Overlapping Image Segmentation for Feature Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional deep learning-based image classification methods fail to fully utilize high-resolution image information, leading to loss of feature details during resizing and difficulties in correlating neighboring image segments, which affects accurate classification, especially in distinguishing objects with similar shapes based on internal characteristics like color, pattern, and texture.
Innovation Solution
An image segmentation method that generates multiple generations of overlapping image segments, allowing for the preservation of small feature areas and enabling deep learning-based classifiers to learn correlated segments without information loss, by dividing images in an overlapped manner and selecting segments based on pixel similarity and size criteria.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the entire image is reduced to a smaller size for classification, then the processing speed and efficiency are improved, but small feature areas and characteristic patterns disappear or change shape leading to loss of important visual information
Solution Approach 1:
The image is divided into multiple overlapping segments, where each segment is processed independently at high resolution. This segmentation allows the system to maintain detailed feature information in each local region while enabling parallel processing of multiple segments, thus resolving the contradiction between processing speed and feature preservation.
Solution Approach 2:
The patent implements multi-scale processing where segments are processed at different resolution levels. Coarse segments provide overall context while finer segments capture detailed features, creating a nested structure of information that preserves both global and local characteristics without requiring full-image downsampling.
2Productivity
If the image is divided into a grid of equal-sized cells for parallel processing, then the processing efficiency is improved, but it becomes difficult to find correlation between neighboring image segments and the segments may contain only partial information
Solution Approach 1:
Adjacent image segments are merged with overlapping regions to form combined segments. This merging process integrates information from multiple neighboring segments, allowing the system to capture contextual relationships and correlations between adjacent regions while maintaining the benefits of parallel processing.
Solution Approach 2:
The patent adds a temporal dimension to the segmentation process by creating multiple generations of segments. First-generation segments provide local details, while second-generation segments (formed by merging adjacent first-generation segments) provide contextual information, effectively adding a dimension of spatial context without sacrificing parallel processing efficiency.
3Measurement precision
If high resolution images are used to preserve feature details, then the classification accuracy for objects with similar shapes is improved, but the processing time and computational resources increase significantly
Solution Approach 1:
By segmenting the high-resolution image into smaller overlapping regions, the system can process multiple segments in parallel rather than processing the entire image sequentially. This segmentation strategy maintains the high resolution necessary for accurate feature detection while reducing the time penalty through parallel computation.
Solution Approach 2:
The patent processes only the most informative segments in full detail while using coarser representations for less critical regions. The overlapping segment strategy ensures that areas containing important features receive excessive processing attention, while other areas use partial processing, optimizing the balance between accuracy and processing time.
Data Source
AI summary
An image segmentation method according to an embodiment of the present invention is performed in a computing device having one or more processors and memory for storing one or more programs executed by means of the one or more processors, and includes the steps of: (a) receiving the input of an image; (b) generating a first-generation image segment set by dividing the input image in an overlapped manner; and (c) generating a second or higher-generation image segment set from the first-generation image segment set, wherein a subsequent-generation image segment set is generated by dividing in an overlapped manner at least one of a plurality of image segments included in the previous-generation image segment set.