Convolution Neural Network Image Search with LambdaMART Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image search systems face challenges in accurately and efficiently comparing features of an image-to-be-identified with a large number of images in a database to find similar images, especially in medical imaging where human assessment becomes unstable with increasing image volumes.
Innovation Solution
The system employs a convolution neural network algorithm for feature extraction and a LambdaMART algorithm for ranking, combined with a Correlation Based Feature Selection Algorithm, to identify and compare features of images, enabling accurate image similarity ranking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If images are compared with the image-to-be-identified based on a plurality of image features to obtain a search result, then the similarity assessment becomes more comprehensive, but the difficulty of detecting and measuring increases with the increase of the number of images
Solution Approach 1:
The patent introduces an automated image analysis system as an intermediary between the image-to-be-identified and the image database. This system extracts multiple image features (color, texture, shape, etc.) and performs automated comparison algorithms to assess similarity, replacing manual human assessment and enabling comprehensive multi-feature comparison while reducing the difficulty of detection and measurement.
Solution Approach 2:
The patent replaces the mechanical human visual assessment system with an automated computer-based image analysis system. This substitution enables the system to handle a large number of images by automatically extracting and comparing multiple features, thereby improving measurement precision while reducing the difficulty of detecting and measuring image similarities.
2Quantity of substance
If the number of images in the image database increases over time, then the comprehensiveness of the search result improves, but the stability of the assessment deteriorates when using human eyes for comparison
Solution Approach 1:
The patent implements an automated image analysis system that performs self-service image comparison and similarity assessment. The system automatically extracts features from the image-to-be-identified and database images, compares multiple features, and generates similarity assessments without human intervention. This ensures consistent and stable assessment reliability regardless of the increasing number of images in the database.
Solution Approach 2:
The patent transforms the assessment process from manual human evaluation to automated multi-parameter feature comparison. By extracting and comparing multiple image features (color, texture, shape, etc.) using standardized algorithms, the system maintains stable and reliable assessments even as the database grows, eliminating the variability inherent in human visual assessment.
3Measurement precision
If features of the image-to-be-identified are compared with features of images as many as possible in the image database, then the accuracy of finding the most similar image improves, but the loss of time increases due to the large number of comparisons required
Solution Approach 1:
The patent applies preliminary action by pre-extracting and storing multiple image features (color, texture, shape, etc.) for all images in the database before the actual search is performed. When a search is initiated, the system only needs to compare the pre-extracted features of the image-to-be-identified with the stored features, significantly reducing the time required for comparison while maintaining high accuracy in finding the most similar image.
Solution Approach 2:
The patent segments the image comparison process into distinct feature extraction and feature comparison stages. By dividing the complex image data into multiple independent feature components (color, texture, shape, etc.) and processing them separately, the system can efficiently compare images based on these segmented features, improving search speed while maintaining measurement precision.
Data Source
AI summary
The present disclosure relates to a method and system for image searching. In the image searching process, different images comprising specific regions are obtained respective, and features in the specific regions of the different images are extracted respectively based on a convolution neural network algorithm. Feature data related to the features corresponding to the different images is calculated, and the different images are ranked based on the feature data so as to identify the image needed.