This is to confirm that the work contained in the thesis entitled A DoG based Approach for Fingerprint Image Enhancement by Shila Samantaray is a report of an original research work carried out under our supervision and guidance in partial fulfillment of the requirements for the award of the degree of Master of Technology in Computer Science and Engineering. Ashok kumar Turuk, for providing support in improving the quality of work in designing this thesis. This thesis focuses on fingerprint enhancement techniques using histogram equalization applied locally to the degraded image.
Biometrics is a method for uniquely identifying individuals based on one or more inherent physiological or behavioral characteristics of the subject.
Biometric Identification and Fingerprint
Finally, the ridge pattern of the fingerprint image can be so short that it even forms a small island. A genetic code in DNA gives general instructions about the way skin should form in a developing fetus, but the specific way it forms is a result of chance events (ie the precise position of the fetus in the womb on a specific moment and the exact composition and density of surrounding amniotic fluid).
Thesis Organization
Fingerprint images are rarely of perfect quality, for reasons such as variations in impression condition, skin condition, scanning equipment, or may be due to the uncooperative attitude of the subject. This degraded image quality can result in creating a significant number of false details and ignoring true details. A vital step in the study of fingerprint statistics is reliable fingerprint feature extraction from fingerprint images.
It is therefore important to apply image enhancement techniques prior to detail extraction in order to obtain a large number of reliable estimates of detail locations. The main objective of fingerprint image enhancement is to enhance the edge characteristics of the image, as these edges carry the information of characteristics required for detail extraction. In a well-defined fingerprint image, the ridges and valleys should ideally alternate and flow in a locally costly direction.
These regularities facilitate the detection of ridges and, in turn, enable accurate extraction of details from thinned ridges [11]. Thus, the corruption or noise needs to be reduced by image enhancement techniques to obtain an improved definition of ridges versus pits in fingerprint images.
Literature Review
The use of fingerprints to identify individuals has been in practice since the late nineteenth century, when Sir Francis Galton defined some of the characteristics from which fingerprints can be used for identification. Results of the technique worked better for improvement and the average error percentage in terms of dropped and false minutiae produced by the proposed approach is significantly lower. During the binarization phase, direct binarization of the image with its orientation field is applied instead of binarization by means of global thresholding.
During segmentation he used a method based on variance thresholding to separate the foreground region of the image from the background regions. 14]in the same year have proposed a mean and standard deviation of the fingerprint image to extract the region of interest which has the advantage of ease of calculation. In 2004 Sen Wang and Yangsheng Wang [15] approached bandpass filters to remove unwanted noise in fingerprint images.
Its disadvantage is that it only focuses on a single point of the image. 17] introduces a new approach for fingerprint enhancement based on Short Time Fourier Transform (STFT) analysis, and the result of the method depends on the choice of window (i.e., 12 × 12 window) to resolve image features in both space and frequency. During the enhancement in the first stage, the author used a raised cosine window that outlines the image and is responsible for calculating all the intrinsic images (i.e., images that represent the important features of the fingerprint image as a fingerprint pixel map) and in the second stage, the image is divided to the overlapping windows and a bandpass Butterworth filter is applied to each window to reconstruct the enhanced image by distributing the enhancement results of each local window.
Problem Statement
It is observed that among the enhancement techniques, histogram-based equalization method is used in various modified ways, which represents the contrast enhancement that gives the best visual performance under certain conditions, but sometimes shows the fatal error of over-enhancement. Keeping the shortcomings in mind, the present research is about the scope of improvements of the histogram equalization based enhancement for fingerprint images to detect the correct number of minutiae points compared to the existing enhancement techniques. In the thesis, emphasis is placed on the implementation of improvement techniques, i.e. instead of applying the enhancement algorithm directly to the input image, the technique transforms the input image and decomposes it into a number of bandpass images. Then, the Gaussian low-pass filtering is applied to obtain Laplacian images from the decomposed images. Finally, the reconstruction is performed considering the highest decomposed layer of the image and the output images of decomposed images applied with difference of Gaussian. The enhancement is followed by the binarization and thinning methods to prepare the fingerprint images for the extraction phase. The operational productivity of the automated fingerprint identification system has boosted individual areas of the fingerprint identification system to provide better performance in defined problem areas.
A fingerprint image does not prove to be a complete image for direct procedures that could be applied to them, such as ridge feature extraction and pattern analysis.
Proposed HE Based Enhancement
Methodology
- DoG based Approach
- Normalization
- Orientation Estimation
- Ridge Frequency Estimation
- Gabor Filtering
- Binarization
- Thinning
The first section deals with higher-order bandpass image enhancement when the original input has to be passed through the Gaussian filter. The algorithm in its first step applies a histogram to the input image to see the brightness range of the image. Now to suppress the quantum jump in the brightness level of the Histogram, clipping is applied.
A global clipping that clips the areas above half the peak, and a local clipping that clips obliquely the areas below half the peak. Second part of the proposed algorithm 3.2. In this part, the details are improved to overcome the inherent gaps of HE-based improvement. Normalization is applied to standardize the intensity value of the image to a pre-specified mean and a pre-specified variance.
Smoothing the orientation field with Gaussian filter gives the final result of the orientation estimation. The local frequency of the ridges is calculated for fingerprint images. The normalized image is divided into blocks of size W ×W. The ridge continuity only gives a smooth ridge frequency, but the discontinuity (i.e. the existence of ridge termination or bifurcations), where small points occur in the block, does not produce a well-defined sun-shaped waveform.
Experimental Results
- DoG based Approach
- Normalization
- Orientation Estimation
- Gabor Filtering
- Binarization and Thinning
Here, a step-by-step implementation of Algorithm 3.1 and subsequent improvement in terms of the graph is made. The resulting image combines the result of contrast enhancement and detail enhancement as described in Algorithm 3.2. The disadvantages of the HE-based methods were in the quantum jump of the luminance value in the pixel, which is changed in the proposed scheme by using the clipping technique during the first section. A distinct result can be observed between the original input histogram and the contrast-enhanced image.
Each fine-enhanced image is normalized to obtain a normalized image with a predefined mean and variance for all images. In order to obtain a smooth orientation through the Gabor filter, the normalized images must be applied with a ridge orientation filter. It can be seen how well the ridge filter has calculated the ridge current pattern, even though there are invisible pattern currents.
The middle part of the image processing is Gabor filtering, which redefines the ridge orientations and ridge frequency, these are calculated in advance. The purpose of the filtering stage is to increase the clarity of the ridge structures and reduce the noise. After all the preprocessing, the fingerprint image is binarized and thinned to prepare it for minutiae extraction.
Minutiae Extraction
Like a ridge pixel with a junction number of one corresponds to the end of the ridge, and a CN of three corresponds to a bifurcation, as shown in fig:cn. a) CN of one, i.e. of the final Ridge pixel. For a candidate ridge endpoint, first the eight connected ridge endpoints are marked with a value of 1, the next step is to count the 0-to-1 transitions at the boundary of the eight. Similarly, if the passage shows 3, it is a ridge bifurcation. a) Input fingerprint image (b) Enhanced fingerprint image.
Global clipping is the clipping of those bins that have more than half the peak value, i.e. p(k) and. A contribution is made to developing an efficient improvement technique that brings out well-improved back patterns. The blurred images or only images with contrast enhancement are insufficient to provide good details about the fingerprint patterns. Therefore, the proposed image enhancement algorithm is presented together with the fingerprint image processing steps.
The algorithm has successfully shown that the number of details generated from the proposed DoG-based approach is relatively less and it has provided a remarkable difference compared to Hong-based enhancement technique. Various methods in the public domain for enhancing fingerprint images have been reviewed and a new method that provides better performance is proposed. To avoid spurious detail generation, the proposed scheme follows A new robust HE algorithm and a modified Laplacian pyramid approach proposed here improve the image quality to overcome the problem of poor enhancement.
To conclude with this thesis, the proposed approaches have been studied and observed with some limitations. The matching part was not included in the research work, which could show the results in a better way.