Bovine Iris Recognition Using SIFT Feature Bags
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing animal iris recognition technologies face challenges with imperfect images due to rotation, shifting, local blockage, or inconsistent scale, leading to inaccurate outer boundary detection and limited application in systems like food traceability.
Innovation Solution
A method utilizing SIFT feature bags for bovine iris recognition, which includes preprocessing, detecting inner boundaries, extracting SIFT feature points, and comparing them to an optimal feature bag to achieve accurate recognition, even with imperfect images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional iris recognition methods are used on animals, then identification accuracy can be achieved under ideal conditions, but the system fails when images are imperfect (rotated, shifted, blocked, or inconsistent scale)
Solution Approach 1:
The patent transforms the iris recognition approach by changing from traditional geometric parameter-based methods to SIFT feature-based methods. SIFT features are invariant to scale, rotation, and affine transformations, allowing the system to maintain recognition accuracy across varying image conditions. The feature histogram comparison further enhances adaptability by comparing distributions rather than exact positions, making the system robust to shifts and rotations.
Solution Approach 2:
The patent segments the iris image processing into distinct stages: detecting the inner boundary contour, extracting SIFT feature points from specific regions, and comparing feature histograms. This segmentation allows the system to focus on invariant features and ignore problematic areas, improving reliability for imperfect images.
2Measurement precision
If traditional iris boundary detection and normalization methods are used, then high-quality normalized images can be obtained for human iris recognition, but these methods cannot accurately detect outer boundaries in animal iris images
Solution Approach 1:
Instead of detecting the outer boundary and normalizing from the outside in (as in human iris recognition), the patent inverts the approach by first detecting the inner boundary contour and then extracting features from regions defined relative to this inner boundary. This inversion makes the method adaptable to animal iris images where outer boundary detection is unreliable.
Solution Approach 2:
The patent applies local quality by extracting SIFT features from specific regions (inner ring and middle ring) defined by the inner boundary, rather than attempting to process the entire iris area. This localized feature extraction focuses computational effort on reliable regions, improving boundary detection accuracy for animal images.
3Loss of information
If SIFT feature extraction is performed on the entire iris image, then comprehensive feature information is obtained, but features from the inner boundary region introduce errors in recognition
Solution Approach 1:
The patent extracts only the necessary SIFT features from specific regions (inner ring and middle ring) while excluding features from the inner boundary region. This selective extraction removes problematic features that would introduce errors while retaining sufficient information for accurate recognition through histogram comparison.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Disclosed are a SIFT feature bag based bovine iris image recognition method, comprising the following steps: preprocessing an iris image to obtain and effective region; obtaining feature points by using a SIFT method; positioning an inner edge by using an active contour line method; removing the feature point in the inner edge to obtain an effective SIFT feature point set; performing comparison with an optimal SIFT feature bag to obtain a feature histogram; calculating the histogram distance between a to-be-recognized image and each image in a target iris library, and using an object corresponding to the target bovine iris image with the smallest histogram distance as an recognition result. The present invention can accurately perform recognition in cases of that the to-be-recognized bovine iris image is rotated, shifted, partially blocked, or inconsistent in scale, so as to improve the accuracy and reliability of bovine iris image recognition, thereby promoting application of the iris-based recognition method in the food traceability system.