Journal of Machine Intelligence and Data Science (JMIDS)

1. Introduction

Fingerprint patterns are distinctive biological characteristics widely employed for personal identification and authentication. They consist of raised ridge structures located on the palmar surface of the fingers and are uniquely individual, even among identical twins. Moreover, fingerprints continue stable through a person’s lifetime, with ridge patterns often becoming more pronounced with age [1]. Due to their uniqueness and permanence, fingerprints are regarded as one of the most reliable biometric identifiers and are widely applied in domains such as law enforcement, security systems, access control, and forensic investigations [2]. Their high accuracy and robustness have sustained fingerprint recognition as a preferred method in applications requiring stringent identity verification standards [3].

Among biometric technologies, fingerprint-based identification stands out as one of the most mature and widely adopted solutions, attracting increasing interest from both academia and industry. A wide range of algorithms and techniques have been proposed to improve recognition performance, which is strongly influenced by the quality of the captured fingerprint images. However, acquiring high-quality fingerprint data remains challenging due to factors such as excessively dry or moist skin, cuts or bruises, uneven finger pressure, and damaged ridge patterns caused by aging or labour-intensive work. According to these issues often result in degraded image quality, ultimately hindering feature extraction and reducing recognition accuracy [4]. Figure 1 illustrates the example of some fingerprint images (a), (b) and (c) before enhancement phase.

To overcome these limitations, fingerprint image enhancement has developed as a critical area of research. Commonly used approaches include Fourier Transform, Wavelet Transform, Histogram Equalization, and Gabor Filtering. Each method offers different advantages and limitations [5].

This study investigates the enhancement of fingerprint image quality using Gabor filtering in combination with adaptive image processing techniques. The proposed method is the adding extra techniques by using Contrast Limited Adaptive Histogram Equalization (CLAHE). The primary objective is to develop an approach capable of improving ridge clarity and feature visibility, thus enhancing the accuracy of subsequent recognition procedures. To quantitatively evaluate enhancement performance, the National Institute of Standards and Technology Fingerprint Image Quality2 (NFIQ2) algorithm is employed to measure post an pre-processing image quality, focusing on clarity, contrast, and feature distinctiveness. Comparative analysis between before and after applied enhancement images is conducted to demonstrate the proposed method’s capability in significantly improving fingerprint image quality.

This paper is organized as follows: Section 2 introduces the related works; Section 3 gives an overview of NFIQ2 quality assessment methodology; Session 4 illustrates the current Enhancement techniques comparison; Section 5 development of the proposed method; Section 6 presents the research methodology; Section 7 the experiments and results. The last session is the conclusions and indicates future research prospects.

2. Related work

Fingerprint recognition stands its efficacy is often challenged by the inherent variability in fingerprint image quality [6]Factors for instance skin conditions or unstable during acquisition process can lead to images with poor contrast, blurred ridges, and gaps, thereby obstructing the accurate extraction of characteristic features [7]. To address these challenges, fingerprint image enhancement techniques are crucial for improving the clarity and continuity of ridge structures, thus helping more reliable matching and proof of identity [9]. There are many enhancement methods for example Histogram equalization, Gabor filters, Wavelet transform and Fourier based approach. Histogram equalization, which apply uniform global contrast adjustments that often amplify both ridges and background noise indiscriminately, leading improves global image contrast [9]. Gabor filters stand out owing to their dual orientation- and frequency-selectivity, which make them remarkably effective at locally enhancing the sinusoidal ridge patterns characteristic of fingerprints [10], [11]. The techniques based on wavelet transforms provide multi-scale decomposition and denoising capability, but without obvious orientation adaptation, their ability to preserve ridge continuity is limited unless combined with additional alignment indications [12], [13]. Fourier-based methods, whether applied to the whole image or in smaller blocks, can strengthen the repeating ridge patterns in fingerprints. However, they cannot localize details well and have difficulty adjusting to changes in ridge direction and curvature [14], [15].

Among the several enhancement techniques, Gabor filtering has occurred as a powerful tool as a result of its ability to selectively enhance specific frequencies and orientations, aligning perfectly with the periodic and directional nature of fingerprint ridges [16]. One more critical technique, Histogram equalization intentions to improve the total contrast of the image by reallocating pixel intensities to utilize the full dynamic range. CLAHE covers this concept by operating on smaller regions of the image and limiting the contrast enhancement to reduce noise amplification, offering a more refined approach to contrast enhancement. By integrating these techniques, a strong fingerprint enhancement pipeline can be developed to significantly improve the performance of fingerprint recognition systems, even under challenging circumstances [17]. The concept of each enchantments describes below.

2.1. Histogram Equalization

Histogram Equalization was developed to overcome the limitations of the standard histogram equalization technique [18]. Instead of calculating the histogram for a whole image, it divides the image in different local regions (filter windows) and calculate histograms for these regions. This makes possible an appropriate contrast enhancement according to the image details in each region. A known disadvantage of

Adaptive Histogram Equalization, though is likelihood of high frequency noise amplification, especially in regions with low contrast, which can create distortion rather than increase clarity. Adaptive Histogram Equalization simply divides image into small blocks (tiles) and then uses histogram equalization to adjust the contrast of the block making it suitable for images containing non-uniform distributions of light [19].

2.2 Gabor Filtering

Gabor filtering is commonly employed in fingerprint image enhancement since, with this method, the filters can be designed to be direction and frequency-selective according to the local ridge direction and scale. This retains the intrinsic fingerprint patterns while enhancing ridge-valley structures and removing background noise, which makes it very helpful in feature extraction and matching with much better image clarity [20]. The key stages of the transformation process are:

Normalization : This step standardizes the grayscale intensity of the fingerprint ridges and valleys to ensure consistent processing in subsequent stages, without altering the fingerprint's ridge structure.

Orientation Estimation : It involves calculating the gradient of pixel intensities in local neighborhoods to determine ridge direction. Accurate orientation estimation enhances both image processing and fingerprint classification.

Region Mask Generation : In this step, the image is segmented into two categories: regions of interest (processable areas) and unrecoverable regions (e.g., areas with excessive noise or missing ridge structures). Only regions of interest are used for further analysis.

Filtering: Gabor filters are applied to the fingerprint image to suppress noise while enhancing the ridge-valley structures. These filters can be tuned for specific frequencies and orientations, making them particularly effective in fingerprint enhancement [21].

2.3 Wavelet Transform

The Wavelet Transform is a powerful method for fingerprint image enhancement, the purpose of this technique is to decompose the image in different levels of resolution, making it a Multi-Scale Representation. This decomposition is beneficial in terms of separating important structural features such as the ridges from those noise high-frequency components [22]. The method clarifies the ridge pattern and definition in the image, by both selectively decreasing noise at certain scales and amplifying ridge related information for other scales. Its improved output will have a clearer and sharper depiction of ridge-valley structures making it better for feature extraction, analysis, and identification in biometric applications.

2.4. Contrast Limited Adaptive Histogram Equalization

This technique is an efficient way to enhance the quality of a fingerprint image based on adaptive contrast enhancement [5] that improves ridge-like structures by locally adjusting intensity levels. This allows for a selective adjustment of the contrast of each region, other regions remain untouched, which increases ridge details without generating too much noise or distortion in the rest of the image, making feature extraction more robust [23]. CLAHE has been used in this study to reduce the impact of low-contrast or uneven illumination factors commonly found in fingerprint images. This approach works clustering the image into non-overlapping untiles.

2.5 Mean and Standard deviation

Quality assessments using the NFIQ2 metric are made on each enhanced fingerprint image after applying each of these all-fingerprint image enhancement techniques. It computes the mean score for quantifying the average quality improvement that each method has brought their consensus, and standard deviation (SD) to evaluate the robustness of using these techniques towards enhancing results across multiple different samples. The low value of SD signals high level stability in the performance. The approach with the highest mean quality score, with a low SD value, is defined as the best performance method and then enhanced for future fingerprint image transformation.

2.5 The percentage of the improvement

The percentage of the improvement be calculated via Eq.1. Percentage Improvement (% Improvement) is a metric used to quantify how much a new value has improved relative to a baseline or original value. A higher percentage Improvement indicates that the applied technique has significantly enhanced performance. Conversely, a low or negative percentage Improvement suggests that the new technique offers minimal benefit or may even degrade performance compared to the original method.

(1)

This paper investigates an alternative approach for fingerprint image enhancement by combining the strengths of Gabor filters, Histogram Equalization, and Contrast Limited Adaptive Histogram Equalization to mitigate the effects of image degradation on recognition performance. This integrated methodology aims to produce well-sharpened fine characteristics and improve the local visibility of the fingerprint image, which is critical for subsequent feature extraction and matching processes.

3. National Institute of Standards and Technology Fingerprint Image Quality 2

This research used the National Institute of Standards and Technology (NIST) Special Database 302 which consist of 90 fingerprint images, resolution 500 PPI (Pixels Per Inch[24]. Figure prints images acquired from the scans of different individuals. This dataset was chosen to confine the adherence of the input data to official international standards, such that additional quality control or improvement experiments could be performed on it afterwards.

Improving fingerprint image quality assessment and raising consistency levels compared to its predecessor is NFIQ2. The NFIQ 2 protocol implemented in C / C++, for all procedures involved in automated fingerprint quality evaluation [25]. The System produces a predictive score of image quality on a scale from 0 to 100 in accordance with ISO / IEC 29794-1:2016 standards. The scale is interpreted as follows: “O ” mean without any usability, the image is so clear visually that parts can be distinguishable. “100 ” is the best for using. From biometric recognition to computational feature extraction, “utility” comprises more than just visual clarity. A grade of “0” points out that image is not usable for biometric workflows, with some parts appearing clear enough.

4.Current Enhancement techniques comparison

The NFIQ2 was used as a benchmark to evaluate image quality before and after enhancement. In this study the different image enchantment method have been conducted to compare the final result of each other. The NFIQ2 score comparison between all the image enhancement, the result illustrated in Table 1.

The table presents a comparative evaluation of four image processing and feature extraction techniques Original image, Adaptive Histogram Equalization, Gabor Filtering, Wavelet Transform, and Fourier Transform, applied across seven test images (denoted as A–G). The numerical values appear to represent quantitative performance measures, possibly accuracy percentages, feature response magnitudes, or classification scores.

Table 1, NFIQ2 Score of the image from difference Enhancement techniques

Image	Original	Histogram	Gabor	Wavelet	Fourier
A	17	12	40	15	9
B	24	32	58	17	8
C	7	22	31	5	7
D	33	37	42	42	16
E	26	22	49	30	8
F	42	45	35	46	16
G	26	10	41	22	8

The "Original" column provides baseline values before the application of enhancement or transformation techniques. Performance varies considerably across images, ranging from 7 (Image C) to 42 (Image F). This variation submits that raw image quality and inherent discriminative features are inconsistent across the dataset, highlighting the necessity of pre-processing or feature extraction methods.

According to the result show that the best overall Performer is Gabor filtering, which demonstrates the highest and most consistent performance, representing its suitability for texture-rich image datasets. Gabor filtering occurs as the most robust approach, while Histogram Equalization and Wavelet Transform exhibit context-dependent strengths. Fourier analysis, although theoretically powerful, appears inadequate for the studied dataset. These findings support the importance of method selection based on image characteristics and application-specific requirements.

5. The proposed methods

This research, the alternative method is introduced and its based-on enhancement of the Gabor filter as the proposed method components illustrated in figure 2. An input images first pass through an adaptive histogram equalization. Local contrast is a technique that changes the local contrast of an image improved, essentially contrasting enhancement. To apply any advanced enhancement techniques some features need to be prominent and hence in this study we have used Contrast Limited Adaptive Histogram Equalization with Contrast enhancement which enhanced the sharpness of image and visibility of fingerprint features. In addition, the Gabor filter is used to capture the ridge-valley patterns of the fingerprint images. The enhanced image, which passed through the proposed method will be appended into process in NFIQ2 to given the score.

This paper will present the experimental result of the enhancement image NFIQ2 score by comparing the Adaptive algorithm with the proposed enhancement method.

6. Methodological Framework

The overall research process is shown in Figure 3, which present the methodology step by step. Fingerprint image enhancement workflow in this research begins with acquisition of the original fingerprint image, followed by initial calculation of its NFIQ2 quality score. A number of techniques are then applied to the source image, including the present algorithm, Wavelet, Gabor filter, Adaptive enhancement and the proposed method. Each technique produces an improved fingerprint image, which is subsequently judged by the NFIQ2 method to give a set of post-enchantment quality scores.

Finally, a comparison of the NFIQ2 scores before and after enhancement can indicate how effective these different methods have been.

7. Experimental Results and Analysis

This section presents the comparative results of the previously discussed enhancement techniques, followed by a detailed analysis of the performance between the Adaptive Histogram method and the proposed system.

Figure 4 illustrated the example of the figure print images before processing by enhancement techniques. These are the examples to visual the result of the improvement image.

7.1 The comparison between previous and adaptive technique.

This experiment demonstrated the comparison between original image, wavelets enchantment, Gabor filter enhancement and the Adaptive algorithm enhancement. The experimental was perform for four example fingerprint images. Each image performs all the enhancement techniques and measure the NFIQ2 score as demonstrated in Table 2. The experimental results indicate that when fingerprint images are processed using the adaptive enhancement algorithm which are based on the Gabor Filtering and combining Adaptive Histogram Equalization, the resulting images yield higher NFIQ2 quality scores compared to those enhanced by other techniques.

Moreover, the percentage of the improvement of each technique have been process as shown in Table 1.

Table 2. The NFIQ2 Score and the percentage of improvement of each technique

Image	Original	Enhancement techniques
		Wavelet	Gabor	Adaptive
Image_01	24	27	58	60
Image_02	27	28	58	60
Image_03	25	31	45	50
Image_04	24	25	55	56
		Percentage of Improvement
Image	Original	Wavelet	Gabor	Adaptive
Image_01	-	12.5	141.7	150.0
Image_02	-	3.7	114.8	122.2
Image_03	-	24.0	80.0	100.0
Image_04	-	4.2	129.2	133.3

After enhancement with the Adaptive algorithm as shown in Figure 5 (i), (j), (k), and (l), the image details are more close-fitting than before. The ridge structures that were previously indistinct have been reconstructed, so the fine detail of the whole fingerprint is a lot clearer. This visual improvement is backed up by their higher NFIQ2 scores, which show the viability of our enrichment method.

From the results, it is evident that the Adaptive method consistently yields the highest NFIQ2 scores across all four images. Figure 6. illustrates the comparative of NFIQ2 scores of four methods: Original, Wavelet Transform, Gabor Filter, and the Adaptive method, evaluated across a dataset of 4 example images.

The Adaptive method consistently achieves higher scores than the other techniques in almost all cases, highlighting its superior enhancement capability and stability. While the Gabor Filter method demonstrates good results for certain images, its performance shows high variability. In contrast, the Wavelet Transform method provides only marginal improvements over the original images. These findings confirm that the Adaptive method offers significant advantages in terms of both average performance and consistency in image processing tasks.

Moreover, Figure.7. demonstrated the comparative analysis of percentage improvement achieved by each technique relative to the original baseline reveals the superior efficacy of the adaptive method across all evaluated cases.

Remarkably, the adaptive algorithm approach demonstrates substantial performance gains in critical instances such as Image_1 and Image_4, exhibiting remarkable improvements of 150.0% and 133.3%, respectively. While the Gabor filter shows significant enhancement in certain images, its performance exhibits considerable variability. The Wavelet Transform, in contrast, yields only modest improvements in numerous scenarios and even results in a performance very small improve in some cases. These findings collectively underscore the consistent and significantly enhanced performance offered by the adaptive method in comparison to other techniques. The results support the efficacy of the adaptive method for improving fingerprint image quality in biometric systems.

7.2 The comparison between adaptive algorithm and the propose system.

This session presents a comparison between the adaptive algorithm which identified in Session 7.1 as the most effective method for enhancing fingerprint images. The proposed method, as previously described, integrates Adaptive Histogram Equalization, the Gabor Filter, and Contrast-Limited Adaptive Histogram Equalization (CLAHE).

Table 2 compares the NFIQ2-based quality improvement results for Adaptive method and Proposed method across all 90 fingerprint images. The results confirm that proposed method achieved the highest mean quality score (53.85) and the greatest percentage improvement (21.73%) after enhancement, outperforming adaptive (Mean: 52.80, Improvement: 19.55%).

Table 2. Comparison of fingerprint image enhancement results using Adaptive method and Proposed method on a total of 90 images.

Method	Mean	SD	Percentage Improvement
Adaptive	52.80	12.62	19.55%
Proposed	53.85	12.00	21.73%

The experimental results indicate that applying Adaptive Histogram Equalization in combination with the Gabor Filter and Contrast Limited Adaptive Histogram Equalization improved the mean fingerprint image quality the mean score to 53.83.

Moreover, the standard deviation (SD) of the quality scores decreased from 21.21 (before enhancement) to 12.62 for adaptive and 12.00 for the proposed method, indicating reduced variability in image quality. This reduction in SD suggests that the enhancement methods not only improved average quality but also produced more consistent results across all images. The overall percentage improvement was 19.55% for Adaptive and 21.73% for Proposed method.

A visual comparison of the enhancement results is provided in Figure 7. Figure 7 (a) shows the original fingerprint image before enhancement. Figure 7(b), processed using Adaptive Histogram Equalization and the Gabor Filter (Adaptive). Figure 7(c), processed using the proposed method of this paper, shows more complete ridge structures, with enhanced continuity and sharpness across the entire fingerprint image.

7.3 Statistical Analysis Using One-Way ANOVA

The hypothesis testing was conducted using Analysis of Variance (ANOVA) to determine whether the differences in mean NFIQ2 scores among the various enhancement techniques were statistically significant. This method was chosen because it allows for comparison of the means from multiple groups while controlling for variability within the data. To examine whether the fingerprint image enhancement methods had a significant effect on the NFIQ2 quality scores, a one-way ANOVA was performed, comparing the following three groups:

• Before Enhancement (original images)
• Adaptive
• Proposed method

The hypotheses were defined as follows:

• H₀: The mean NFIQ2 scores of the three groups (Before Enhancement, Adaptive, and Proposed) are not significantly different.
• H₁: The mean NFIQ2 scores of the three groups are significantly different.

The analysis was conducted at a significance level of α = 0.05. The results are presented in Table 3, which reports the F-value and P-value obtained from the ANOVA test.

Table 3 Analysis of Variance (ANOVA)

Source	DF	Adj SS	Adj MS	F-Value	P-Value
Factor	2	5031	2515.4	10.15	0.000
Error	267	66188	247.9
Total	269	71219

The F-value is 10.15, which is substantially greater than 1, indicating notable variation between groups. Furthermore, the P-value is 0.000, which is less than the significance level α = 0.05. Therefore, the null hypothesis (H₀) is rejected, confirming that there are statistically significant differences in mean NFIQ2 scores among the three enhancement methods.

Moreover, the mean value of the proposed method was the highest at 53.85, indicating that this technique achieved the greatest improvement in image quality compared with Adaptive method (mean = 52.80) and Before Enhancement (mean = 44.24).

In addition, the SD values reflect the stability of the enhancement results. Before Enhancement had the highest SD at 21.39, indicating the greatest variability and least stability in the results. In contrast, Adaptive method and Proposed method achieved lower SD values of 12.62 and 12.00, respectively, demonstrating more consistent and stable outcomes.

The proposed fingerprint enhancement method has strong practical relevance for real-world biometric applications, where degraded or low-quality fingerprints often reduce matching accuracy. By combining Adaptive Histogram Equalization, CLAHE, and Gabor filtering, the method is leading to more reliable automated recognition. However, its reliance on accurate orientation estimation for Gabor filtering may limit performance in extremely noisy or distorted images, and real-time implementation may require computational optimization. These findings highlight both the practical value and theoretical significance of integrating global contrast enhancement with locally adaptive, orientation-specific filtering.

8. Conclusion

This study presented an enhanced fingerprint image processing method that integrates Adaptive Histogram Equalization, Gabor filtering, and Contrast-Limited Adaptive Histogram Equalization. Two variations of the approach Adaptive method and Proposed method, were evaluated against the original (unenhanced) images using the NFIQ2 quality score as the primary performance metric.

The fingerprint images are from the NIST Special Database. The experiment result demonstrated that both adaptive method and the proposed method significantly outperformed the baseline images in terms of clarity, ridge-valley continuity, and overall fingerprint quality. Statistical analysis using one-way ANOVA (F = 10.15, p < 0.001) confirmed that the observed differences in mean NFIQ2 scores among the three groups, Before Enhancement, Adaptive, and Proposed method were statistically significant.

The proposed methods achieved the highest mean NFIQ2 score (53.86) compared to the adaptive method (52.87) and before Enhancement (44.24). It also produced a narrower standard deviation (SD = 12.00) than the baseline (SD = 21.39), indicating greater stability and consistency in enhancement results.

The proposed methods effectively reduced noise, enhanced ridge structures, and filled in minor discontinuities, especially in low-contrast fingerprint images. This improvement makes the enhanced images more suitable for reliable biometric matching and automated identification. For future work, further enhancement can be achieved by integrating multi-frequency filtering methods or advanced image restoration techniques, thereby increasing the applicability and robustness of fingerprint enhancement systems in biometric applications.

Acknowledgements

We would like to express our gratitude to the Department of Industrial Engineering, Faculty of Engineering, Thammasat University for supporting the budget and the laboratory and the case study company that providing the sample products.

References