Fingerprint Compression Based on Sparse Representation
Transcription
Fingerprint Compression Based on Sparse Representation
IJSART - Volume 1 Issue 4 –APRIL 2015 ISSN [ONLINE]: 2395-1052 Fingerprint Compression Based on Sparse Representation GavitJayesh1, JadhavUmesh2, SawantTushar3, Latkar Mahesh4 1, 2, 3, 4 Department of Computer Engineering K.K.Wagh Institute of Engineering, Education & Reaserch Abstract- Another unique finger impression pressure calculation taking into account scanty representation is resented. Getting an over complete lexicon from an arrangement of unique finger impression patches permits us to speak to them as a scanty straight blend of word reference iotas. In the calculation, we first build a lexicon for predefined finger impression picture patches. For another given unique mark pictures, speak to its fixes as per the lexicon by figuring l0-minimization and after that quantize and encode the representation. In this paper, we consider the impact of different elements on pressure results. Three gatherings of unique mark pictures are tried. The investigations exhibit that our calculation is effective contrasted and a few contending pressure systems (JPEG, JPEG 2000, and WSQ), particularly at high pressure degrees. The tests additionally delineate that the proposed calculation is earty to concentrate detail. Keywords- Fingerprint, compression, sparse representation, JPEG 2000, JPEG, WSQ, PSNR. I. INTRODUCTION Recognition persons by method for biometric attributes is an imperative innovation in the general public, since biometric identifiers can't be imparted and they naturally speak to the individual's substantial personality. Among numerous biometric distinguishment advances, unique finger impression distinguishment is exceptionally mainstream for individual distinguishing proof because of the uniqueness, all inclusiveness, collectability and invariance. Extensive volumes of unique finger impression are gathered and put away every day in an extensive variety of uses, including crime scene investigation and access control. In 1995, the measure of the FBI unique finger impression card chronicle contained more than 200 million things and file size was expanding at the rate of 30 000 to 50 000 new cards every day. Vast volume of information devour the measure of memory. Unique mark picture pressure is a key method to illuminate the issue. By and large, pressure innovations can be classed into lossless and lossy. Lossless pressure permits the careful unique pictures to be reproduced from the compacted nformation. Lossless pressure advancements are utilized as a part of situations where it is essential that the unique and the decompressed information are indistinguishable. Maintaining a strategic Page | 36 distance from bending confines their pressure productivity. At the point when utilized in picture pressure where slight bending is worthy, lossless pres0sure advancements are frequently utilized in the yield coefficients of lossyressure. Lossy pressure advancements typically change a picture into an alternate area, quantize and encode its coefficients. Amid the most recent three decades, change based picture pressure advancements have been broadly scrutinized and a few benchmarks have showed up. Two most basic choices of change are the Discrete Cosine Transform (DCT) also, the Discrete Wavelet Transform (DWT). The DCT-based encoder can be considered pressure of a surge of 8 × 8 little square of pictures. This change has been embraced in JPEG. The JPEG pressure plan has numerous favorable circumstances, for example, effortlessness, comprehensiveness and accessibility. Be that as it may, it has an awful execution at low bit-rates essentially in view of the basic square based DCT plan. Thus, as ahead of schedule as 1995, the JPEG-council started to build up another wavelet-based pressure standard for still pictures, in particular JPEG 2000. The DWT-based calculations incorporate three stages: a DWT processing of the standardized picture, quantization of the DWT coefficients and lossless coding of the quantized coefficients. The point of interest can be found in and.Contrasted and JPEG, JPEG 2000 gives numerous highlights that bolster adaptable and intelligent access to expansive measured picture. It additionally permits extraction of diverse resolutions, pixel devotions, districts of investment, segments and and so forth. There are a few other DWT-based calculations, for example, Set Apportioning in Hierarchical Trees (SPIHT) Algorithm . The above calculations are for general picture pressure. Focused at unique mark pictures, there are exceptional pressure calculations. The most widely recognized is Wavelet Scalar Quantization (WSQ). It turned into the FBI standard for the pressure of 500 dpi unique mark pictures. Motivated by the WSQ calculation, a couple of wavelet parcel based finger impression pressure plans have been created. Notwithstanding WSQ, there are other calculations for finger impression pressure, for example, Contourlet Change (CT). These calculations have a typical inadequacy, specifically, without the apacity of learning. The finger impression pictures can't be compacted well at this point. They won't be compacted well later. In this paper, a novel methodology in light of inadequate representation is www.ijsart.com IJSART - Volume 1 Issue 4 –APRIL 2015 given. The proposed system has the capacity by overhauling thword reference. . The particular procedure is as per the following: build a base network whose segments speak to highlights of the finger impression pictures, alluding the framework lexicon whose segments are called particles; for a given entire unique finger impression, separate it into little squares called patches whose number of pixels are equivalent to the measurement of the particles; utilize the strategy for inadequate representation to acquire the coefficients; then, quantize the coefficients; last, encode the coefficients and other related data utilizing lossless coding techniques. In many examples, the assessment of pressure execution of the calculations is confined to Peak Signal to Noise Degree (PSNR) reckoning. The consequences for real unique finger impression coordinating or distinguishment are not researched. In this paper, we will mull over it. In most Automatic Fingerprint recognizable proof System (AFIS), the fundamental highlight used to match two unique mark pictures are particulars (edges endings and bifurcations). Subsequently, the distinction of the particulars between preand post-pressure is considered in the paper. II. RELATED WORK The field of scanty representation is moderately youthful. Early indications of its center thoughts showed up in a spearheading work. In that paper, the creators presented the idea of lexicons also, set forward a percentage of the center thoughts which later got to be crucial in the field, for example, a voracious interest method. From that point, S. S. Chen, D. Donoho and M. Saunders presented an alternate interest method which utilized l 1-standard for scanty. It is astounding that the best possible arrangement frequently could be gotten by unraveling a raised programming assignment. Since the two original works, specialists have contributed an awesome arrangement in the field. The action in this field is spread over different disciplines. There are as of now numerous effective applications in different fields, for example, face distinguishment, picture denoising , object location and superdetermination picture reproduction . In paper ,the creators proposed a general characterization calculation for article distinguishment in light of a meager representation registered by l 1-minimization. On one hand, the calculation in light of inadequate representation has a superior execution than different calculations, for example, closest neighbor, closest subspace also, direct SVM; then again, the new system gave new bits of knowledge into face distinguishment: with sparsity legitimately saddled, the decision of highlights gets to be less vital than the quantity of highlights. Without a doubt, this sensation is regular in the fields of inadequate representation. It doesn't just exist in the face distinguishment, additionally shows up in different circumstances. In paper taking into Page | 37 ISSN [ONLINE]: 2395-1052 account inadequate and repetitive representations on overcomplete word reference, the creators composed a calculation that could evacuate the zero-mean white and homogeneous Gaussian added substance clamor from a given picture. In this paper, we can see that the substance of the word reference is of significance. The significance is epitomized in two viewpoints. On one hand, the word reference ought to effectively mirror the substance of the pictures; then again, the word reference is sufficiently vast that the given picture can be spoken to inadequately. These two focuses are completely indispensable for the routines in view of scanty representation. Meager representation has officially a few applications in picture pressure . In paper , the analyses demonstrate that the proposed calculation has great execution. Nonetheless, its pressure effectiveness is reliably lower than JPEG 2000's. On the off chance that more general characteristic pictures are tried, this marvel will be more evident that the pressure productivity is lower than the cutting edge pressure advances. In this paper, we demonstrate the finger impression pictures can be compacted better under an over-complete lexicon in the event that it is legitimately built.In paper , the creators proposed a calculation of finger impression pressure in view of Nonnegative Matrix Factorization (NMF). In spite of the fact that NMF has some fruitful applications, it likewise has inadequacies. In some cases, non-antagonism is a bit much. Case in point, in the picture pressure, what is considered is the manner by which to lessen the contrast between preand postpressure as opposed to nonnegativity. Moreover, we think the techniques in light of meager representation don't work exceptionally well in the general picture pressure field. The reasons are as per the following: the substance of the general pictures are rich to the point that there is no legitimate lexicon under which the given picture can be spoken to scantily; regardless of the fact that there is one, the extent of the word reference may be too huge to be processed successfully. Case in point, the deformity, turn, interpretation and the clamor all can make the lexicon gotten to be too substantial. Accordingly, inadequate representation ought to be utilized in exceptional picture pressure field in which there are no above deficiencies. The field of unique finger impression picture pressure is one of them. III. THE MODEL AND ALGORITHMS OF SPARSE REPRESENTATION A. The Model of Sparse Representation Given A = [a1, a2,..., A ] ∈ RM×N , any new test y ∈ RM×1, is thought to be spoken to as a straight mix of few segments from the word reference An, as indicated in recipe (1). This is the main former information about the word reference in our www.ijsart.com IJSART - Volume 1 Issue 4 –APRIL 2015 ISSN [ONLINE]: 2395-1052 calculation. Later, we will see the property can be guaranteed by building the lexicon legitimately. y = Ax……… (1) where y ∈RM×1, A ∈ RM×N and x = [x1, x2,..., xN ] T∈ RN×1. Clearly, the framework y = Ax is underdetermined when M < N. Accordingly, its answer is not one of a kind. As indicated by the supposition, the representation is scanty. A legitimate arrangement can be gotten by taking care of the accompanying enhancement issue: (l0) : min II xII0s.t. Ax = y ………..(2) Arrangement of the advancement issue is relied upon to be extremely scanty, in particular, IIx0II << N. The documentation IIx0II checks the nonzero entrances in x. Really it is not a standard. Then again, without uncertainty, despite everything we call it l0-standard. Actually, the pressure of y can be attained to by packing x. In the first place, record the areas of its non-zero entrances and their sizes. Second, quantize and encode the records. This is the thing that we will do. Next, procedures for illuminating the enhancement issue are given. Fig. The behavior of xp for various values of p. As p tends to zero, approaches the l-norm. Algorithm-Fingerprint compression algorithm based on sparse representation i. B. Meager Solution by Greedy Algorithm ii. Analysts' first believed is to tackle the streamlining issue l0 straightforwardly. The Matching Pursuit (MP) on account of its effortlessness and proficiency is frequently used to pretty nearly illuminate the l0 issue. Numerous variations of the calculation are accessible, offering upgrades either in exactness or/and in many-sided quality. Despite the fact that the hypothetical investigation of these calculations is troublesome, examinations demonstrate that they act well when the quantity of non-zero passages is low. iii. C. Meager Solution by l1-Minimization It is a characteristic thought that the enhancement issue (2) can be approximated by tackling the accompanying enhancement issue: (lp) : min IIxIIps.t. Ax = y…………. (3) where p > 0 Clearly, the littler p is, the closer the arrangements of the two enhancement issues l,0 and l p are, as shown in Fig. 1. This is because the magnitude of x is not important when p is very small. What does matter is whether x is equal to 0 or not. Therefore, p is theoretically chosen as small as possible. However, the optimization problem (3) is not convex if 0 <p< 1. It makes p= 1 the most ideal situation, namely, the following problems. (l Page | 38 1) : min x 1 s.t. Ax = y…………... (4) iv. To construct a base matrix whose columns represent features of the fingerprint images, referring the matrix dictionary whose columns are called atoms For a given whole fingerprint divide it into small block called patches For all patches no. of pixels is equal to the dimensions of the atoms To use the method of sparse representation: Obtain t he coefficients quantize coefficients Encode the coefficients and other related information lossless coding methods. IV. OUR WORK We create a dictionary which is neatly maintain and fingerprint images are divided into patches. A. Construct dictionary The first method: choose fingerprint patches from thetraining samples at random and arrange these patches as columns of the dictionary matrix.The second method: in general, patches from foreground of a fingerprint have an orientation while the patchesfrom the background don’t have, as shown in Fig. 2. This fact can be used to construct the dictionary. Divide the interval [00, . . . ,1800] into equal-size intervals. Each interval isrepresented by an orientation (the middle value of eachinterval is chosen). Choose the same number of patchesfor each interval and arrange them into the dictionary. www.ijsart.com IJSART - Volume 1 Issue 4 –APRIL 2015 ISSN [ONLINE]: 2395-1052 first index and other indexes are coded by the same arithmetic encoder. In the following experiments, the first coefficient is quantized with 6 bits and other coefficients are quantized with 4 bits. V. CONCLUSION Fig. fingerprint divided into patches and store in sparse format B. Compression of a Given Fingerprint Given a new fingerprint, slice it into square patches which have the same size with the training patches. The size of the patches has a direct impact on the compression efficiency.The algorithm becomes more efficient as the size increases. However, the computation complexity and the size of the dictionary also increase rapidly. The proper size should be chosen. The size of the patches are defined as per the size of database. In addition, to make the patches fit the dictionary better, the mean of each patch needs to be calculated and subtracted from the patch. After that, compute the sparse representation for each patch by solving the l0 problem. Those coefficients whose absolute values are less than a given threshold are treated as zero. For each patch, four kinds of information need to be recorded. They are the mean value, the number about how many atoms to use, the coefficients and their locations. The tests show that many image patches require few coefficients. Consequently, compared with the use of a fixed number of coefficients, the method reduces the coding complexity andimproves the compression ratio. Another pressure calculation adjusted to unique finger impression pictures is presented. Notwithstanding the straightforwardness of our proposed calculations, they contrast positively and existing more modern calculations, particularly at high pressure degrees. Due to the square by-piece preparing system, on the other hand, the calculation has higher complexities. The examinations demonstrate that the piece impact of our calculation is less genuine than that of JPEG. We consider the impact of three separate word references on unique finger impression pressure. The examinations mirror that the lexicon got by the K-SVD calculation works best. Also, the bigger the quantity of the preparing set is, the better the pressure result is. One of the principle troubles in creating pressure calculations for fingerprints lives in the requirement for protecting the particulars which are utilized as a part of the ID. The examinations demonstrate that our calculation can hold a large portion of the particulars vigorously amid the pressure and remaking. There are numerous charming inquiries that future work ought to consider. To begin with, the highlights and the techniques for developing word references ought to be thoroughly considered. Furthermore, the preparation tests ought to incorporate fingerprints with diverse quality ("great", "awful", "monstrous"). Thirdly, the improvement calculations for unraveling the meager representation need to be examined. Fourthly, streamline the code to diminish unpredictability of our proposed technique. At long last, different applications in light of meager representation for unique mark pictures ought to be investigated. ACKNOWLEDGMENT C. Coding and Quantization Entropy coding of the atom number of each patch, the mean value of each patch, the coefficients and the indexes is carried out by static arithmetic coders. The atom number of each patch is separately coded. The mean value of each patch is also separately coded. The quantization of coefficients is performed using the Lloyd algorithm, learnt off-line from the coefficients which are obtained from the training set by the MP algorithm over the dictionary. The first coefficient of each block is quantized with a larger number of bits than other coefficients and entropy-coded using a separate arithmetic coder. The model for the indexes is estimated by using the source statistics obtained off-line from the training set. The Page | 39 The authors would like to thank the Associate Editor and the anonymous reviewers for their valuable comments and suggestions that helped greatly improve the quality of this paper. REFERENCES [1] D. Maltoni, D. Miao, A. K. Jain, and S. Prabhakar, Handbook ofFingerprint Recognition, 2nd ed. London, U.K.: Springer-Verlag, 2009. [2] N. Ahmed, T. Natarajan, and K. R. Rao, “Discrete cosine transform,”IEEE Trans. Comput., vol. C-23, no. 1, pp. 90–93, Jan. 1974. www.ijsart.com IJSART - Volume 1 Issue 4 –APRIL 2015 [3] C. S. Burrus, R. A. Gopinath, and H. Guo, Introduction to Wavelets andWavelet Transforms: A Primer. Upper Saddle River, NJ, USA: Prentice-Hall, 1998. [4] W. Pennebaker and J. Mitchell, JPEG—Still Image Compression Standard.New York, NY, USA: Van Nostrand Reinhold, 1993. [5] M. W. Marcellin, M. J. Gormish, A. Bilgin, and M. P. Boliek,“An overview of JPEG-2000,” in Proc. IEEE Data Compress. Conf.,Mar. 2000, pp. 523–541. [6] A. Skodras, C. Christopoulos, and T. Ebrahimi, “The JPEG 2000 stillimage compression standard,” IEEE Signal Process. Mag., vol. 11, no. 5,pp. 36–58, Sep. 2001. [7] T. Hopper, C. Brislawn, and J. Bradley, “WSQ gray-scale fingerprintimage compression specification,” Federal Bureau of Investigation,Criminal Justice Information Services, Washington, DC, USA, Tech.Rep. IAFIS-IC0110-V2, Feb. 1993. [8] C. M. Brislawn, J. N. Bradley, R. J. Onyshczak, and T. Hopper, “FBIcompression standard for digitized fingerprint images,” Proc. SPIE,vol. 2847, pp. 344–355, Aug. 1996. [9] A. Said and W. A. Pearlman, “A new, fast, and efficient image codecbased on set partitioning in hierarchical trees,” IEEE Trans. CircuitsSyst. Video Technol., vol. 6, no. 3, pp. 243–250, Jun. 1996. [10] R. Sudhakar, R. Karthiga, and S. Jayaraman, “Fingerprint compressionusing contourlet transform with modified SPIHT algorithm,” IJECEIranian J. Electr. Comput. Eng., vol. 5, no. 1, pp. 3–10, 2005. [11] S. G. Mallat and Z. Zhang, “Matching pursuits with timefrequencydictionaries,” IEEE Trans. Signal Process., vol. 41, no. 12,pp. 3397–3415, Dec. 1993. [12] S. S. Chen, D. Donoho, and M. Saunders, “Atomic decomposition bybasis pursuit,” SIAM Rev., vol. 43, no. 1, pp. 129–159, 2001. [13] J. Wright, A. Y. Yang, A. Ganesh, S. S. Sastry, and Y. Ma, “Robust facerecognition via sparse representation,” IEEE Trans. Pattern Anal. Mach.Intell., vol. 31, no. 2, pp. 210–227, Feb. 2009. [14] M. Elad and M. Aharon, “Image denoising via sparse and redundantrepresentation over learned dictionaries,” IEEE Trans. Image Process.,vol. 15, no. 12, pp. 3736–3745, Dec. 2006. [15] S. Agarwal and D. Roth, “Learning a sparse representation for objectdetection,” in Proc. Eur. Conf. Comput. Vis., 2002, pp. 113–127. [16] J. Yang, J. Wright, T. Huang, and Y. Ma, “Image superresolutionas sparse representation of raw image patches,” in Proc. IEEE Conf.Comput. Vis. Pattern Recognit., Jun. 2008, pp. 1–8. [17] K. Skretting and K. Engan, “Image compression using learned dictionariesby RLS-DLA and compared with KSVD,” in Proc. IEEE ICASSP,May 2011, pp. 1517–1520. [18] O. Bryt and M. Elad, “Compression of facial images using theK-SVD algorithm,” J. Vis. Commun. Image Represent., vol. 19, no. 4,pp. 270–282, 2008. Page | 40 ISSN [ONLINE]: 2395-1052 [19] Y. Y. Zhou, T. D. Guo, and M. Wu, “Fingerprint image compressionalgorithm based on matrix optimization,” in Proc. 6th Int. Conf. DigitalContent, Multimedia Technol. Appl., 2010, pp. 14–19. [20] A. J. Ferreira and M. A. T. Figueiredo, “Class-adapted image compressionusing independent component analysis,” in Proc. Int. Conf. ImageProcess., vol. 1. 2003, pp. 625–628. [21] A. J. Ferreira and M. A. T. Figueiredo, “On the use of independentcomponent analysis for image compression,” Signal Process., ImageCommun., vol. 21, no. 5, pp. 378– 389, 2006. [22] P. Paatero and U. Tapper, “Positive matrix factorization: A nonnegativefactor model with optimal utilization of error estimates of data values,”Environmetrics, vol. 5, no. 1, pp. 111–126, 1994. [23] D. D. Leeand and H. S. Seung, “Learning the parts of objects by nonnegativematrix factorization,” Nature, vol. 401, pp. 799–791, Oct. 1999. www.ijsart.com