Weighted Binary Descriptor Image Registration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current feature-based image registration methods, such as SIFT and SURF, are computationally expensive and less efficient for real-time processing, especially in mobile devices, and binary descriptors like ORB suffer from lower accuracy in challenging conditions.
Innovation Solution
The method improves binary descriptors by learning a small number of parameters and using a weighted Hamming distance with adaptive thresholds to account for image distortions, enhancing accuracy and efficiency without compromising computational speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional feature-based methods (SIFT, SURF) are used for image registration, then robustness to image size, illumination, and rotation changes is improved, but computational complexity increases and processing speed decreases
Solution Approach 1:
The patent replaces expensive, complex feature descriptors (SIFT's 128-vector, SURF's 64-vector) with inexpensive binary descriptors that can be computed and compared rapidly. The binary descriptor uses simple pixel intensity comparisons instead of complex gradient calculations, making it computationally lightweight while maintaining sufficient discrimination capability for real-time applications.
Solution Approach 2:
The patent substitutes the mechanical/computational complexity of gradient-based descriptor computation with a simpler binary comparison system. Instead of calculating gradients, determinants, and high-order derivatives, the system uses direct pixel intensity thresholding and binary encoding, dramatically reducing computational burden while preserving key matching capabilities.
2Productivity
If binary descriptors (ORB) are used to improve computational speed, then processing efficiency increases, but matching accuracy decreases in challenging conditions
Solution Approach 1:
The patent modifies the binary descriptor generation process by introducing adaptive thresholding parameters and weighting schemes. Instead of using fixed thresholds for pixel comparisons, the system learns optimal thresholds from training data and adapts them to different image conditions. This allows the binary descriptor to maintain speed advantages while improving discrimination accuracy in challenging scenarios.
Solution Approach 2:
The patent performs preliminary learning and adaptation of descriptor parameters before actual image matching. By pre-computing optimal thresholds, weights, and other parameters from training data, the system prepares the binary descriptor to handle various image conditions effectively, thereby improving matching accuracy without compromising real-time processing speed during actual operation.
3Measurement precision
If high-dimensional descriptors (SIFT 128-vector) are used to improve discrimination capability, then matching accuracy improves, but memory consumption and computation time increase
Solution Approach 1:
The patent extracts only the essential discriminative information needed for image matching and encodes it in a compact binary form. By taking out and retaining only the most critical features (encoded as binary values rather than floating-point vectors), the system achieves sufficient discrimination capability with dramatically reduced memory requirements compared to 128-vector or 64-vector descriptors.
Solution Approach 2:
The patent transforms the descriptor representation from a high-dimensional continuous space (128 or 64 floating-point values) to a low-dimensional discrete binary space (256 bits). This dimensional transformation maintains the essential discrimination capability while reducing memory consumption by a factor of 512 compared to SIFT, enabling efficient storage and processing of large numbers of descriptors.
Data Source
AI summary
An image registrations includes determining a first binary descriptor of a first key point in a first image, determining a second binary descriptor of a second key point in a second image, determining a weighted Hamming distance between the first binary descriptor and the second binary descriptor, and registering the first key point with the second key point when the weighted Hamming distance is below a noise threshold. At least one element in the first or the second binary descriptor is a result of a comparison of a difference between intensities of at least two pixels of the first or the second image with a threshold. At least two weights of the weighted Hamming distance for comparing at least two elements of the first or the second binary descriptors are different.