Patch-Based Object Re-identification Using Feature Vector Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for re-identifying individuals in images are error-prone due to significant data loss and resolution degradation when comparing low-resolution images, especially under varying lighting and pose changes.
Innovation Solution
The system divides images into patches, extracts feature vectors, repositions patches based on deformation costs, and compares them using a patch-based metric learning approach to determine object similarity, combining color and texture descriptors for accurate re-identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional methods compare total images directly, then the process is simple, but significant data loss and resolution degradation occur
Solution Approach 1:
The patent divides images into multiple patches instead of comparing entire images directly. Each patch is processed independently to extract local features, which are then aggregated for comparison. This segmentation approach preserves local image details and reduces information loss while maintaining computational feasibility.
2Productivity
If image compression is applied to enable comparison, then computational efficiency improves, but re-identification accuracy decreases due to data loss
Solution Approach 1:
The patent extracts local features from image patches using feature vectors that capture color, texture, and structural information. These extracted features are then compared directly without requiring full image compression, thereby maintaining accuracy while achieving computational efficiency through selective feature processing.
3Productivity
If conventional total image comparison is used, then the method is computationally efficient, but it is error-prone under varying lighting and pose conditions
Solution Approach 1:
The patent applies different processing and comparison strategies to different local regions (patches) of the image. Each patch is evaluated independently for its local characteristics, allowing the system to be more robust to global variations in lighting and pose while maintaining computational efficiency through localized analysis.
4Measurement precision
If high-resolution images are used for comparison, then re-identification accuracy improves, but data processing requirements and computational cost increase significantly
Solution Approach 1:
The patent processes only the necessary local features from image patches rather than analyzing entire high-resolution images. By extracting and comparing only the relevant local characteristics needed for identification, the system achieves high accuracy while significantly reducing computational cost and data processing requirements.
Data Source
AI summary
There is provided a system including a memory and a processor configured to receive a first image depicting a first object and a second image depicting a second object, divide the first image into a first plurality of patches and the second image into a second plurality of patches, extract plurality of feature vectors from each of the first plurality of patches and a second plurality of feature vectors from the second plurality of patches, determine dissimilarities based on a plurality of patch metrics, each patch dissimilarity measure being a dissimilarity between corresponding patches of the first plurality of patches and the second plurality of patches, compute an image dissimilarity between the first image and the second image based on an aggregate of the plurality of patch dissimilarity measures, evaluate the image dissimilarity to determine a probability of whether the first object and the second object are the same.