Image Search System Region-Specific Feature Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image search systems struggle to find images that differ in specific parts from an original image, such as clothing images differing only in sleeve patterns.
Innovation Solution
An image search system that allows users to designate and process specific regions of an original image, using feature information from both processed and unprocessed regions to search for images with similar or differing characteristics, and outputs results based on similarity degrees.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If image search is performed based on the original image as a whole, then the search covers all regions of the image, but it becomes difficult to search for images that differ in specific parts only
Solution Approach 1:
The patent divides the original image into multiple regions (processed region and non-processed region) and performs separate feature extraction and matching for each region. This allows the system to search for images that match the non-processed regions exactly while allowing variations in processed regions, thereby enabling region-specific search flexibility.
Solution Approach 2:
The patent applies different processing and matching strategies to different regions of the image. The non-processed regions are matched with high precision using multiple feature information, while the processed regions allow for more flexible matching. This local differentiation resolves the contradiction between overall search coverage and region-specific matching accuracy.
2Measurement precision
If all feature information from the original image is used for search, then comprehensive matching is achieved, but the system cannot find images with specific regional differences
Solution Approach 1:
The patent extracts and separates the feature information of the processed region from the overall image feature information. By taking out the processed region's features and handling them differently (with lower weight or optional matching), the system maintains comprehensive matching for non-processed regions while enabling flexible search for images with specific regional differences.
Solution Approach 2:
The patent segments the feature information into multiple categories (color, texture, shape, etc.) and further segments them by region (processed vs. non-processed). This segmentation allows selective application of matching criteria, achieving both comprehensive matching where needed and flexibility where variations are expected.
3Productivity
If the original image is included in search results, then complete search coverage is maintained, but the original image may be prioritized unnecessarily
Solution Approach 1:
The patent calculates the degree of similarity between search results and the original image, and uses this feedback to control the output order. By incorporating this similarity feedback into the ranking mechanism, the system maintains complete search result coverage while preventing the original image from being inappropriately prioritized, as users can control whether to include exact matches based on their search intent.
Data Source
AI summary
Provided is an image search system capable of searching for an image that differs in a specific part from an original image. A reception unit receives designation of a partial region in an original image to be processed and a process content for the partial region. A search unit searches for an image identical or similar to a processed image, which is obtained by applying a process of the process content to the partial region of the original image, based on: a plurality of kinds of feature information on a region other than the partial region of the processed image or of the original image; and at least one piece of feature information selected based on the process content from among a plurality of kinds of feature information on the partial region of the processed image or of the original image.