Folded bill identification method and device

A method for recognizing a folded bill, performed by a folded bill recognizing device, comprising:

• obtaining
• an infrared transmission image Ts and an infrared reflection image Fs of a to-be-recognized bill;
• performing high-pass filtering on the infrared transmission image Ts to obtain a high-pass infrared transmission filtering image gTs;
• performing low-pass filtering on the infrared transmission image Ts to obtain a low-pass infrared transmission filtering image dTs;
• performing high-pass filtering on the infrared reflection image Fs, to obtain a high-pass infrared reflection filtering image gFs, wherein the high-pass filtering on the infrared reflection image Fs and the low-pass filtering on the infrared transmission image Ts are performed synchronously according to a geometric coordinate mapping relationship;
• performing low-pass filtering on the infrared reflection image Fs, to obtain a low-pass infrared reflection filtering image dFs, wherein the low-pass filtering on the infrared reflection image Fs and the high-pass filtering on the infrared transmission image Ts are performed synchronously according to the geometric coordinate mapping relationship;
• performing a differential operation on the high-pass infrared reflection filtering image gFs and the low-pass infrared transmission filtering image dTs to obtain a differential image cFTs;
• performing a first characteristic extraction on the high-pass infrared transmission filtering image gTs by calculating an average gray value gT_Gs of the gTs as a characteristic value;
• performing a second characteristic extraction on the low-pass infrared reflection filtering image dFs by calculating an average gray value dF_Gs of the dFs as a characteristic value;
• performing a third characteristic extraction on the differential image cFTs by calculating an average gray value cFT_Gs of the cFTs as a characteristic value;
• substituting the characteristic value gT_Gs, the characteristic value dF_Gs and the characteristic value cFT_Gs respectively into three models y1,y2,y3 for distinguishing folded bills and non-folded bills,
• y1=f1(gT_G);
• y2=f2(dF_G)
• y3=f3(cFT_G)
• to obtain
• p1=f1(gT_Gs);
• p2=f2(dF_Gs)
• p3=f3(cFT_Gs);
• wherein p1, p2 and p3 are confidence levels for determining the to-be-recognized bill as a folded bill;
• f1, f2 and f3 indicate learnt multi-characteristic classifying probability distribution models;
• in a case that p1>T1, p2>T2, p3>T3 are all true, the to-be-recognized bill is recognized as a folded bill;
• in a case that p1>T1, p2>T2, p3>T3 are not all true, the to-be-recognized bill is recognized as a non-folded bill, where T1, T2 and T3 are three confidence level thresholds.