Font Size: a A A

Research Of Finger Vein Feature Extraction Method Based On Beyond Wavelets Transform

Posted on:2014-03-20Degree:MasterType:Thesis
Country:ChinaCandidate:Z TianFull Text:PDF
GTID:2268330425966721Subject:Pattern Recognition and Intelligent Systems
Abstract/Summary:PDF Full Text Request
Feature extraction is a critical step for the statistical recognition system based on theimage statistical recognition system. At present, there are mainly four methods for featureextraction of the image characteristics, and the feature extraction based on the objectfrequency domain transform coefficients is very strongly to represent object character; thusthis paper concerns the finger vein image, the finger vein image were operated bypreprocessing firstly, and then the preprocessed image was applied by beyond wavelettransform to extract the feature of the image.During preprocessing the finger vein image, the acquired finger vein image wasperformed by grayscaling operation; and then the finger region of the grayscaled image waspositioned and captured, thereby the interested region of the finger vein image was obtainedby the above operation. In order to verify the fact that there are lots of obvious directioninformation in the obtained finger region through observation intuitively, this paper calculatedthe direction graph of the finger vein region; thus the method of directional filtersegmentation was performed next to obtain the vein texture of finger region; during theprocess of acquiring the finger vein image, because of the difference of the size of the finger、the placement and so on, it is necessary to perform normalization operation on the finger veinimage above operated.Considering that the wavelet analysis theory dealing with the two-dimensional and abovefunction possessing the singularity shows unsatisfactory results, the beyond wavelet analysistheory is proposed. In order to verify the superiority of the beyond wavelet transform in thefeature extraction aspect, this paper performed the feature extraction based on the wavelettransform, the wavelet transform decomposing coefficients was obtained trough applying thewavelet transform on the finger vein image processing above, Principal Component Analysisand Wavelet Moment Analysis were applied in the low-frequency part of the waveletcoefficients, and at the same time, the statistical analysis were applied in the horizontal,vertical and diagonal direction of the wavelet transform decomposing coefficients, then thefeature for finger vein recognition was obtained. In this paper, the low frequency coefficientsand the horizontal, vertical, diagonal direction high frequency coefficients of the wavelettransform decomposition, being applied Principal Component Analysis, was input into SVM(Support Vector Machine) classifier to identify it, this paper compared the recognitionperformance superior of the four categories of coefficients; there are the coefficient’s dimension by calculating the recognition rate reaching maximum, considering thesedecomposition coefficients’ identifying performance, the feature constructor method wasproposed. Finally, the superiority of the feature extraction method based on the wavelettransform was validated by compared experiments.This paper mainly perform Ridgelet Transform and Curvelet Transform in the beyondwavelet transform to extract the feature of the finger vein image. During the feature methodbased on the ridgelet transform, since the ridgelet transform is recognized as the wavelettransform in the Radon domain, therefore the finger vein image was performed by theapproximate discrete Radon transform; one-dimensional discrete wavelet transform wasrespectively performed on the approximate discrete Radon transform image according to thepolar coordinate, the Ridgelet transform coefficients of the finger vein image were obtained,the coefficients calculated by principal component analysis were input into the NearestNeighbor(NN) Classifier to classify and recognition them, this paper analyze the classificationand recognition results and proposed the method of constructing the feature based on theridgelet decomposition coefficients; the experimental results shows that the feature extractionbased on the ridgelet transform is better than that based on the wavelet transform. During thecurvelet transform method, this paper described two kinds of discrete algorithm based on thesecond generation curvelet transform, comparing the advantages of these two algorithms, thealgorithm of USFFT was selected to process the finger vein image; the finger vein imagecalculated by the selected algorithm were given, this paper performed the different scalecoefficients calculated by principal component analysis respectively put into the SVM andNN to obtain their recognition rate, the related experiment analyze and verified the classifyingand recognizing the ability of the coefficients of the various scale, according to the analysisresults, this paper constructed the feature based on the curvelet transform, at last, theconstructed curvelet feature was input into the SVM classifier for classification andrecognition.
Keywords/Search Tags:Finger vein recognition, Feature extraction, Wavelet transform, Ridgelettransform, Curvelet transform
PDF Full Text Request
Related items