Unionpay Logos Recognition Methods By 2D-compressed Measurement And FDDL Classification Based On The Sample Size Difference

Nowadays, as the information and the number of images increase significantly, identifying the image accurately, effectively and quickly has become a difficult problem. By using traditional features for huge amounts of samples, distinguishing different features become harder with the number and sub kinds of images increaseing, memory consumption becomes bigger and recognition rate drops greatly.Using images features basing on the sparse representation has reduced memory by increasing the computational complexity.This paper proposed a two-dimensional CS(2D-CS) measurement feature based on Compressed Sensing observation targeting the identification of unionpay logos, and proposed the new Bags Of Words(BOW) to choose the number of words denpending on the sample size.Meanwhile, with 2D-CS features removed the background interference, the difference between features and the recognition performance could be improved by the use of the algorithm of Fisher Discrimination Dictionary Learning(FDDL).1, Two-dimensional CS measurement feature based on Compressed Sensing(CS) observation are proposed. The Non-separable BOW of 2D-CS texture feature is constructed to improved the difference between samples. For different samples size of unionpay symbols,realize the weigted or direct fusion using the spatio-temporal field features, measurement features of 1D-CS, measurement features of the Non-separable 2D-CS, measurement features of 2D-CS with words to adjustable sample size because single features are with effvective information The results of the experiment in the paper could be divided into two respects. 1) when the sample size is small, recognition rate of the single Non-seperable 2D-CS texture feature increases by 6% compared with that of the 1D-CS features, while recognition rate of the fusion feature including the Non-seperable 2D-CS texture feature increases by 2% compared with that of the single BOW of 2D-CS texture feature. 2) when the sample size is large, recognition rate of the single BOW of 2D-CS texture feature increases by 9.84% and 12.85% compared with that of the 1D-CS features and the canny features respectively, while recognition rate of the fusion feature including the BOW of 2D-CS texture feature increases by 0.8% compared with that of the single BOW of 2D-CS texture feature.2, The selecting rule of global or local FDDL basing on the different sample size of unionpay logo is proposed. The rule is based on the difference of 2D-CS feature of unionpay symbols without background and amount of features data to selectparameters of FDDL. The aims are to reduce the amount of data and improve the effective difference between features. The results showed in the paper could be divided into two respects.1)when the sample size is small, in single features, recognition rate of the Global Classifier of FDDL(GC-FDDL) of the BOW of 2D-CS texture feature whose number of words is variable and 2D-CS feature of unionpay logo without background increases by 3% and 4% compared with that of the traditional single CS features respectively, while recognition rate of GC-FDDL of HSV-HOG in fusion features increases by 11% compared with that of the single features;2) when the sample size is large,recognition rate of the Local Classifier of FDDL(LC-FDDL) of the BOW of 2D-CS texture feature whose number of words is variable and 2D-CS feature of unionpay logo without background increases by 2.02% and 11.11% compared with that of the traditional single CS features respectively.