Fingerprint enhancement

Question

Fingerprint sensor (Persona) is being used to get a fingerprint image. I am trying to enhance this image. I am using OpenCV for this purpose. Here is my original image:

I have applied otsu transform on it and got this image:

Now I have applied Gabor filter from OpenCV on orientations of 0, 45, 90, 135. I have got this result:

Here is my code in Python OpenCV for application of gabor filter:

import numpy as np
import cv2

from matplotlib import pyplot as plt

//cv2.getGaborKernel(ksize, sigma, theta, lambda, gamma, psi, ktype)
// ksize - size of gabor filter (n, n)
// sigma - standard deviation of the gaussian function
// theta - orientation of the normal to the parallel stripes
// lambda - wavelength of the sunusoidal factor
// gamma - spatial aspect ratio
// psi - phase offset
// ktype - type and range of values that each pixel in the gabor kernel 
//canhold

g_kernel = cv2.getGaborKernel((25, 25), 6.0, np.pi/4, 8.0, 0.5, 0, ktype=cv2.CV_32F)
g_kernel1 = cv2.getGaborKernel((30, 30), 6.0, (3*np.pi)/4, 8.0, 0.5, 0, ktype=cv2.CV_32F)
g_kernel2 = cv2.getGaborKernel((30, 30),4 , 0, 8, 0.5, 0, ktype=cv2.CV_32F)
g_kernel3 = cv2.getGaborKernel((30, 30),4 , np.pi, 8, 0.5, 0, ktype=cv2.CV_32F)

print np.pi/4
img = cv2.imread('C:/Users/admin123/Desktop/p.png')
img1 = cv2.imread('C:/Users/admin123/Desktop/p.png')
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
img1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)

// Otsu thresholding
ret2,img1 = cv2.threshold(img1,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU)
cv2.imshow('otsu', img1)

filtered_img = cv2.filter2D(img, cv2.CV_8UC3, g_kernel)
filtered_img1 = cv2.filter2D(img, cv2.CV_8UC3, g_kernel1)
filtered_img2 = cv2.filter2D(img, cv2.CV_8UC3, g_kernel2)
filtered_img3 = cv2.filter2D(img, cv2.CV_8UC3, g_kernel3)

cv2.imshow('0', filtered_img)
cv2.imshow('1', filtered_img1)
cv2.imshow('2', filtered_img2)
cv2.imshow('image', img)

cv2.addWeighted(filtered_img2,0.4,filtered_img1,0.8,0,img) #0 degree and 90
cv2.addWeighted(img,0.4,filtered_img,0.6,0,img) #0 degree and 90
cv2.addWeighted(img,0.4,filtered_img3,0.6,0,img)
cv2.addWeighted(img,0.4,img1,0.6,0.3,img)

cv2.imshow('per',img)

//threshold will convert it plain zero and white image
ret,thresh1 = cv2.threshold(img,150,255,cv2.THRESH_BINARY)#127 instead of 200

cv2.imshow('per1',thresh1)

h, w = g_kernel.shape[:2]
g_kernel = cv2.resize(g_kernel, (3*w, 3*h), interpolation=cv2.INTER_CUBIC)
g_kernel1 = cv2.resize(g_kernel1, (3*w, 3*h), interpolation=cv2.INTER_CUBIC)

cv2.imshow('gabor kernel (resized)', g_kernel)
cv2.imshow('gabor kernel1 (resized)', g_kernel1)

cv2.waitKey(0)
cv2.destroyAllWindows()

I want robust fingerprint recognition. For this I want image of this level to get accurate Minutiae points:

How can I get this much result from enhancement? What changes are required in code to get the enhanced result?

Answer 1

Well I dont have a python/opencv answer but I can point you the resource where you can fiddle around with a Matlab code. You can find the code here Simple Fingerprint Matcher The code basically holds all the code for enhacement/minutiae extraction/matching. Though not very robust on matching but enhancements are pretty good.

I ran the code on the sample you uploaded, which came out as follows.

Note that the code uses two different approaches combined for fingerprint enhancement. One is based on Gabor filters and the other is called STFT (Short Time Fourier Transform), But it is likely that you will only need the Gabor filter part. Actually depends upon the image quality. If you need the Gabor filter code in Matlab, you can find it herehttp://www.peterkovesi.com/matlabfns/#fingerprints But I did modify the code to show up the images and process only a single finger. The following is the main file calling in steps the extraction of enhanced fingerprint. The matlab function doing that is f_enhance.m

function [ binim, mask, cimg1, cimg2, oimg1, oimg2 ] = f_enhance( img )
enhimg =  fft_enhance_cubs(img,6);             % Enhance with Blocks 6x6
enhimg =  fft_enhance_cubs(enhimg,12);         % Enhance with Blocks 12x12
[enhimg,cimg2] =  fft_enhance_cubs(enhimg,24); % Enhance with Blocks 24x24
blksze = 5;   thresh = 0.085;                  
normim = ridgesegment(enhimg, blksze, thresh);
oimg1 = ridgeorient(normim, 1, 3, 3);                 
[enhimg,cimg1] =  fft_enhance_cubs(img, -1);
[normim, mask] = ridgesegment(enhimg, blksze, thresh);
oimg2 = ridgeorient(normim, 1, 3, 3); 
[freq, medfreq] = ridgefreq(normim, mask, oimg2, 32, 5, 5, 15);
binim = ridgefilter(normim, oimg2, medfreq.*mask, 0.5, 0.5, 1);
figure,imshow(binim,[]); % Normalize to grayscale
binim = ridgefilter(normim, oimg2, medfreq.*mask, 0.5, 0.5, 1) > 0;
figure;
figure,imshow(binim);
figure;
end

Either you revert to Matlab or you can always translate the code :) Good luck

Answer 2

I may be late here. But this might be useful for other people later on.

Take a look at this reporitory: https://github.com/Utkarsh-Deshmukh/Fingerprint-Enhancement-Python

It performs Fingerprint Enhancement using oriented Gabor filters in python.

Answer 3

installation:

pip install fingerprint_enhancer

usage:

import fingerprint_enhancer # Load the library

img = cv2.imread('image_path', 0) # read input image

out = fingerprint_enhancer.enhance_Fingerprint(img) # enhance the fingerprint image

cv2.imshow('enhanced_image', out); # display the result

cv2.waitKey(0)