Dynamic Image Matching Method Selection for Resolution Mismatch
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face recognition systems face significant accuracy degradation when comparing images of differing resolutions, particularly in surveillance environments where high-resolution mugshots are matched with low-resolution CCTV images, due to issues like upscaled noise and reduced information during downscaling.
Innovation Solution
A method dynamically selects the most appropriate image matching method based on the underlying resolution of the probe and gallery images, allowing for tuning to work effectively with both high and low resolution images, using techniques such as Multi-Region Histogram analysis and discrete cosine transformation to determine and classify image resolutions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If images are resized to a common intermediate format size, then the face matching method can be applied consistently, but recognition accuracy degrades significantly when comparing images of differing underlying resolutions
Solution Approach 1:
The system dynamically selects between different face matching methods (holistic vs. local-feature based) based on the underlying resolution of the probe image. For high-resolution probe images, the system uses holistic methods; for low-resolution probe images, it uses local_feature methods. This dynamic adaptation resolves the contradiction by allowing the system to maintain both method compatibility and recognition accuracy across varying resolution conditions.
Solution Approach 2:
The system changes the parameter of matching method selection based on the resolution characteristic of the input images. By detecting the underlying resolution and adjusting the matching method accordingly, the system optimizes recognition accuracy while maintaining versatility across different resolution scenarios.
2Shape
If low-resolution images are upscaled to match high-resolution images, then image size consistency is achieved, but noise and artifacts are introduced and recognition accuracy drops
Solution Approach 1:
Instead of always upscaling low-resolution images to match high-resolution images, the system inverts the approach by detecting the underlying resolution and selecting appropriate matching methods for each case. This prevents the introduction of noise and artifacts that would result from forced upscaling, while still achieving image size consistency through selective processing.
3Shape
If high-resolution images are downscaled to match low-resolution images, then image size consistency is achieved, but the amount of information is reduced
Solution Approach 1:
The system applies local_feature based methods for low-resolution probe images, which focus on extracting key local features rather than requiring full image information. This allows the system to maintain information availability by concentrating on the most discriminative local characteristics, while still achieving image size consistency through selective feature extraction.
4Device complexity
If a single face matching method is used for all image resolutions, then system simplicity is maintained, but performance degrades on resolution-mismatched images
Solution Approach 1:
The system introduces dynamic method selection based on probe image resolution detection. By automatically choosing between holistic and local_feature based methods according to the underlying resolution, the system maintains relative simplicity while significantly improving performance on resolution-mismatched images compared to using a single fixed method.
Data Source
AI summary
The disclosure concerns face recognition systems. The aim is to identify candidate matching images to a probe image. There is provided methods, software and computer system to select (22, 24) a method of matching images from two or more methods of matching images based on an underlying resolution (20) of the probe image (8). Comparing two images of differing resolutions is common in surveillance environments. To alleviate this degradation, the method advantageously dynamically selects the most appropriate matching method for a probe image. The disclosure also provided methods to determine the underlying resolution of a probe (8) or gallery image (14).