Content-Based Image Retrieval Using Multi-Color Texture Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Content-Based Image Retrieval (CBIR) systems face challenges in accurately and efficiently searching for identical and similar images or objects across large databases, particularly in handling complex high-level visual information, scalability, and adaptability to varying capture conditions and new data, due to limitations in feature descriptor selection and similarity analysis methods.
Innovation Solution
A CBIR mechanism that employs a combination of computer vision and statistical methods, using high-level feature descriptors such as multi-color and multi-texture features, along with shape and geometry descriptors, and an objective similarity function analysis based on Euclidean and Mahalanobis distances, to provide real-time scalability and robustness to various capture conditions without requiring learning or training methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If direct 1:1 comparisons or pattern matching methods are applied to entire images, then accuracy in recognizing global structure is improved, but the system cannot cope well with partial occlusion, significant changes in viewpoint, or deformable transformations
Solution Approach 1:
The patent divides the image into multiple regions of interest (ROIs) and extracts features from each region separately. This segmentation allows the system to handle partial occlusion and viewpoint changes by focusing on visible regions while maintaining overall accuracy through aggregation of regional features.
Solution Approach 2:
The patent applies different feature extraction strategies to different regions of the image based on their specific characteristics. By tailoring the analysis to local properties of each ROI, the system achieves both high accuracy in recognizing global structure and robustness to local variations such as occlusion and deformation.
2Reliability
If low-level feature based descriptors are used, then robustness to different capture conditions is improved, but the system cannot describe high-level features which are more complex and costly
Solution Approach 1:
The patent combines multiple low-level feature descriptors (color, texture, shape) into a unified feature representation that captures both robustness to capture conditions and high-level semantic information. By merging these complementary features, the system achieves comprehensive description without relying on complex high-level features alone.
Solution Approach 2:
The patent creates a composite feature descriptor that integrates multiple types of low-level features (color histograms, texture descriptors, shape metrics) into a single comprehensive representation. This composite approach provides both the robustness of low-level features and the descriptive power needed for high-level feature characterization.
3Measurement precision
If machine learning and training methods are applied over previously known limited databases, then accuracy for specific objects is improved, but the system is not extendable to other objects and categories without corresponding prior training
Solution Approach 1:
The patent develops a universal feature extraction framework that can be applied to any type of image or object without requiring specific training. The system uses general-purpose descriptors (color, texture, shape) that work across diverse categories, enabling the system to handle both trained and new objects with consistent performance.
Solution Approach 2:
The patent implements a self-adaptive system that automatically adjusts its feature extraction and matching parameters based on the input image characteristics without requiring external training. The system serves itself by dynamically optimizing its operation for each query, eliminating the need for prior training on specific object categories.
4Ease of operation
If text-based retrieval with keywords and tags is used, then image indexing capability is improved, but images need to be indexed and labeled entailing a great deal of time and resources
Solution Approach 1:
The patent implements automatic feature extraction and image indexing without requiring manual labeling or keyword assignment. The system autonomously analyzes image content, extracts relevant features, and creates indexes based on these features, eliminating the time-consuming manual indexing process while maintaining effective retrieval capability.
Data Source
AI summary
A method and non-transitory computer readable medium for content based image retrieval. The method includes selecting a query image, segmenting the selected query image by applying a segmentation technique, extracting features from the segmented query image by determining at least two feature descriptors, including color feature descriptors and texture feature descriptors, and determining a similarity of the query image to a plurality of images included in a database using the determined at least two feature descriptors of the segmented query image, features being extracted from each of the plurality of images included in the database by determining the at least two feature descriptors, the color feature descriptors and the texture feature descriptors including a simultaneous combination of different color spaces, and global and local statistical measurements being carried out on the simultaneous combination of the different color spaces.