CN104658097A - RMB paper currency denomination identification method based on histogram matching of images - Google Patents

Abstract

The invention provides an RMB paper currency denomination identification method based on histogram matching of images. The method is applied to RMB paper currency denomination identification. The training process comprises the following steps: acquiring a two-dimensional matrix of image training samples; filtering each image training sample; extracting a first histogram vector of a red-green-blue three-color component of each image training sample; and respectively acquiring the average value of the histogram of each color component in the multiple image training samples to serve as a template of the paper currency with the denomination. The identification process specifically comprises the following steps: acquiring a two-dimensional matrix of to-be-identified images; filtering the to-be-identified images; extracting a second histogram of the red-green-blue three-color component with an identification image, respectively acquiring the distance between the second histogram and the template of the paper currency of each denomination, comparing the distances, and judging the denomination which corresponds to the template of the paper currency with the denomination with the smallest distance away from the histogram as the denomination of the to-be-identified paper currency. According to the method provided by the invention, the noise robustness is guaranteed, the identification precision is greatly improved, and the identification time is short.

Description

A kind of rmb paper currency denomination identifying method of the Histogram Matching based on image

Technical field

The present invention relates to financial field, particularly relate to a kind of rmb paper currency denomination identifying banknote denomination recognition methods of the Histogram Matching based on image.

Background technology

Along with the prosperity and development of economy, the circulation of bank note is increasing, and the core technology of the multi-optical spectrum paper money counting machine that domestic many banks use, cleaning-sorting machine, ATM is all from abroad, and not only expensive and existence jeopardizes the hidden danger of financial security.

The core technology basis that multi-optical spectrum paper money counting machine, cleaning-sorting machine, ATM want is real-time banknote image process and identification.At home, the image denomination recognition technology that multi-optical spectrum paper money counting machine adopts, adopts manual features local more, due to be difficult to avoid image shift and get the limited problem of false distinguishing feature, its stability and recognition capability are all difficult to reach requirement.

Because bank note is that other technical difficult points is: recognition speed requires that high, characteristics of image is difficult to extract, and while raising recognition capability, also must ensure the robustness of algorithm to banknote image.Therefore work out good stability, recognition efficiency rmb paper currency recognition methods that is high, that can carry out global recognition is necessary.

Summary of the invention

The invention provides a kind of rmb paper currency denomination identifying method of the Histogram Matching based on image, it is according to the coupling of three Color Histograms of different denominations image, identified the denomination of bank note by the tolerance of proper vector, ensure that banknote denomination recognition speed improves the robustness of image to noise while high with this.

The rmb paper currency denomination method of the Histogram Matching based on image provided by the invention, its technical scheme is as follows:

The rmb paper currency denomination method of this Histogram Matching based on image, by the high-velocity scanning of multi-optical spectrum paper money counting machine, obtain the common N comprising often kind of same denominations image and open image training sample, by carrying out medium filtering to often opening training sample, to be reduced in the noise from scanister in image acquisition process, then, its histogram being extracted to red, green, blue three colouring component of the every pictures obtained, the banknote image histogram of same denomination being got average as training the standard form obtained; Finally, the sample of the bank note of denomination to be identified is mated with standard form, identify the real denomination of bank note sample to be identified; Concrete steps are as follows:

One, the dependent image data of each training sample denominations is obtained

S11 obtains the two-dimensional matrix of the image training sample of multiple this kind of training sample denominations;

S12 carries out filtering to often opening image training sample described in training sample;

S13 extracts the histogram vectors of often opening red, green, blue three colouring component of image training sample described in training sample, comprises red component first histogram vectors H _r, green component first histogram vectors H _g, and blue component first histogram vectors H _b;

S14 obtains the histogrammic mean value of often kind of color component in multiple training sample image described respectively as the standard form of this kind of denominations, comprises red component standard form T (H _r), green component standard form T (H _g), and blue component standard form T (H _b), wherein;

Two, denominations image real time transfer to be identified is carried out

S21 obtains the two-dimensional matrix of the image to be identified of described denominations to be identified;

S22 carries out filtering to described denominations image to be identified;

S23 extracts the histogram of the redgreenblue component of described image to be identified, comprises red component second party figure vector h _r, green component second histogram vectors h _g, and blue component second histogram vectors h _b;

Three, the real denomination of described denominations to be identified is confirmed

Obtain described second histogram and the distance described in each between denominations template (distance) respectively:

$\mathrm{d\; is}\tan \mathrm{ce}=\underset{1\≤k\≤M}{\min }\sqrt{\frac{1}{3}\underset{j=0}{\overset{255}{\mathrm{\Σ}}}{{(h_r\left(j\right)-T\left(H_r\right)\left(j\right))}^2+\frac{1}{3}\underset{j=0}{\overset{255}{\mathrm{\Σ}}}{(h_g\left(j\right)-T\left(H_g\right)\left(j\right))}^2+\frac{1}{3}\underset{j=0}{\overset{255}{\mathrm{\Σ}}}(h_b\left(j\right)-T\left(H_b\right)\left(j\right))}^2};$

And the distance between described second Histogram distance obtained and each denomination training sample bank note template is compared, namely denomination corresponding to the denominations template minimum apart from described second histogram be judged to be the denomination of described bank note to be identified.

Be applied to banknote denomination identification, the banknote denomination said here comprises: 100 yuan, 50 yuan, 20 yuan, 10 yuan, 5 yuan, and 1 yuan.

Preferably, in step S11 and step S21, the high-speed scanning device in multi-optical spectrum paper money counting machine is used to obtain the two-dimensional matrix of image training sample and the two-dimensional matrix of image to be identified respectively.

Preferably, in step S12 and step S22, respectively medium filtering is carried out to image training sample and image to be identified.

Preferably, in step S12 and step S22,3 × 3 windows are adopted to carry out medium filtering to image training sample and image to be identified respectively.

Preferably, in step s 24 which, specifically comprise:

The round values corresponding with the lowest distance value of described denominations template according to described second histogram judges the denomination of bank note described to be identified corresponding with it.

Histogram matching for identifying rmb paper currency denomination provided by the invention, at least can bring following beneficial effect:

According to the intrinsic image attributes of the bank note of different denomination, namely the different denomination to bank note of histogrammic distribution identifies, efficiently solve and traditional identify by selected characteristic the drawback that the method for banknote denomination is brought, compared with the recognition methods in traditional banknote denomination, method provided by the invention ensure that substantially increasing accuracy of identification while noise robustness and recognition time is short, well meets the demand that the finance devices such as paper money counter, cleaning-sorting machine, ATM run identification fast.

Accompanying drawing explanation

Fig. 1 is the control flow chart based on the rmb paper currency denomination identifying method of the Histogram Matching of image in the present invention;

Fig. 2 is the concrete steps process flow diagram of the template obtaining each denominations in the present invention;

Fig. 3 is medium filtering 3 × 3 window schematic diagram in the present invention;

Fig. 4 a is the standard form schematic diagram of 100 yuans of red components in the present invention;

Fig. 4 b is the standard form schematic diagram of 100 yuans of green components in the present invention;

Fig. 4 c is the standard form schematic diagram of 100 yuans of blue components in the present invention.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is described in further detail:

In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, below in conjunction with drawings and Examples, the present invention is specifically described.Accompanying drawing in the following describes is only some embodiments of the present invention.For those of ordinary skill in the art, under the prerequisite not paying creative work, other accompanying drawing can also be obtained according to these accompanying drawings.

As shown in Figure 1, the invention provides a kind of rmb paper currency denomination identifying method of the Histogram Matching based on image, based on principle be being made up of RGB three kinds of basic colors components of image, as long as we extract the histogram vectors of often opening three kinds of basic colors components of bank note bank note to be identified, and mate with trained bank note template, the denomination of bank note to be identified can be identified.It should be noted that, the present invention is applicable to all current paper moneys in circulation on the market, wherein, Renminbi comprises 100 yuan, 50 yuan, 20 yuan, 10 yuan, 5 yuan and 1 yuan, as long as have demand in actual applications, be equally applicable to the identification of the bank note of foreign different denomination, here we are not specifically limited it.

In the method for the Histogram Matching based on image provided by the invention, first we need to train banknotes of different denominations image to be trained in training storehouse, to compare as the image of template and bank note to be identified.Concrete steps comprise:

As shown in Figure 2, the concrete steps obtaining the template of each denominations comprise:

Specifically, here, we can be obtained N and open image training sample by the high-speed scanning device in multi-optical spectrum paper money counting machine.In a specific embodiment, can for the bank note of different denomination, the image training sample of selected some respectively, as, the image training sample of often kind of banknote denomination chooses 10, to ensure the precision of follow-up identification bank note to be identified, certainly, we do not do concrete restriction to the quantity of image training sample, although it is more that the quantity of image training sample is chosen, the degree of accuracy identified is higher, but mean that the time of carrying out training sample also can lengthen, thus can select according to the actual requirements, image training sample as often kind of banknote denomination chooses 5, 15, 20, 25 etc., even when not high to precise requirements, a training sample as image only chosen by bank note for a certain denomination, as long as the object of current application can be reached, all be included in content of the present invention.

S12 carries out filtering to often opening image training sample;

After acquiring the image training sample of some, namely we carry out filtering to often opening training sample, in the present invention, in order to improve the robustness to noise, carry out medium filtering to often opening image training sample, particularly, in the present invention, as shown in Figure 3,3 × 3 windows are used to carry out medium filtering to image training sample image, wherein, p ₀, p ₁, p ₂, p ₃, p ₄, p ₅, p ₆, p ₇, and p ₈represent respectively to should filter window pixel value and arrange according to size, then P ₄=med{P ₀, P ₁, P ₂, P ₃, P ₄, P ₅, P ₆, P ₇, P ₈be the result of medium filtering.The medium filtering in the medium filtering in red component image, the medium filtering in green component image, blue component image is obtained respectively after medium filtering.

S13 extracts first histogram vectors of often opening the redgreenblue component of image training sample, comprises red component first histogram vectors H _r, green component first histogram vectors H _g, and blue component first histogram vectors H _b;

S14 obtains the standard form of histogrammic mean value as this kind of denominations of often kind of color component in multiple image training samples respectively, comprises red component template T (H _r), green component template T (H _g), and blue component template T (H _b), wherein;

$\begin{array}{ccc}T\left(H_r\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{H}_{\mathrm{ri}}& T\left(H_g\right)=\frac{1}{N}\underset{i=0}{\overset{N}{\mathrm{\Σ}}}{H}_{\mathrm{gi}}& T\left(H_b\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{H}_{\mathrm{bi}}\end{array}$

Wherein, N represents the quantity of often kind of banknote image training sample, completes the training process to training sample image by above-mentioned steps, and obtains the denominations template of often kind of denominations.As Fig. 4 a, 100 yuans are respectively through training the red component module map (horizontal ordinate i represents the scope 0-255 of the pixel value of image) obtained shown in Fig. 4 b and Fig. 4 c, green component Prototype drawing and blue component Prototype drawing (horizontal ordinate i represents the scope 0-255 of the pixel value of image), as the benchmark (horizontal ordinate i represents the scope 0-255 of the pixel value of image) of identification 100 yuans.

The concrete steps of carrying out denominations image real time transfer to be identified comprise:

S21 obtains the two-dimensional matrix of the image to be identified of denominations to be identified, and to obtain in template procedure identical for expression and the training sample image of two-dimensional matrix here, and therefore not to repeat here.

S22 treats recognition image and carries out filtering.Similar, here, we adopt the method for medium filtering to carry out filtering to its image to be identified, and therefore not to repeat here.

S23 extracts the histogram of the redgreenblue component of image to be identified, comprises red component histogram vectors h _r, green component histogram vectors h _g, and blue component histogram vectors h _b.

Confirm that the concrete steps of the real denomination of described denominations to be identified are as follows:

Calculate the distance (distance) between image histogram to be identified and each denominations template respectively, i.e. the Euclidean distance (scope of pixel is 0-255) of sample image three Color Histogram to be identified and template samples three Color Histogram:

$\mathrm{d\; is}\tan \mathrm{ce}=\underset{1\≤k\≤M}{\min }\sqrt{\frac{1}{3}\underset{j=0}{\overset{255}{\mathrm{\Σ}}}{{(h_r\left(j\right)-T\left(H_r\right)\left(j\right))}^2+\frac{1}{3}\underset{j=0}{\overset{255}{\mathrm{\Σ}}}{(h_g\left(j\right)-T\left(H_g\right)\left(j\right))}^2+\frac{1}{3}\underset{j=0}{\overset{255}{\mathrm{\Σ}}}(h_b\left(j\right)-T\left(H_b\right)\left(j\right))}^2};$

Compared apart from the distance between each denominations and training sample denominations template by image histogram to be identified, namely the denomination corresponding apart from minimum denominations template be judged to be the denomination of bank note to be identified.

In this step, it should be noted that in a particular embodiment, before the denomination identifying bank note to be identified, first we be numbered from big to small according to banknote denomination, as: 100 yuan of corresponding sequence numbers are 1,50 yuan of corresponding sequence numbers are 2, other the like, M represents the kind of denomination.Thus, in the process identified, when we obtain the minor increment between each denominations template of the second Histogram distance, round values corresponding to this minor increment judges the denomination of bank note to be identified corresponding with it.

Be described in detail the specific embodiment of invention above, but the present invention is not restricted to specific embodiment described above, it is just as example.To those skilled in the art, any equivalent modifications that this system is carried out and substituting also all among category of the present invention.Therefore, equalization conversion done under the spirit and scope not departing from invention and amendment, all should contain within the scope of the invention.

Claims (6)

1. the rmb paper currency denomination identifying method based on the Histogram Matching of image, be applied to banknote denomination identification, it is characterized in that: by the high-velocity scanning of multi-optical spectrum paper money counting machine, obtain the common N comprising often kind of same denominations image and open image training sample, by carrying out filtering to often opening training sample, to be reduced in the noise from scanister in image acquisition process; Then, its histogram being extracted to red, green, blue three colouring component of the every pictures obtained, the banknote image histogram of same denomination being got average as training the standard form obtained; Finally, the sample of the bank note of denomination to be identified is mated with standard form, identify the real denomination of bank note sample to be identified; Concrete steps are as follows:

One, the dependent image data of each training sample denominations is obtained

S11 obtains the two-dimensional matrix that N opens the image training sample of this kind of training sample denominations;

S12 carries out filtering to often opening image training sample described in training sample;

S13 extracts the histogram vectors of often opening red, green, blue three colouring component of image training sample described in training sample, comprises red component first histogram vectors H _r, green component first histogram vectors H _g, and blue component first histogram vectors H _b;

S14 obtains the histogrammic mean value of often kind of color component in multiple training sample image described respectively as the standard form of this kind of denominations, comprises red component standard form T (H _r), green component standard form T (H _g), and blue component standard form T (H _b), wherein N represents the quantity of training sample;

$T\left(H_r\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{H}_{\mathrm{ri}},$

$T\left(H_g\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{H}_{\mathrm{gi}},$

$T\left(H_b\right)=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{H}_{\mathrm{bi}},$

Two, denominations image real time transfer to be identified is carried out

S21 obtains the two-dimensional matrix of the image to be identified of described denominations to be identified;

S22 carries out filtering to described denominations image to be identified;

S23 extracts the second histogram of the redgreenblue component of described band recognition image, comprises red component second party figure vector h _r, green component second histogram vectors h _g, and blue component second histogram vectors h _g;

Three, the real denomination of described denominations to be identified is confirmed

Obtain described second histogram and the distance described in each between denominations template (distance) respectively:

$\mathrm{dis}\tan \mathrm{ce}=\underset{1\≤k\≤M}{\min }\sqrt{\frac{1}{3}\underset{j=0}{\overset{255}{\mathrm{\Σ}}}{(h_r\left(j\right)-T\left(H_r\right)\left(j\right))}^2+\frac{1}{3}\underset{j=0}{\overset{255}{\mathrm{\Σ}}}{(h_g\left(j\right)-T\left(H_g\right)\left(j\right))}^2+\frac{1}{3}\underset{j=0}{\overset{255}{\mathrm{\Σ}}}{(h_b\left(j\right)-T\left(H_b\right)\left(j\right))}^2};$

And the distance between described second Histogram distance obtained and each denomination training sample bank note template is compared, namely denomination corresponding to the denominations template minimum apart from described second histogram be judged to be the denomination of described bank note to be identified.

2. the rmb paper currency denomination identifying method of a kind of Histogram Matching based on image as claimed in claim 1, it is characterized in that, described banknote denomination comprises: 100 yuan, 50 yuan, 20 yuan, 10 yuan, 5 yuan, and 1 yuan.

3. the rmb paper currency denomination identifying method of a kind of Histogram Matching based on image as claimed in claim 1, it is characterized in that: in step S11 and step S21, use the high-speed scanning device in multi-optical spectrum paper money counting machine to obtain the two-dimensional matrix of image training sample and the two-dimensional matrix of image to be identified respectively.

4. the rmb paper currency denomination identifying method of a kind of Histogram Matching based on image as claimed in claim 1, is characterized in that: in step S12 and step S22, carries out medium filtering respectively to image training sample and image to be identified.

5. the rmb paper currency denomination identifying method of a kind of Histogram Matching based on image as claimed in claim 4, is characterized in that: in step S12 and step S22, adopts 3 × 3 windows to carry out medium filtering to image training sample and image to be identified respectively.

6. a kind of histogram matching based on image as claimed in claim 1, is characterized in that: in step s 24 which, specifically comprises:

The round values corresponding with the lowest distance value of described denominations template according to described second histogram judges the denomination of bank note described to be identified corresponding with it.