CN103400161A - Handwritten numeral recognition method and system - Google Patents

Abstract

The invention provides a handwritten numeral recognition method and a handwritten numeral recognition device. The method comprises the following steps of: receiving a handwritten numeral image input by a user; extracting M covariance characteristics of the handwritten numeral image, wherein the value of M is any one positive integer which is greater than 1; and recognizing handwritten numerals according to the preset M covariance characteristics of each training image, a preset type tag carried by each training image and the M covariance characteristics of the handwritten numeral image. The handwritten numerals are recognized by extracting the M covariance characteristics of the handwritten numeral image, wherein the value of the M is any one positive integer which is greater than 1, so that the problem of inaccurate recognition of the handwritten numerals caused by adoption of a single covariance characteristic in the recognition process of the handwritten numerals in the prior art is solved.

Description

A kind of Handwritten Numeral Recognition Method and system

Technical field

The application relates to digital recognition technology field, particularly relates to a kind of Handwritten Numeral Recognition Method and system.

Background technology

Along with the develop rapidly of computer technology and digital image processing techniques, the Handwritten Digital Recognition technology is at extensive data statistics in recent years, and mail sorts, and finance have all been carried out application widely in the fields such as the tax and finance.The Handwritten Digital Recognition technology is as a major issue of area of pattern recognition, important theory value is also arranged, often represent accurate numerical value in every field due to numeral, small mistake is probably brought unpredictable consequence, therefore the accuracy of handwriting digital recognition technology is had high requirement.

in prior art, often by a Lie group k nearest neighbor sorter, handwriting digital is identified, Lie group k nearest neighbor sorter is as characteristics of image with the covariance of area image, then by this covariance, handwriting digital is identified, but in the process of the covariance with area image as characteristics of image, the structural scheme of covariance has various ways, namely there is multiple covariance feature, and each Lie group k nearest neighbor sorter corresponding a kind of covariance feature only, make prior art in the process of carrying out Handwritten Digital Recognition, because can only adopt single covariance feature, cause Handwritten Digital Recognition inaccurate.

Summary of the invention

In view of this, the embodiment of the present application provides a kind of Handwritten Numeral Recognition Method and system,, to solve prior art in the process of carrying out Handwritten Digital Recognition,, because can only adopt single covariance feature, causes the inaccurate problem of Handwritten Digital Recognition.

To achieve these goals, the technical scheme that provides of the embodiment of the present application is as follows:

A kind of Handwritten Numeral Recognition Method comprises:

Receive the handwriting digital image of user's input;

Extract the M kind covariance feature of described handwriting digital image, and the M kind covariance feature of described handwriting digital image is corresponding one by one with the M kind covariance feature of each training image that sets in advance respectively; The value of M is any one positive integer greater than 1;

The class label entrained according to the M kind covariance feature of each the described training image that sets in advance, each described training image of setting in advance and the M kind covariance feature of described handwriting digital image are identified described handwriting digital.

Preferably, the setting up procedure of the M kind covariance feature of described each training image that sets in advance specifically comprises:

Receive at least one training image of user's input;

Extract respectively the M kind covariance feature of each described training image; The value of M is any one positive integer greater than 1.

Preferably, the described training image of each that sets in advance all carries class label;

The class label that the M kind covariance feature of each described training image that described basis sets in advance, each the described training image that sets in advance carry and the M kind covariance feature of described handwriting digital image are identified described handwriting digital, specifically comprise:

Calculate respectively the distance between the covariance feature of every kind of covariance feature of described handwriting digital image and each the described training image that sets in advance corresponding with every kind of covariance feature of described handwriting digital image, for every kind of covariance feature in described handwriting digital image, generate a distance set; The all corresponding training image that sets in advance of each element in described distance set;

Respectively to the element in each described distance set according to sorting from small to large;

Receive neighbour's number K of user's input; The value of K is any one positive integer more than or equal to 1;

According to described neighbour's number K, obtain respectively front K element in each described distance set, and according to neighbour's set of front K Element generation in each distance set;

Each described neighbour is merged, generate final neighbour's set;

According to the entrained class label of the training image that sets in advance corresponding to each element in described final neighbour's set, described handwriting digital is identified.

Preferably, the corresponding unique class label of each described handwriting digital;

Describedly according to the entrained class label of the training image that sets in advance corresponding to each element in described final neighbour set, described handwriting digital is identified, is specifically comprised:

Obtain the entrained class label of the training image that set in advance corresponding with each element in described final neighbour set;

Respectively the quantity of each class label in all described class labels that get is added up, the class labels that class label quantity is maximum are as the purpose class label;

Search the numeral corresponding with described purpose class label as purpose numeral according to described purpose class label in the corresponding relation of the class label that sets in advance and numeral;

Described purpose numeral is identified as handwriting digital.

Preferably, the M kind covariance feature of the described handwriting digital image of described extraction specifically comprises:

Extract respectively the M kind pixel characteristic φ that each point (x, y) on each described handwriting digital image is located _j(I, x, y), j=1,2...M, wherein, I represents handwriting digital image, φ _jThe j kind pixel characteristic that each point (x, y) of (I, x, y) expression handwriting digital image is located;

According to the M kind pixel characteristic that each point (x, y) on described handwriting digital image is located, extract respectively the M kind covariance feature of described handwriting digital image, computing formula is as follows:

$C^j=\frac{1}{\mathrm{mn}}\underset{x=1}{\overset{m}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}({\mathrm{\φ}}_j(I,x,y)-{\stackrel{\‾}{\mathrm{\φ}}}_j\left(I\right)){({\mathrm{\φ}}_j(I,x,y)-{\stackrel{\‾}{\mathrm{\φ}}}_j\left(I\right))}^T,j=\mathrm{1,2}...M$

Wherein,

Pixel characteristic average for the handwriting digital image I.

Preferably, the described M kind covariance feature that extracts respectively each described training image; The value of M is any one positive integer greater than 1, specifically comprises:

Extract respectively the M kind pixel characteristic φ that each point (x, y) on each described training image is located _j(I _i, x, y), j=1,2...M, wherein, I _iRepresent i training image, i=1,2...N, φ _j(I _i, x, y) and the j kind pixel characteristic located of each point (x, y) of i training image of expression;

The M kind pixel characteristic that each point (x, y) on training image according to each is located, extract respectively the M kind covariance feature of each described training image, and computing formula is as follows:

$C_i^j=\frac{1}{\mathrm{mn}}\underset{x=1}{\overset{m}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}({\mathrm{\φ}}_j(I_i,x,y)-{\stackrel{\‾}{\mathrm{\φ}}}_j\left(I_i\right)){({\mathrm{\φ}}_j(I_i,x,y)-{\stackrel{\‾}{\mathrm{\φ}}}_j\left(I_i\right))}^T,j=\mathrm{1,2}...M,i=\mathrm{1,2}...N$

Wherein,

For training image I _iThe pixel characteristic average.

Preferably, the M kind covariance feature of described handwriting digital image is corresponding one by one with the M kind covariance feature of each described training image respectively, specifically comprises:

The extracting mode of every kind of covariance feature of described handwriting digital image is identical one by one with the extracting mode of every kind of covariance feature of each described training image respectively.

A kind of device for Identification of Handwritten Numerals comprises: handwriting digital image receiving unit, handwriting digital image covariance feature extraction unit and Handwritten Digital Recognition unit, wherein,

Described handwriting digital image receiving unit is used for receiving the handwriting digital image of user's input;

Described handwriting digital image covariance feature extraction unit is connected with described handwriting digital image receiving unit, be used for extracting the M kind covariance feature of described handwriting digital image, and the M kind covariance feature of described handwriting digital image is corresponding one by one with the M kind covariance feature of each training image that sets in advance respectively; The value of M is any one positive integer greater than 1;

Described Handwritten Digital Recognition unit is connected with described handwriting digital image covariance feature extraction unit, is used for the entrained class label of the M kind covariance feature according to each the described training image that sets in advance, each described training image of setting in advance and the M kind covariance feature of described handwriting digital image described handwriting digital is identified.

Preferably, the described training image of each that sets in advance all carries class label, described Handwritten Digital Recognition unit comprises: metrics calculation unit, sequencing unit, neighbour's number receiving element, neighbour gather generation unit, final neighbour gathers generation unit and Handwritten Digital Recognition subelement, wherein

Described metrics calculation unit is connected with described handwriting digital image covariance feature extraction unit, be used for calculating respectively the distance between the covariance feature of every kind of covariance feature of described handwriting digital image and each the described training image that sets in advance corresponding with every kind of covariance feature of described handwriting digital image, for every kind of covariance feature in described handwriting digital image, generate a distance set; The all corresponding training image that sets in advance of each element in described distance set;

Described sequencing unit is connected with described metrics calculation unit, is used for respectively element to each described distance set according to sorting from small to large;

Described neighbour's number receiving element is used for receiving neighbour's number K of user's input; The value of K is any one positive integer more than or equal to 1;

The end that the neighbour gathers generation unit is connected with described sequencing unit, the other end is connected with described neighbour's number receiving element, be used for according to described neighbour's number K, obtain respectively front K element in each described distance set, and according to neighbour's set of front K Element generation in each distance set;

Described final neighbour gathers generation unit and gathers generation unit with described neighbour and be connected, and is used for each described neighbour is merged, and generates final neighbour and gathers;

Described Handwritten Digital Recognition subelement is gathered generation unit with described final neighbour and is connected, be used for the entrained class label of the training image that sets in advance corresponding to each element according to described final neighbour's set, described handwriting digital is identified.

Preferably, the corresponding unique class label of each described handwriting digital; Described Handwritten Digital Recognition subelement comprises: class label acquiring unit, purpose class label determining unit and purpose numeral determining unit and recognition unit, wherein,

Described class label acquiring unit is gathered generation unit with described final neighbour and is connected, and is used for obtaining the training image that the set in advance entrained class label corresponding with each element of described final neighbour's set;

Described purpose class label determining unit is connected with described class label acquiring unit, be used for respectively the quantity of each class label of all described class labels of getting is added up, the class labels that class label quantity is maximum are as the purpose class label;

Described purpose numeral determining unit is connected with described purpose class label determining unit, is used for searching the numeral corresponding with described purpose class label as the purpose numeral according to described purpose class label at the class label that sets in advance and the corresponding relation of numeral;

Described recognition unit is connected with described purpose numeral determining unit, is used for described purpose numeral is identified as handwriting digital.

this shows, the application provides a kind of Handwritten Numeral Recognition Method and device, and the method is by receiving the handwriting digital image of user's input, extract the M kind covariance feature of handwriting digital image, the value of M is any one positive integer greater than 1, the M kind covariance feature of last each training image according to setting in advance, the M kind covariance feature of the class label that each training image that sets in advance is entrained and handwriting digital image is identified handwriting digital, the application identifies handwriting digital by the M kind covariance feature that extracts the handwriting digital image, wherein, the value of M is any one positive integer greater than 1, avoided prior art in the process of carrying out Handwritten Digital Recognition, because can only adopt single covariance feature, cause the inaccurate problem of Handwritten Digital Recognition.

Description of drawings

In order to be illustrated more clearly in the embodiment of the present application or technical scheme of the prior art, below will the accompanying drawing of required use in embodiment or description of the Prior Art be briefly described, apparently, the accompanying drawing that the following describes is only some embodiment that put down in writing in the application, for those of ordinary skills, under the prerequisite of not paying creative work, can also obtain according to these accompanying drawings other accompanying drawing.

A kind of Handwritten Numeral Recognition Method process flow diagram that Fig. 1 provides for the embodiment of the present application one;

A kind of Handwritten Numeral Recognition Method process flow diagram that Fig. 2 provides for the embodiment of the present application two;

The entrained class label of the training image that sets in advance corresponding to each element in the final neighbour's set of a kind of basis that Fig. 3 provides for the embodiment of the present application two, the method flow diagram that handwriting digital is identified;

The method flow diagram of the setting up procedure of the M kind covariance feature of a kind of each training image that sets in advance that Fig. 4 provides for the embodiment of the present application three;

A kind of concrete grammar process flow diagram that extracts the M kind covariance feature of each training image that Fig. 5 provides for the embodiment of the present application three;

A kind of concrete grammar process flow diagram that extracts the M kind covariance feature of described handwriting digital image that Fig. 6 provides for the embodiment of the present application four;

The structural representation of a kind of device for Identification of Handwritten Numerals that Fig. 7 provides for the embodiment of the present application five;

The structural representation of a kind of device for Identification of Handwritten Numerals that Fig. 8 provides for the embodiment of the present application six;

The detailed construction schematic diagram of a kind of Handwritten Digital Recognition subelement that Fig. 9 provides for the embodiment of the present application six;

Figure 10 for the embodiment of the present application seven provide along with parameter K changes situations to numeral 1,7 and 9 classification;

Figure 11 for the embodiment of the present application seven provide along with parameter K changes situations to numeral 1,2,7 and 9 classification;

Figure 12 for the embodiment of the present application seven provide along with parameter K changes situation to digital 0-9 classification.

Embodiment

In order to make those skilled in the art person understand better technical scheme in the application, below in conjunction with the accompanying drawing in the embodiment of the present application, technical scheme in the embodiment of the present application is clearly and completely described, obviously, described embodiment is only the application's part embodiment, rather than whole embodiment.Based on the embodiment in the application, those of ordinary skills are not making under the creative work prerequisite the every other embodiment that obtains, and all should belong to the scope of the application's protection.

Embodiment one:

A kind of Handwritten Numeral Recognition Method process flow diagram that Fig. 1 provides for the embodiment of the present application one.

As shown in Figure 1, the method comprises:

The handwriting digital image of S101, reception user input.

In the embodiment of the present application, the user can input handwriting digital by hand-written mode on flat board, handset touch panel, and at first the method receives the user inputs on the instruments such as handset touch panel or flat board handwriting digital image.

The M kind covariance feature of S102, extraction handwriting digital image.

in the embodiment of the present application, the user can pre-enter at least one training image, and extract M kind covariance feature for each training image respectively, the number of the covariance of extracting for each training image is all identical, and the extracting mode of the M kind covariance of each training image in a plurality of training images is corresponding one by one, if namely this user has pre-entered 5 training images, respectively each training image is extracted M kind covariance feature, when M equals 4, also just to say respectively and extract 4 kinds of covariance features for each training image, and 4 kinds of covariance extracting modes of each training image are all corresponding one by one with 4 kinds of covariance extracting modes of other 4 training images.

After the handwriting digital image that receives user's input, at first the method extracts the M kind covariance feature of handwriting digital image, the M here is any one positive integer greater than 1, and the value of the M here is identical with the number of the covariance that each training image extracts, and the extracting mode of the M kind covariance feature of this handwriting digital image is corresponding one by one with the extracting mode of the M kind covariance of each training image.

S103, the class label entrained according to the M kind covariance feature of each training image of setting in advance, each training image of setting in advance and the M kind covariance feature of handwriting digital image are identified handwriting digital.

In the embodiment of the present application, the user pre-enters in each training image and all carries class label, after the M kind covariance feature of the Handwritten Numeral Recognition System that extracts user input, the class label entrained according to the M kind covariance feature of each training image that sets in advance, each training image of setting in advance and the M kind covariance feature of handwriting digital image can be completed the identification to handwriting digital.

this shows, the application provides a kind of Handwritten Numeral Recognition Method and device, and the method is by receiving the handwriting digital image of user's input, extract the M kind covariance feature of handwriting digital image, the value of M is any one positive integer greater than 1, the M kind covariance feature of last each training image according to setting in advance, the M kind covariance feature of the class label that each training image that sets in advance is entrained and handwriting digital image is identified handwriting digital, the application identifies handwriting digital by the M kind covariance feature that extracts the handwriting digital image, wherein, the value of M is any one positive integer greater than 1, avoided prior art in the process of carrying out Handwritten Digital Recognition, because can only adopt single covariance feature, cause the inaccurate problem of Handwritten Digital Recognition.

Embodiment two:

A kind of Handwritten Numeral Recognition Method process flow diagram that Fig. 2 provides for the embodiment of the present application two.

As shown in Figure 2, the method comprises:

The handwriting digital image of S201, reception user input.

The M kind covariance feature of S202, extraction handwriting digital image.

The implementation of the step S101-S102 that provides in the implementation of the step S201-S202 that the application implements to provide in two and embodiment one is identical, the detailed description of the step S201-S202 that the embodiment of the present application two provides sees also the step S101-S102 in the embodiment of the present application one, does not repeat them here.

Distance between the covariance feature of S203, the every kind of covariance feature that calculates respectively the handwriting digital image and each training image that sets in advance corresponding with every kind of covariance feature of handwriting digital image, generate a distance set for every kind of covariance feature in the handwriting digital image.

in the embodiment of the present application, after the M kind covariance feature that extracts the handwriting digital image, at first calculate the distance between the covariance feature of every kind of covariance feature of handwriting digital image and each training image that sets in advance corresponding with every kind of covariance feature of this handwriting digital image, for example: when having set in advance in the embodiment of the present application 5 training images, and respectively each training image is extracted 3 kinds of covariance features in advance, after receiving the handwriting digital image, also 3 kinds of covariance features that extract this handwriting digital image, and the extracting mode of every kind of covariance feature of this handwriting digital all with 5 training images that set in advance in the extracting mode of every kind of covariance feature of each training image corresponding one by one, then the first covariance feature that calculates this handwriting digital image respectively and the difference between the first covariance feature of each training image, and for all differences, form a distance set, because extracted 3 kinds of covariance features when extracting covariance feature, therefore the difference between the other two kinds of covariance features that also need to calculate respectively the handwriting digital image and other two kinds of covariance features of each training image, namely more respectively for distance set of every kind of covariance feature formation of handwriting digital image.As fully visible, this process is all to generate a distance set for every kind of covariance feature in the handwriting digital image, and each element in each distance set in this process is the training image to there being to set in advance all.

204, respectively to the element in each distance set according to sorting from small to large.

In the embodiment of the present application, respectively each element in each distance set is arranged according to order from small to large.

Neighbour's number K of S205, reception user input.

In the embodiment of the present application, can receive neighbour's number K of user's input, the value of this neighbour's number K is any one positive integer more than or equal to 1.

S206, according to neighbour's number K, obtain respectively front K element in each distance set, and according to neighbour of front K Element generation in each distance set set.

In the embodiment of the present application, after the neighbour's number K that receives user's input, neighbour's set of K Element generation before can selecting in each distance set respectively, as: suppose 3 set of current existence, select the neighbour's set of front K Element generation in first set, extract the neighbour's set of front K Element generation in second set and gather for neighbour of front K Element generation in the 3rd set finally.

S207, each neighbour is merged, generate final neighbour's set.

In the embodiment of the present application, will merge in all neighbours set obtained above, generate final neighbour's set.

S208, handwriting digital is identified.

Be according to the entrained class label of the training image that sets in advance corresponding to each element in final neighbour's set in the embodiment of the present application, handwriting digital is identified.

The entrained class label of the training image that sets in advance corresponding to each element in the final neighbour's set of a kind of basis that Fig. 3 provides for the embodiment of the present application two, the method flow diagram that handwriting digital is identified.

As shown in Figure 3, the method comprises:

S301, obtain the entrained class label of the training image that set in advance corresponding with each element in final neighbour set.

In the embodiment of the present application, all carry class label in each training image that sets in advance, and the corresponding unique class label of each handwriting digital.After the neighbour with all merges the final neighbour's set of generation, obtain the entrained class label of the training image that sets in advance corresponding to each element in this final neighbour's set.

S302, respectively the quantity of each class label in all categories label that gets is added up, the class labels that class label quantity is maximum are as the purpose class label.

Respectively the quantity of each class label in all categories label that gets is added up, calculate respectively the number of each class label in described class label, the class labels that quantity is maximum are as the purpose class label.

S303, according to the purpose class label, search the numeral corresponding with the purpose class label as purpose numeral in the corresponding relation of the class label that sets in advance and numeral.

In the application implements, set in advance the corresponding relation of class label and numeral, after obtaining the purpose class label, obtain and the corresponding numeral of this purpose class label according to the class label that sets in advance and the corresponding relation of numeral, and should numeral as the purpose numeral.

S304, the purpose numeral is identified as handwriting digital.

This shows, the application provides a kind of Handwritten Numeral Recognition Method and device, the class label that each training image that how to specifically disclose M kind covariance feature according to each training image that sets in advance in the method, sets in advance is entrained and the M kind covariance feature of handwriting digital image are identified handwriting digital, make the disclosed scheme of the application more clear, complete.

Embodiment three:

The method flow diagram of the setting up procedure of the M kind covariance feature of a kind of each training image that sets in advance that Fig. 4 provides for the embodiment of the present application three.

As shown in Figure 4, the method comprises:

At least one training image of S401, reception user input.

In the embodiment of the present application, can receive in advance at least one training image of user's input.

S402, extract respectively the M kind covariance feature of each training image.

in the embodiment of the present application, the user can pre-enter at least one training image, and extract M kind covariance feature for each training image respectively, the number of the covariance of extracting for each training image is all identical, and the extracting mode of the M kind covariance of each training image in a plurality of training images is corresponding one by one, if namely this user has pre-entered 5 training images, respectively each training image is extracted M kind covariance feature, when M equals 4, also just to say respectively and extract 4 kinds of covariance features for each training image, and 4 kinds of covariance extracting modes of each training image are all corresponding one by one with 4 kinds of covariance extracting modes of other 4 training images.

A kind of concrete grammar process flow diagram that extracts the M kind covariance feature of each training image that Fig. 5 provides for the embodiment of the present application three.

As shown in Figure 5, the method comprises:

S501, extract the M kind pixel characteristic that each point (x, y) on each training image is located respectively.

In the embodiment of the present application, the M kind pixel characteristic φ that locates of each point (x, y) on each training image of extraction _j(I _i, x, y), j=1,2...M represent, wherein, I _iRepresent i training image, i=1,2...N, φ _j(I _i, x, y) and the j kind pixel characteristic located of each point (x, y) of i training image of expression.

Illustrate the following computing formula that has provided three kinds of pixel characteristic extracting respectively each some place on first training image:

${\mathrm{\φ}}_1(I_i,x,y)={(x,y,I_i(x,y),|\frac{\∂}{\∂x}{I}_i(x,y)|,|\frac{\∂}{\∂y}{I}_i(x,y)\left|\right)}^T---\left(1\right)$

${\mathrm{\φ}}_2(I_i,x,y)={(I_i(x,y),|\frac{\∂}{\∂x}{I}_i(x,y)|,|\frac{\∂}{\∂y}{I}_i(x,y)|,|\frac{{\∂}^2}{\∂x\∂x}{I}_i(x,y)|,|\frac{{\∂}^2}{\∂y\∂y}{I}_i(x,y)\left|\right)}^T---\left(2\right)$

${\mathrm{\φ}}_3(I_i,x,y)={\left(\begin{array}{c}x,y,I_i(x,y),\left|\frac{\∂}{\∂x}{I}_i(x,y)\right|,\left|\frac{\∂}{\∂y}{I}_i(x,y)\right|,\sqrt{{\left|\frac{\∂}{\∂x}{I}_i(x,y)\right|}^2+{\left|\frac{\∂}{\∂y}{I}_i(x,y)\right|}^2},\\ \left|\frac{{\∂}^2}{\∂x\∂x}{I}_i(x,y)\right|,\left|\frac{{\∂}^2}{\∂y\∂y}{I}_i(x,y)\right|,a\tan \left(\frac{\left|\frac{\∂}{\∂x}{I}_i(x,y)\right|}{\left|\frac{\∂}{\∂y}{I}_i(x,y)\right|}\right)\end{array}\right)}^T---\left(3\right)$

Wherein, 1≤x≤m and 1≤y≤n, I _i(x, y) is illustrated in the gray-scale value at this some place,

With

Be illustrated respectively in the single order local derviation on this some x of place and y direction,

With

Be illustrated respectively in the second order local derviation on this some x of place and y direction.

In the embodiment of the present application, the computing formula of 3 kinds of calculating pixel features mentioning is 3 kinds of optimal ways, the inventor can according to the demand of oneself add arbitrarily or the project training image on the pixel characteristic computing formula at each some place.

S502, according to the M kind pixel characteristic that each point (x, y) on each training image is located, extract respectively the M kind covariance feature of each training image.

According to the M kind pixel characteristic that each point (x, y) on each training image is located, the computing formula of extracting respectively the M kind covariance feature of each training image is:

$C_i^j=\frac{1}{\mathrm{mn}}\underset{x=1}{\overset{m}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}({\mathrm{\φ}}_j(I_i,x,y)-{\stackrel{\‾}{\mathrm{\φ}}}_j\left(I_i\right)){({\mathrm{\φ}}_j(I_i,x,y)-{\stackrel{\‾}{\mathrm{\φ}}}_j\left(I_i\right))}^T,j=\mathrm{1,2}...M,i=\mathrm{1,2}...N$

Wherein,

For training image I _iThe pixel characteristic average.

As seen, the embodiment of the present application provides the concrete grammar of setting up procedure of the M kind covariance feature of each training image that sets in advance, and makes the concrete grammar of the Handwritten Numeral Recognition System that the embodiment of the present application provides more clear, perfect.

Embodiment four:

A kind of concrete grammar process flow diagram that extracts the M kind covariance feature of handwriting digital image that Fig. 6 provides for the embodiment of the present application four.

As shown in Figure 6, the method comprises:

S601, extract the M kind pixel characteristic that each point (x, y) on each handwriting digital image is located respectively.

In the embodiment of the present application, the M kind pixel characteristic φ that each point (x, y) on each handwriting digital image that extracts is respectively located _j(I, x, y), j=1,2...M represent, wherein, I represents handwriting digital image, φ _jThe j kind pixel characteristic that each point (x, y) of (I, x, y) expression handwriting digital image is located.

Illustrate the following computing formula that has provided three kinds of pixel characteristic extracting respectively each some place on the handwriting digital image:

${\mathrm{\φ}}_1(I_i,x,y)={(x,y,I_i(x,y),|\frac{\∂}{\∂x}{I}_i(x,y)|,|\frac{\∂}{\∂y}{I}_i(x,y)\left|\right)}^T---\left(1\right)$

${\mathrm{\φ}}_2(I_i,x,y)={(I_i(x,y),|\frac{\∂}{\∂x}{I}_i(x,y)|,|\frac{\∂}{\∂y}{I}_i(x,y)|,|\frac{{\∂}^2}{\∂x\∂x}{I}_i(x,y)|,|\frac{{\∂}^2}{\∂y\∂y}{I}_i(x,y)\left|\right)}^T---\left(2\right)$

${\mathrm{\φ}}_3(I_i,x,y)={\left(\begin{array}{c}x,y,I_i(x,y),\left|\frac{\∂}{\∂x}{I}_i(x,y)\right|,\left|\frac{\∂}{\∂y}{I}_i(x,y)\right|,\sqrt{{\left|\frac{\∂}{\∂x}{I}_i(x,y)\right|}^2+{\left|\frac{\∂}{\∂y}{I}_i(x,y)\right|}^2},\\ \left|\frac{{\∂}^2}{\∂x\∂x}{I}_i(x,y)\right|,\left|\frac{{\∂}^2}{\∂y\∂y}{I}_i(x,y)\right|,a\tan \left(\frac{\left|\frac{\∂}{\∂x}{I}_i(x,y)\right|}{\left|\frac{\∂}{\∂y}{I}_i(x,y)\right|}\right)\end{array}\right)}^T---\left(3\right)$

Wherein, 1≤x≤m and 1≤y≤n, I _i(x, y) is illustrated in the gray-scale value at this some place,

With

Be illustrated respectively in the single order local derviation on this some x of place and y direction,

With

Be illustrated respectively in the second order local derviation on this some x of place and y direction.

In the embodiment of the present application, the computing formula of 3 kinds of calculating pixel features mentioning is 3 kinds of optimal ways, and the inventor can add or design arbitrarily the pixel characteristic computing formula at each some place on the handwriting digital image according to the demand of oneself.

S602, according to the M kind pixel characteristic that each point (x, y) on the handwriting digital image is located, extract respectively the M kind covariance feature of handwriting digital image.

In the embodiment of the present application, according to the M kind pixel characteristic that each point (x, y) on the handwriting digital image is located, the computing formula of M kind covariance feature of extracting respectively the handwriting digital image is as follows:

$C^j=\frac{1}{\mathrm{mn}}\underset{x=1}{\overset{m}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}({\mathrm{\φ}}_j(I,x,y)-{\stackrel{\‾}{\mathrm{\φ}}}_j\left(I\right)){({\mathrm{\φ}}_j(I,x,y)-{\stackrel{\‾}{\mathrm{\φ}}}_j\left(I\right))}^T,j=\mathrm{1,2}...M$

Wherein,

Pixel characteristic average for the handwriting digital image I.

As seen, the embodiment of the present application discloses the concrete grammar of the M kind covariance feature that extracts the handwriting digital image, makes the concrete grammar of the Handwritten Numeral Recognition System that the embodiment of the present application provides more clear, perfect.

Embodiment five:

A kind of device for Identification of Handwritten Numerals structural representation that Fig. 7 provides for the embodiment of the present application five.

As shown in Figure 7, the device for Identification of Handwritten Numerals that the embodiment of the present application provides comprises: handwriting digital image receiving unit 1, handwriting digital image covariance feature extraction unit 2 and Handwritten Digital Recognition unit 3, wherein, handwriting digital image receiving unit 1, handwriting digital image covariance feature extraction unit 2 and Handwritten Digital Recognition unit 3 are connected successively.

Handwriting digital image receiving unit 1 is used for receiving the handwriting digital image of user's input.

In the embodiment of the present application, the user can input handwriting digital by hand-written mode on flat board, handset touch panel, and at first handwriting digital image receiving unit 1 receives the user inputs on the instruments such as handset touch panel or flat board handwriting digital image.

Handwriting digital image covariance feature extraction unit 2 is connected with handwriting digital image receiving unit 1, handwriting digital image covariance feature extraction unit 2 receives the handwriting digital image that handwriting digital receiving element 1 sends, and extracts the M kind covariance feature of handwriting digital image.

in the embodiment of the present application, the user can pre-enter at least one training image, and extract M kind covariance feature for each training image respectively, the number of the covariance of extracting for each training image is all identical, and the extracting mode of the M kind covariance of each training image in a plurality of training images is corresponding one by one, if namely this user has pre-entered 5 training images, respectively each training image is extracted M kind covariance feature, when M equals 4, also just to say respectively and extract 4 kinds of covariance features for each training image, and 4 kinds of covariance extracting modes of each training image are all corresponding one by one with 4 kinds of covariance extracting modes of other 4 training images.

After the handwriting digital image that receives 1 transmission of handwriting digital image receiving unit, at first handwriting digital image covariance feature extraction unit 2 extracts the M kind covariance feature of handwriting digital image, the M here is any one positive integer greater than 1, and the value of the M here is identical with the number of the covariance that each training image extracts, and the extracting mode of the M kind covariance feature of this handwriting digital image is corresponding one by one with the extracting mode of the M kind covariance of each training image.

Handwritten Digital Recognition unit 3 is connected with handwriting digital image covariance feature extraction unit 2, be used for to receive the M kind covariance feature of the handwriting digital image that handwriting digital image covariance feature extraction unit 2 sends, and the class label entrained according to the M kind covariance feature of each training image that sets in advance, each training image of setting in advance and the M kind covariance feature of handwriting digital image are identified to handwriting digital.

In the embodiment of the present application, the user pre-enters each training image and all carries class label, after the M kind covariance feature of the Handwritten Numeral Recognition System that extracts user input, the class label entrained according to the M kind covariance feature of each training image that sets in advance, each training image of setting in advance and the M kind covariance feature of handwriting digital image can be completed the identification to handwriting digital.

this shows, the application provides a kind of Handwritten Numeral Recognition Method and device, this device comprises: handwriting digital image receiving unit, handwriting digital image covariance feature extraction unit and Handwritten Digital Recognition unit, wherein, the handwriting digital image receiving unit is by receiving the handwriting digital image of user's input, handwriting digital image covariance feature extraction unit extracts the M kind covariance feature of handwriting digital image, the value of M is any one positive integer greater than 1, last Handwritten Digital Recognition unit is according to the M kind covariance feature of each training image that sets in advance, the M kind covariance feature of the class label that each training image that sets in advance is entrained and handwriting digital image is identified handwriting digital, the application extracts the M kind covariance feature of handwriting digital image by handwriting digital image covariance feature extraction unit, then by the Handwritten Digital Recognition unit, handwriting digital is identified, the value of M is any one positive integer greater than 1, avoided prior art in the process of carrying out Handwritten Digital Recognition, because can only adopt single covariance feature, cause the inaccurate problem of Handwritten Digital Recognition.

Embodiment six:

The structural representation of a kind of device for Identification of Handwritten Numerals that Fig. 8 provides for the embodiment of the present application six.

as shown in Figure 8, the device for Identification of Handwritten Numerals that the embodiment of the present application provides comprises: handwriting digital image receiving unit 1, handwriting digital image covariance feature extraction unit 2, metrics calculation unit 31, sequencing unit 32, neighbour's number receiving element 33, the neighbour gathers generation unit 34, final neighbour gathers generation unit 35 and Handwritten Digital Recognition subelement 36, wherein, handwriting digital image receiving unit 1, handwriting digital image covariance feature extraction unit 2, metrics calculation unit 31 is connected successively with sequencing unit 32, the end that the neighbour gathers generation unit 34 is connected with sequencing unit 32, the other end is connected with neighbour's number receiving element 33, and the neighbour gathers generation unit 34, final neighbour gathers generation unit 35 and is connected successively with Handwritten Digital Recognition subelement 36.

Wherein, handwriting digital image receiving unit 1 is used for receiving the handwriting digital image of user's input.

Handwriting digital image covariance feature extraction unit 2 is connected with handwriting digital image receiving unit 1, receives the handwriting digital image that handwriting digital receiving element 1 sends, and extracts the M kind covariance feature of handwriting digital image.

Metrics calculation unit 31 is connected with handwriting digital image covariance feature extraction unit 2, metrics calculation unit 31 is calculated respectively the distance between the covariance feature of every kind of covariance feature of handwriting digital image and each training image that sets in advance corresponding with every kind of covariance feature of handwriting digital image, for every kind of covariance feature in the handwriting digital image, generates a distance set; The all corresponding training image that sets in advance of each element in distance set.

Sequencing unit 32 is connected with metrics calculation unit 31, sequencing unit 32 respectively to the element in each distance set according to sorting from small to large.

Neighbour's number receiving element 33 is used for receiving neighbour's number K of user's input, and wherein, the value of K is any one positive integer more than or equal to 1.

The end that the neighbour gathers generation unit 34 is connected with sequencing unit 32, the other end is connected with neighbour's number receiving element 33, the neighbour gathers generation unit 34 according to neighbour's number K, obtain respectively front K element in each distance set, and according to neighbour's set of front K Element generation in each distance set.

Final neighbour gathers generation unit 35 and gathers generation unit 34 with the neighbour and be connected, and final neighbour gathers generation unit 35 each neighbour is merged, and generates final neighbour and gathers.

Handwritten Digital Recognition subelement 36 is gathered generation unit 35 with final neighbour and is connected, Handwritten Digital Recognition subelement 36, according to the entrained class label of the training image that sets in advance corresponding to each element in final neighbour's set, is identified handwriting digital.

The detailed construction schematic diagram of a kind of Handwritten Digital Recognition subelement that Fig. 9 provides for the embodiment of the present application six.

As shown in Figure 9, in the embodiment of the present application, the Handwritten Digital Recognition subelement comprises: class label acquiring unit 361, purpose class label determining unit 362 and purpose numeral determining unit 363 and recognition unit 364, wherein, class label acquiring unit 361, purpose class label determining unit 362 and purpose numeral determining unit 363 are connected successively with recognition unit 364.

Class label acquiring unit 361 is gathered generation unit 35 with final neighbour and is connected, class label acquiring unit 361 obtains the entrained class label of the training image that set in advance corresponding with each element in final neighbour set, the corresponding unique class label of each handwriting digital.

Purpose class label determining unit 362 is connected with class label acquiring unit 361, purpose class label determining unit 362 is added up the quantity of each class label in all categories label that gets respectively, and the class labels that class label quantity is maximum are as the purpose class label.

Purpose numeral determining unit 363 is connected with purpose class label determining unit 362, and purpose numeral determining unit 363 is searched the numeral corresponding with the purpose class label in the class label that sets in advance and digital corresponding relation according to the purpose class label digital as purpose.

Recognition unit 364 is connected with purpose numeral determining unit 363, and recognition unit 364 is identified as handwriting digital with the purpose numeral.

This shows, the application provides a kind of Handwritten Numeral Recognition Method and device, specifically disclose the detailed construction schematic diagram of Handwritten Digital Recognition unit and the detailed construction schematic diagram of Handwritten Digital Recognition subelement in this device, make the disclosed scheme of the application more clear, complete.

Embodiment seven:

Below in conjunction with concrete case, a kind of Handwritten Numeral Recognition Method and the device that the embodiment of the present application provides is elaborated:

The embodiment of the present application is the test of carrying out in MNIST handwriting digital data centralization, MNIST is the subset of famous American data set NIST, pattern-recognition common experimental data set, this data centralization has 60000 training samples and 10000 test samples, and test samples is the handwriting digital image of user's input of mentioning in the embodiment of the present application.The training sample of every class is 100 of stochastic samplings from training set in an embodiment, and test samples is 200 of stochastic samplings from inspection set.Concrete implementation step is as follows:

1) input training image set

I wherein _i∈ R ^{M * n}Be i training image, m and n be capable pixel value and the row pixel value of presentation video respectively, l _i{ 1, L, c} are I to ∈ _iThe class label that carries, namely represent I _iBe which numeral, N represents total number of training image.

In the present embodiment, m=n=28, classification is counted c and is got respectively 3,4 and 10, N=100c.

2) all training images are proceeded as follows: to training image I _iOn the following three kinds of pixel characteristic of pixel extraction located of each point (x, y):

${\mathrm{\φ}}_1(I_i,x,y)={(x,y,I_i(x,y),|\frac{\∂}{\∂x}{I}_i(x,y)|,|\frac{\∂}{\∂y}{I}_i(x,y)\left|\right)}^T---\left(1\right)$

${\mathrm{\φ}}_2(I_i,x,y)={(I_i(x,y),|\frac{\∂}{\∂x}{I}_i(x,y)|,|\frac{\∂}{\∂y}{I}_i(x,y)|,|\frac{{\∂}^2}{\∂x\∂x}{I}_i(x,y)|,|\frac{{\∂}^2}{\∂y\∂y}{I}_i(x,y)\left|\right)}^T---\left(2\right)$

${\mathrm{\φ}}_3(I_i,x,y)={\left(\begin{array}{c}x,y,I_i(x,y),\left|\frac{\∂}{\∂x}{I}_i(x,y)\right|,\left|\frac{\∂}{\∂y}{I}_i(x,y)\right|,\sqrt{{\left|\frac{\∂}{\∂x}{I}_i(x,y)\right|}^2+{\left|\frac{\∂}{\∂y}{I}_i(x,y)\right|}^2},\\ \left|\frac{{\∂}^2}{\∂x\∂x}{I}_i(x,y)\right|,\left|\frac{{\∂}^2}{\∂y\∂y}{I}_i(x,y)\right|,a\tan \left(\frac{\left|\frac{\∂}{\∂x}{I}_i(x,y)\right|}{\left|\frac{\∂}{\∂y}{I}_i(x,y)\right|}\right)\end{array}\right)}^T---\left(3\right)$

Wherein, 1≤x≤m and 1≤y≤n, I _i(x, y) is illustrated in the gray-scale value at this some place,

With

Be illustrated respectively in the single order local derviation on this some x of place and y direction,

With

Be illustrated respectively in the second order local derviation on this some x of place and y direction.

3) according to the pixel characteristic φ that extracts _j(I _i, x, y), j=1,2,3, obtain as follows training image I _iCorresponding covariance

$C_i^j=\frac{1}{\mathrm{mn}}\underset{x=1}{\overset{m}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}({\mathrm{\φ}}_j(I_i,x,y)-{\stackrel{\‾}{\mathrm{\φ}}}_j\left(I_i\right)){({\mathrm{\φ}}_j(I_i,x,y)-{\stackrel{\‾}{\mathrm{\φ}}}_j\left(I_i\right))}^T,j=\mathrm{1,2,3},i=1,L,N---\left(4\right)$

Wherein,

For training image I _iThe average of pixel characteristic.

4) incoming inspection image I, wherein I ∈ R ^{M * n}Utilize formula (1), formula (2), and formula (3) goes up to checking image I the pixel that point (x, y) is located respectively, extraction pixel characteristic φ _j(I, x, y), j=1,2,3, utilize the pixel characteristic of extracting, then extract the covariance C of checking image I according to formula (4) ^j

5) input neighbour number K, in the present embodiment, K is in set { 1,3,5, L, value in 29}; Calculate checking image covariance C ^jCovariance with training image

Between distance, namely

$d_i^j(C^j,C_i^j)=\sqrt{\underset{k=1}{\overset{d_j}{\mathrm{\Σ}}}\ln \left({\mathrm{\λ}}_k^2\right)},i=1,L,N,j=\mathrm{1,2,3}$

D wherein _jThe row or column number of expression covariance feature matrix, λ _kC ^jWith Generalized eigenvalue.Its d of covariance matrix by formula (1), (2) and (3) acquisition _jBe respectively 5,5 and 9.

6) respectively to three distance sets

With

, according to order sequence from small to large, with this, obtain three neighbour's set { I of image I _j ^K, j=1,2,3;

7) merge three neighbour's set and be I={I ₁ ^K, I ₂ ^K, I ₃ ^K, contain 3K training image in this set.

8) belong to the p class (1≤p≤c) is the p class with regard to the class label of determination check image if the neighbour gathers in I most training images.

The detailed description that a kind of Handwritten Numeral Recognition Method that the embodiment of the present application provides the embodiment of the present application in conjunction with concrete case and device carry out is only a kind of optimal way, and the inventor can adjust execution step in this detailed description according to the demand of oneself.

Effect of the present invention can be by following experimental verification:

Every class training sample 100 of stochastic samplings from training set, test samples 200 of stochastic samplings from inspection set.Repeat this sampling process 10 times, the result of output is this average result of 10 times finally.The method of experiment contrast has single Lie group k nearest neighbor sorter and the present invention.Have 3 Lie group k nearest neighbor sorters, represent to adopt respectively the Lie group k nearest neighbor sorter of different covariances with covariance 1, covariance 2 and covariance 3.

Having carried out altogether three groups of experiments, is respectively that numeral (1,7,9) is carried out three classification, and numeral (1,2,7,9) is carried out four classification, and numeral (0～9) is carried out very class.Experimental result is respectively as Figure 10, Figure 11 and shown in Figure 12, and wherein, Figure 10 is along with parameter K changes the situation that numeral 1,7 and 9 is classified; Figure 11 is along with parameter K changes the situation that numeral 1,2,7 and 9 is classified; Figure 12 is along with parameter K changes the situation that digital 0-9 is classified.Can see from these figure, the present invention has better recognition effect.Covariance 1 and covariance 2 in not testing on the same group, show different performances.Comparatively speaking, covariance 3 is best, but in the literature not with this as classification application, be apparent that very much, the present invention has better recognition performance than covariance 3.

In this instructions, each embodiment adopts the mode of going forward one by one to describe, and what each embodiment stressed is and the difference of other embodiment that between each embodiment, identical similar part is mutually referring to getting final product.For the disclosed device of embodiment, because it corresponds to the method disclosed in Example, so description is fairly simple, relevant part partly illustrates and gets final product referring to method.

Be below only the application's preferred implementation, make those skilled in the art can understand or realize the application.Multiple modification to these embodiment will be apparent to one skilled in the art, and General Principle as defined herein can be in the situation that do not break away from the application's spirit or scope, realization in other embodiments.Therefore, the application will can not be restricted to these embodiment shown in this article, but will meet the widest scope consistent with principle disclosed herein and features of novelty.

Claims (10)

1. a Handwritten Numeral Recognition Method, is characterized in that, comprising:

Receive the handwriting digital image of user's input;

Extract the M kind covariance feature of described handwriting digital image, and the M kind covariance feature of described handwriting digital image is corresponding one by one with the M kind covariance feature of each training image that sets in advance respectively; The value of M is any one positive integer greater than 1;

The class label entrained according to the M kind covariance feature of each the described training image that sets in advance, each described training image of setting in advance and the M kind covariance feature of described handwriting digital image are identified described handwriting digital.

2. method according to claim 1, is characterized in that, the setting up procedure of the M kind covariance feature of described each training image that sets in advance specifically comprises:

Receive at least one training image of user's input;

Extract respectively the M kind covariance feature of each described training image; The value of M is any one positive integer greater than 1.

3. method according to claim 1 and 2, is characterized in that, the described training image of each that sets in advance all carries class label;

The class label that the M kind covariance feature of each described training image that described basis sets in advance, each the described training image that sets in advance are entrained and the M kind covariance feature of described handwriting digital image are identified described handwriting digital, specifically comprise:

Calculate respectively the distance between the covariance feature of every kind of covariance feature of described handwriting digital image and each the described training image that sets in advance corresponding with every kind of covariance feature of described handwriting digital image, for every kind of covariance feature in described handwriting digital image, generate a distance set; The all corresponding training image that sets in advance of each element in described distance set;

Respectively to the element in each described distance set according to sorting from small to large;

Receive neighbour's number K of user's input; The value of K is any one positive integer more than or equal to 1;

According to described neighbour's number K, obtain respectively front K element in each described distance set, and according to neighbour's set of front K Element generation in each distance set;

Each described neighbour is merged, generate final neighbour's set;

According to the entrained class label of the training image that sets in advance corresponding to each element in described final neighbour's set, described handwriting digital is identified.

4. method according to claim 3, is characterized in that, the corresponding unique class label of each described handwriting digital; Every kind of corresponding unique handwriting digital of described class label,

Describedly according to the entrained class label of the training image that sets in advance corresponding to each element in described final neighbour set, described handwriting digital is identified, is specifically comprised:

Obtain the entrained class label of the training image that set in advance corresponding with each element in described final neighbour set;

Respectively the quantity of each class label in all described class labels that get is added up, the class labels that class label quantity is maximum are as the purpose class label;

Search the numeral corresponding with described purpose class label as purpose numeral according to described purpose class label in the corresponding relation of the class label that sets in advance and numeral;

Described purpose numeral is identified as handwriting digital.

5. require 1 or 2 described methods according to claim, it is characterized in that, the M kind covariance feature of the described handwriting digital image of described extraction specifically comprises:

Extract respectively the M kind pixel characteristic φ that each point (x, y) on each described handwriting digital image is located _j(I, x, y), j=1,2...M, wherein, I represents handwriting digital image, φ _jThe j kind pixel characteristic that each point (x, y) of (I, x, y) expression handwriting digital image is located;

According to the M kind pixel characteristic that each point (x, y) on described handwriting digital image is located, extract respectively the M kind covariance feature of described handwriting digital image, computing formula is as follows:

$C^j=\frac{1}{\mathrm{mn}}\underset{x=1}{\overset{m}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}({\mathrm{\φ}}_j(I,x,y)-{\stackrel{\‾}{\mathrm{\φ}}}_j\left(I\right)){({\mathrm{\φ}}_j(I,x,y)-{\stackrel{\‾}{\mathrm{\φ}}}_j\left(I\right))}^T,j=\mathrm{1,2}...M$

Wherein,

Pixel characteristic average for the handwriting digital image I.

6. method according to claim 2, is characterized in that, the described M kind covariance feature that extracts respectively each described training image; The value of M is any one positive integer greater than 1, specifically comprises:

Extract respectively the M kind pixel characteristic φ that each point (x, y) on each described training image is located _j(I _i, x, y), j=1,2...M, wherein, I _iRepresent i training image, i=1,2...N, φ _j(I _i, x, y) and the j kind pixel characteristic located of each point (x, y) of i training image of expression;

The M kind pixel characteristic that each point (x, y) on training image according to each is located, extract respectively the M kind covariance feature of each described training image, and computing formula is as follows:

$C_i^j=\frac{1}{\mathrm{mn}}\underset{x=1}{\overset{m}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}({\mathrm{\φ}}_j(I_i,x,y)-{\stackrel{\‾}{\mathrm{\φ}}}_j\left(I_i\right)){({\mathrm{\φ}}_j(I_i,x,y)-{\stackrel{\‾}{\mathrm{\φ}}}_j\left(I_i\right))}^T,j=\mathrm{1,2}...M,i=\mathrm{1,2}...N$

Wherein,

For training image I _iThe pixel characteristic average.

7. method according to claim 1 and 2, is characterized in that, the M kind covariance feature of described handwriting digital image is corresponding one by one with the M kind covariance feature of each described training image respectively, specifically comprises:

The extracting mode of every kind of covariance feature of described handwriting digital image is identical one by one with the extracting mode of every kind of covariance feature of each described training image respectively.

8. a device for Identification of Handwritten Numerals, is characterized in that, comprising: handwriting digital image receiving unit, handwriting digital image covariance feature extraction unit and Handwritten Digital Recognition unit, wherein,

Described handwriting digital image receiving unit is used for receiving the handwriting digital image of user's input;

Described handwriting digital image covariance feature extraction unit is connected with described handwriting digital image receiving unit, be used for extracting the M kind covariance feature of described handwriting digital image, and the M kind covariance feature of described handwriting digital image is corresponding one by one with the M kind covariance feature of each training image that sets in advance respectively; The value of M is any one positive integer greater than 1;

Described Handwritten Digital Recognition unit is connected with described handwriting digital image covariance feature extraction unit, is used for the entrained class label of the M kind covariance feature according to each the described training image that sets in advance, each described training image of setting in advance and the M kind covariance feature of described handwriting digital image described handwriting digital is identified.

9. device according to claim 8, it is characterized in that, the described training image of each that sets in advance all carries class label, described Handwritten Digital Recognition unit comprises: metrics calculation unit, sequencing unit, neighbour's number receiving element, neighbour gather generation unit, final neighbour gathers generation unit and Handwritten Digital Recognition subelement, wherein

Described metrics calculation unit is connected with described handwriting digital image covariance feature extraction unit, be used for calculating respectively the distance between the covariance feature of every kind of covariance feature of described handwriting digital image and each the described training image that sets in advance corresponding with every kind of covariance feature of described handwriting digital image, for every kind of covariance feature in described handwriting digital image, generate a distance set; The all corresponding training image that sets in advance of each element in described distance set;

Described sequencing unit is connected with described metrics calculation unit, is used for respectively element to each described distance set according to sorting from small to large;

Described neighbour's number receiving element is used for receiving neighbour's number K of user's input; The value of K is any one positive integer more than or equal to 1;

The end that the neighbour gathers generation unit is connected with described sequencing unit, the other end is connected with described neighbour's number receiving element, be used for according to described neighbour's number K, obtain respectively front K element in each described distance set, and according to neighbour's set of front K Element generation in each distance set;

Described final neighbour gathers generation unit and gathers generation unit with described neighbour and be connected, and is used for each described neighbour is merged, and generates final neighbour and gathers;

Described Handwritten Digital Recognition subelement is gathered generation unit with described final neighbour and is connected, be used for the entrained class label of the training image that sets in advance corresponding to each element according to described final neighbour's set, described handwriting digital is identified.

10. device according to claim 9, is characterized in that, the corresponding unique class label of each described handwriting digital; Described Handwritten Digital Recognition subelement comprises: class label acquiring unit, purpose class label determining unit and purpose numeral determining unit and recognition unit, wherein,

Described class label acquiring unit is gathered generation unit with described final neighbour and is connected, and is used for obtaining the training image that the set in advance entrained class label corresponding with each element of described final neighbour's set;

Described purpose class label determining unit is connected with described class label acquiring unit, be used for respectively the quantity of each class label of all described class labels of getting is added up, the class labels that class label quantity is maximum are as the purpose class label;

Described purpose numeral determining unit is connected with described purpose class label determining unit, is used for searching the numeral corresponding with described purpose class label as the purpose numeral according to described purpose class label at the class label that sets in advance and the corresponding relation of numeral;

Described recognition unit is connected with described purpose numeral determining unit, is used for described purpose numeral is identified as handwriting digital.

Images