach 1:
Low-resolution processing is performed as a preliminary action to identify candidate regions before applying high-resolution processing. This preliminary screening at low resolution reduces the amount of data requiring expensive high-resolution processing, improving overall processing speed while maintaining accuracy through the subsequent high-resolution analysis of candidates.
Solution Approach 2:
The low-resolution image processing acts as an intermediary step that identifies candidate regions, which then serve as input for the high-resolution processing stage. This intermediary processing at low resolution filters out non-candidate regions, reducing the computational burden of high-resolution processing while ensuring accurate identification of objects of interest.
Data Source
AI summary
An image processing device includes: an image reception unit for receiving an input image; an image generation unit for generating an image for extracting a feature from the input image; a feature extraction unit for extracting a feature from the generation image generated by the image generation unit; an identification unit for identifying an object in the image using the feature output from the feature extraction unit; an output unit for outputting an identification result output from the identification unit; and a feature map generation unit for instructing the image generation unit generating a new generation image based on the feature output from the feature extraction unit, and generating a feature map indicating a feature extraction condition for the new generation image and output the generated feature map to the feature extraction unit. With this configuration, a target object in the image is identified quickly and accurately.