Iris Recognition System

This repository provides a complete, Python-based biometric identification pipeline for iris recognition. Contained entirely within iris.py, the system leverages OpenCV and classical computer vision techniques to segment the iris, extract robust features, and verify identity by comparing local scale-invariant descriptors.

Core Capabilities

• Accurate Boundary Detection: Utilizes Canny edge detection, median blurring, and custom Hough Circle Transform heuristics to isolate the pupil (inner boundary) and the exterior iris (outer boundary).
• Image Normalization: Applies intelligent masking to remove eyelids and eyelashes (specifically targeting the top sector between 50 and 130 degrees) and uses histogram equalization to standardize lighting.
• ROI Segmentation: Unwraps and isolates the iris into discrete geometric patches: right-side, left-side, bottom, and complete for localized feature analysis.
• SIFT Feature Extraction: Computes Scale-Invariant Feature Transform (SIFT) keypoints and descriptors within the defined iris regions, actively filtering out noisy keypoints that fall inside the pupil or outside the iris bounds.
• Robust Feature Matching: Matches geometric keypoints between two images using a Brute-Force Matcher (KNN). The matches are rigorously filtered using a spatial distance ratio, as well as angle and distance standard deviation checks to eliminate false positives.
• Binary Serialization: Supports storing and loading pre-computed keypoints and descriptors using gzip and pickle to drastically speed up repetitive comparisons.

Prerequisites

To run the recognition pipeline, you will need a Python 3 environment with the following dependencies:

• numpy
• opencv-python (cv2) - Note: SIFT is utilized, requiring OpenCV version 4.4.0 or higher.
• matplotlib

Standard library modules used include os, sys, math, random, pickle, copy, gzip, inspect, and itertools.

File Structure

The repository relies on a single modular script:

File	Description
iris.py	The main execution script containing the full computer vision pipeline, from boundary segmentation to feature matching.

Usage

Direct Execution

By default, executing iris.py runs a localized comparison test on two provided sample images. Ensure the sample images (IMG_002_L_2.JPG and IMG_002_R_3.JPG) are located in the same working directory.

python iris.py

Programmatic Integration

You can import iris.py into other systems or evaluation scripts to utilize specific stages of the pipeline:

Python

import iris

# Compare two raw iris images directly
iris.compare_images('path/to/image1.JPG', 'path/to/image2.JPG')

# Compare two serialized/pickled ROI binaries
iris.compare_binfiles('path/to/data1.gz', 'path/to/data2.gz')

Pipeline Architecture

1. Preprocessing (load_image): Loads the input images in grayscale mode.
2. Segmentation (get_iris_boundaries, find_pupil, find_ext_iris): Iteratively scans for circular gradients to isolate the distinct inner (pupil) and outer (sclera) boundaries.
3. Enhancement (get_equalized_iris): Masks noisy physiological regions and equalizes the pixel intensity distribution.
4. Feature Mapping (get_rois, load_keypoints, load_descriptors): Flattens regions of interest and computes SIFT spatial features.
5. Matching & Verification (getall_matches, get_matches): Computes k-NN matches and applies rigid geometric validation (median differences and standard deviation thresholds) to return a quantitative similarity assessment.