Dynamic Image Segmentation via Recursive Neural Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing document layout analysis methods face challenges in determining appropriate segmented area sizes for composite images, as the size varies with document type and analysis purpose, making it difficult to effectively segment and process images.
Innovation Solution
A computer-readable recording medium stores a program that uses machine learning to input images to a first model for segmentation, with the segmented images then being processed by a second model, determining whether to further segment based on the output, allowing for appropriate segmentation of images with varying area sizes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a fixed segmentation size is used for composite images, then the segmentation process is simple, but it cannot adapt to different document types and analysis purposes
Solution Approach 1:
The patent implements dynamic segmentation by using a neural network model that automatically determines segmentation sizes based on the input image characteristics. The model evaluates features such as text density, image resolution, and document type to dynamically adjust the number and size of segmented areas, replacing fixed segmentation schemes with an adaptive, data-driven approach that varies according to the specific document being processed.
Solution Approach 2:
The patent changes the segmentation parameters (number of segments, segment size, segmentation granularity) based on the input image characteristics. The neural network model processes features extracted from the composite image and outputs optimized segmentation parameters that adapt to different document types and analysis requirements, allowing the same system to handle various segmentation needs without manual reconfiguration.
2Productivity
If manual extraction of segmented areas is performed, then segmentation accuracy can be controlled, but processing time and labor increase
Solution Approach 1:
The patent implements self-service segmentation where the neural network model automatically performs the segmentation task without requiring manual intervention. The model independently analyzes the composite image, determines appropriate segmentation sizes, and generates segmented areas autonomously based on learned patterns from training data, eliminating the need for manual extraction while maintaining high segmentation accuracy through its trained parameters.
Solution Approach 2:
The patent replaces manual mechanical extraction processes with an automated neural network-based system. Instead of human operators manually selecting and extracting segmented areas, the neural network model uses computational algorithms to automatically detect, segment, and process images, substituting human cognitive and manual operations with machine learning-based automation that achieves both speed and accuracy.
3Reliability
If recursive segmentation is applied to segment composite images into unit images, then segmentation completeness improves, but the number of processing steps increases
Solution Approach 1:
The patent implements feedback-based recursive segmentation where the neural network model continuously evaluates the segmentation results and adjusts the segmentation process accordingly. The model processes segmented images, evaluates their characteristics, and determines whether further segmentation is needed based on feedback from the segmentation output. This feedback mechanism ensures complete segmentation into unit images while efficiently stopping the recursive process when further segmentation yields no additional benefits, reducing unnecessary processing steps.
Data Source
AI summary
A non-transitory computer-readable recording medium stores a program for causing a computer to execute a process, the process includes inputting an accepted image to a first model generated through machine learning based on a composite image and information, the composite image being obtained by combining a first plurality of images each of which includes one area, the information indicating a combination state of the first plurality of images in the composite image, inputting a first image among a second plurality of images output by the first model to a second model generated through machine learning based on an image which includes one area and an image which includes a plurality of areas, and determining whether to input the first image to the first model, based on a result output by the second model in response to the inputting of the first image.