Fingerprint Enhancement Based on Discrete Cosine Transform

Abstract

This paper proposes a novel fingerprint enhancement algorithm based on contextual filtering in DCT domain. All intrinsic fingerprint features including ridge orientation and frequency are estimated simultaneously from DCT analysis, resulting in fast and efficient implementation. In addition, the proposed approach takes advantage of frequency-domain enhancement resulting in best performance in high curvature area. Comparing with DFT domain, DCT has better signal energy compaction and perform faster transform with real coefficients. Moreover, the experimental results show that the DCT approach is out-performed the traditional Gabor filtering, including the fastest separable Gabor filter, in both quality and computational complexity.

Efficient Fingerprint Recognition Through Improvement of Feature Level Clustering, Indexing and Matching Using Discrete Cosine Transform

“…The energy is one of the image properties using signal processing technique, and it means characteristics of images. From the result of DCT transformed from image F(u,v) of size N x N, the energy using DCT coefficients is defined as following equation [16]. …”

Performance and Computational Complexity Comparison of Block-Based Fingerprint Enhancement

“…However, with low quality fingerprint, the highest peak could not pinpoint. [4], (c) Enhanced results from STFT approach modified from [5] in frequency domain, (d) Enhanced results from DCT approach method [6] in frequency domain…”

“…For DCT approach, coefficient is real and no computation needs because spectrum can be referred directly to its absolute value. However, DCT approach needs an additional phase analysis to classify ridge orientation [6]. This step requires two 2-D filters with size of 5×3 pixels for calculating average directional strength.…”

“…The fingerprint image is partitioned into blocks of 16×16 pixels, and a basic segmentation scheme using mean and variance is employed. Five fingerprint enhancement filtering types are evaluated as follows; Traditional Gabor filtering with non-quantized orientation (TG) [2], Separable Gabor filtering with 8-quantized orientation (SG8) [3], Separable Gabor filtering with non-quantized orientation (SG) [4], Short Time Fourier Transform approach (STFT) [5], and Discrete Cosine Transform approach (DCT) [6]. In the spatial domain approaches, the discrete Gabor filters are the same 25×25 fixed-window size.…”

“…In frequency-domain approaches, Gabor filter can be implemented in Short Time Fourier Transform (STFT) domain, resulting in better enhancement around high curvature zone [5]. Enhancement can be applied in Discrete Cosine Transform (DCT) domain [6]. The STFT and DCT approaches take advantage of frequency domain, which could design arbitrary 2-D filter shape, corresponding to improve quality of fingerprint in high curvature zones.…”