FINGERPRINT IMAGE QUALITY ESTIMATION USING STATISTICAL & GRAPHI

Abstract

The use of biometrics is evolving day by day in our society. Fingerprint recognition is well known for its high acceptability and popularity in the world of biometric systems. An ideally sensed or scanned fingerprint image has clear and distinct ridges and valleys. An automatic fingerprint recognition system can perform well on such fingerprint images. However, precise fingerprint acquisition has some peculiar and challenging aspects. Often skin condition or imperfect acquisitions cause the captured fingerprint image to be far from ideal. Unclean sensor plates, non-uniform and inconsistent contact can result in poor samples and feature extraction artifacts during image processing and hence increase false accept/reject rate. It is desirable to assess “quality” of a fingerprint before any matching process. This allows poor image acquisition to be corrected before poor quality is entered into users’ databases. This means presenting the matcher with good quality fingerprint images will result in high matcher performance, and vice versa, the matcher will perform poorly for poor quality fingerprints. Moreover it is fruitless effort to apply matching techniques on poor quality image. It will be wastage of time, effort and resources. We have purposed a new methodology to estimate the quality of fingerprint. ‘Graphical Representation’ as well as ‘Statistical Measures’ will be carried out to analyze the quality of input image. Graphical Representation assists in global features extraction. Entropy , Uniformity , Smoothness , PCA Components (Latent , T-square ,Score ) & Spectral Analysis constitute the graphical representation portion of our methodology . Whereas on the other hand local feature extraction is carried out by Mean, Standard Deviation, Variance, Dry Percentage, Humidity Percentage, Background Pixel Percentage, Foreground Pixel Percentage, and Quality Index, Mean to Standard Deviation Ratio, Average Gray Level to Variance Ratio, Uniformity, Smoothness & In-homogeneity factors. Global features extraction analyze the overall image quality but Local features extraction process the image in depth at block level in effort to estimate quality. Threshold value has been set for each factor and input image is classified after comparison from predefined threshold. Good quality images require minor preprocessing and enhancement. Bad quality image (dry or wet) requires different preprocessing & enhancement techniques. The purpose of this research is that to estimate the quality of image before matching process starts. By achieving this we not only save precious time & effort but also maintain the integrity of our database and achieve high performance. Difference of our methodology with previous work is that most of already existing technologies for said propose relies on complex mathematical model that requires much knowledge of mathematics. Our image quality estimatator has been implemented in MATLAB and tested on Database DB1 of FVC 2002 which contains 800 images. Results from experiments clearly depict the effectiveness of our proposal.