IRIS SEGMENTATION
Group No:8 Arpit Agrawal (07D07007) Satyaprakash Pareek (07D07030)
Content
Introduction Overview of IRIS Recognition System Conventional Methods Proposed Methods Selection Of Wavelet Function Non-separable wavelets Expected Output Reference
Introduction
Biometric
uniqueness, stability and noninvasiveness
IRIS Recognition:
IRIS Segmentation Normalization Enhancement Matching
Isolating actual iris area from image of eye Conventional methods assume Iris boundaries circular Edge detection techniques used for segmentation These approaches have problems
Conventional Methods
Normal Edge Detection
Use edge detectors to produce edges to detect iris boundaries. Additional edges are produced due to eye lashes and iris texture. leads to inaccurate iris boundaries.
Conventional Methods
Assumed shape of IRIS boundaries
Circular or Elliptical
Past edge detection based on convolution with directional derivatives (kernels) Failed in case of noise images. Soln- Gaussian filtering than detection Not optimal smoothens sharp peaks Canny was proposed as a soln Wavelet based edge detection similar to canny was proposed
Derived non separable wavelet filters based on this matrix. m0 is low pass filter m1 and m2 are high pass anti-symmetric filters very effective in detecting edges Use m1 and m2 for modulus maxima Decimation not performed
Ease in mapping Avoids error in mapping
Comparison
Expected Output
Reference