Image Matching System Using Normalized Shape and Color Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image matching algorithms fail to effectively handle size, position, and 2D rotation invariance, particularly struggling with smooth-curved objects and missing overall shape information, leading to false matches and limited adaptability.
Innovation Solution
An image matching system that identifies an object area, normalizes it, calculates shape and color vectors, and determines matches based on these vectors, using edge detection, silhouette creation, and vectorization to achieve size, position, and rotation invariance, while addressing issues like patch color differences and sub-area variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If handcrafted algorithms extract distinctive features from raw pixel values, then matching reliability is improved for objects with sharp corners or junctions, but matching fails for smooth-curved objects without distinctive features
Solution Approach 1:
The patent transforms the image representation from raw pixel values to multiple parameter spaces including gradient magnitude, gradient direction, and normalized coordinates. This parameter transformation enables the algorithm to capture both local feature details and global shape information, making it effective for both sharp-cornered and smooth-curved objects by changing how object characteristics are represented and compared
2Measurement precision
If machine learning algorithms use CNN for feature extraction, then classification accuracy is improved, but the ability to handle color variations and allow easy algorithm modification is reduced
Solution Approach 1:
The patent segments the feature extraction process into distinct computational stages: gradient magnitude calculation, gradient direction computation, coordinate normalization, and shape vector construction. This segmentation allows independent optimization and modification of each stage, enabling easy adaptation to different color spaces and object types without retraining entire models, while maintaining high matching accuracy through the coordinated operation of all segments
3Measurement precision
If image matching algorithms focus on local features, then feature detection precision is improved, but overall shape information is lost leading to false matches
Solution Approach 1:
The patent extends the feature representation from 2D spatial coordinates to 3D shape vectors by adding the gradient magnitude dimension. Each feature point is represented as (normalized_x, normalized_y, gradient_magnitude), creating a three-dimensional feature space. This dimensional extension preserves both the precise local feature location and the global shape context, as the distribution of points in this 3D space encodes the overall object shape while maintaining local feature detection precision
Data Source
AI summary
An image matching system includes a non-transitory computer-readable medium and a processor. The non-transitory computer-readable medium is configured to store information of a plurality of images. The processor is configured to identify an object area in an original image that illustrates an object. The processor is configured to normalize the object area, resulting in a normalized image. The processor is configured to calculate a shape vector and a color vector from the normalized image. The processor is configured to calculate a match score using the shape vector and the color vector. The processor is configured to determine if the non-transitory computer-readable medium stores an identical match for the original image based on the match score.