Bank Card Number Identification Algorithm Based On Gabor Feature

With the rising and development of mobile payments,the APPs that can provide mobile payment function are also increasing rapidly.These APPs require users to bind their bank cards,thus bank card number input has become an indispensable step.However,since bank card numbers usually consist of many numeric characters and are not easy to remember,thus card number manual input process often cause errors and the efficiency is poor.Mobile payment business providers hope to capture the bank card image by the camera of the mobile device,then identify the card numbers automatically by software and input them to the mobile device.This kind of method of inputting card will bring great convenience to people's life.Card number characters of some bank cards are flat and black.The background pattern has no effect on the number area,so eyes can identify the card numbers through chromatism.Card number characters of other bank cards are concavo-convex,formed by embossing machine.Eyes can identify this kind of card numbers through the three-dimensional information and the luminance information.The color of the characters on the concavo-convex bank card are not very different from the background area,thus it is difficultly to separate the characters from the background pattern,which brings great difficulties to the segmentation and identification of card numbers.This thesis studies a bank card number identification method suitable for the concavo-convex character bank cards.First,the tilt correction is carried out by employing the Radon transform on the bank card image.Then,the card number area of the corrected bank card image is roughly extracted according to the apriori position of the card number.Most parts of the background area of the bank card are removed and the card number area is roughly segmented from the card image.Finally,the edge detection is performed on the roughly extracted image area.The accurate area of the card numbers is obtained according to the distribution of pixels in the projection image.Since the color of the concave convex characters is similar to the background color,it is difficult to separate the characters from the background.The background patterns make it harder to separate every single character.So this thesis uses the sliding recognition method to identify the card number.The gray levels of the edges and the backgrounds of the concave convex characters are different.The 2D Gabor filter can be used to extract the width and the texture features of characters from different angles.The 2D Gabor filter has a certain anti-interference ability for background patterns.The gray level image texture features are extracted using 2D Gabor filter for each character from several angles,and then character features are reduced by PCA and LDA.Card number identification is performed by using the feature template matching algorithm based on sliding identification.Finally the identification results are verified through the Luhn algorithm.