CN102541347A - Automatic identification system and method of handwriting Chinese character - Google Patents

Abstract

The invention relates to a system and method for automatic recognition of handwriting. The system includes the EMB-4650 main board, and its extended sub-interfaces include LVDS interface, DVI interface, USB interface and COM interface; these interfaces are respectively connected with an AUO display screen, a DVI-LVDS signal conversion board, a USB controller and a GSM module The DVI-LVDS signal conversion board is connected with another AUO display, and the USB controller is connected with the touch screen. The operation steps of the recognition method are (1) input of handwriting data, (2) preprocessing of handwriting data, (3) shape feature extraction of handwriting data, and (4) recognition of handwriting data. The invention adopts a new method of polar coordinate feature recognition with font centroid as the reference point. Users can use handwriting to record, save, and send important information in real time; to achieve matching from handwriting to handwriting. The system integrates high confidentiality and high-speed transmission, and is suitable for intelligence personnel to collect and screen important information.

Description

System and method for automatic handwriting recognition

technical field

The invention relates to a method for automatically extracting note features by using a computer, and develops a handwriting automatic recognition system and method, which uses an image segmentation algorithm and a discriminant algorithm based on statistical features to search and extract handwriting. This handwriting identification method can be widely used. Applied in finance, law and other fields.

Background technique

Handwriting is a fairly stable behavioral characteristic, different people have different handwriting, and handwriting is easy to obtain. With the rapid development of information digitization, handwriting recognition has become an active research topic in the fields of computer vision and pattern recognition. As we all know, literature experts can identify the authenticity of handwriting, but it is still difficult for a computer to automatically extract handwriting features and identify its authenticity, especially for a small amount of handwriting.

The research on computer handwriting identification first began in the 1960s. In recent years, due to the wide application of computer handwriting identification technology in public security, finance, education evaluation and other fields, it is gradually becoming a very active but difficult research topic in the field of computer vision and pattern recognition. Extensive research has been carried out, but no satisfactory solution has been achieved so far.

Due to the early start and rapid development of information and communication technology in developed countries, its popularity and use have reached a relatively high level, and their computer management and automation technology also has a considerable penetration rate and a high degree of modernization, so computer identity authentication Technology has gradually received extensive attention and research. At present, there have been several related products for signature identification in foreign markets, but due to the particularity of Chinese character structure, research on Chinese handwriting identification has progressed slowly. Due to differences in cultural background, etc., Western countries have less research on the identification of oriental characters. Therefore, most of the research institutions on related issues are located in Asia. At present, the main research institutions include Pusan University in South Korea, the Pattern Recognition Laboratory of the Institute of Automation of the Chinese Academy of Sciences, and the Chinese Academy of Sciences. Computing office. In addition, many companies in China are currently devoting themselves to the development of handwriting identification systems and have achieved some results, but none have been put into use.

Contents of the invention

The purpose of the present invention is to provide a system and method for automatic recognition of handwriting in view of the defects in the prior art. It takes the diary in daily life as the prototype of the design, combines personal double-page note writing habits, and uses computer technology and innovation Search algorithm for personal handwriting to achieve information search. Starting from the many structural characteristics of Chinese characters, the system comprehensively analyzes the characteristics of handwriting, abandons the huge standard font library, and directly searches from "handwriting to handwriting" for convenience. It not only pays attention to extracting commonness from individuality, but also emphasizes the grasp of individuality .

In order to achieve the above object, the idea of the present invention is as follows: the present invention is based on image-based handwriting recognition to realize the search function. This technology belongs to the category of offline character recognition. The system search function only needs to find the same or similar characters in the library written by the user, and does not need to determine which character in the Chinese characters the character written by the user corresponds to. . The object to be processed by this system is a two-dimensional image. After separating the written Chinese characters from the background, through a series of preprocessing, find out the location of the picture file with the same or similar characters as the written Chinese characters and display them at the same time. Can. The entire search process can be divided into the following parts: information acquisition (handwritten character input), preprocessing (including background removal, character segmentation, etc.), single character feature extraction, comparison and selection.

According to above-mentioned inventive concept, the present invention adopts following technical scheme:

A stylus automatic recognition system comprises an EMB-4650 main board made by Intel® Atom Z510P chip master control, it is characterized in that the sub-interfaces of said main board expansion have LVDS interface, DVI interface, USB interface and COM interface; These interfaces are respectively connected with AUO display, DVI-LVDS signal conversion board, USB controller, and GSM module; thus the DVI-LVDS signal conversion board is connected with another AUO display, and the USB controller is connected with the touch screen. When the user writes data on the touch screen, the data is transmitted to the EMB-4650 main board through the USB controller, and the EMB-4650 main board transmits the data to the display screen to realize real-time display; at the same time, the EMB-4650 main board performs a recognition algorithm on these data processing, and then send the processed results to the display screen to achieve the effect of simple human-computer interaction and real-time operation. If necessary, the edited font data can be sent to the GSM module through the COM port to realize real-time data transmission.

The user writes the font on the touch screen, the font data is input to the storage device, and after preprocessing by the processor, the shape feature of a single character is extracted by using the polar coordinate method for protection in this paper, and the input character is selected and determined. At the same time, the user's input and processed character results are displayed on the liquid crystal display. If necessary, the GSM module can be used to wirelessly transmit the input information, making it more convenient and timely to transmit information.

In the above system, the power supply of the system is powered by an external power supply. Obtain the 18V AC power from the mains 220V50Hz through a power filter and a transformer, connect it to the voltage conversion circuit, and the voltage conversion circuit converts the 18V AC power into 12V, 5V, 3.3V and 1.8V power supplies, Provide working power for each module of the system.

In the above system, the touch screen is connected to the USB interface of the main board through the conversion board. The conversion board converts the input signal into a USB signal and connects with the system host, and controls the performance of the touch screen through the corresponding software.

Among the above systems, B089AW01 V1 is a color dot-matrix liquid crystal display composed of a TFT LCD panel, a set of drive circuits and an LED backlight system. The screen display supports WSVGA (1024(H) x 600(V)) format and is equipped with 262k color data drive with LED backlight drive circuit. All input signals are compatible with LVDS interface. One of the two display screens is connected to the LVDS interface of the motherboard, and the other is connected to the DVI interface of the motherboard through a DVI-LVDS conversion board. DVI-LVDS completes the conversion of the input DVI signal into an LVDS signal suitable for the LCD screen.

In the above system, the two GSM modules are Zehan T-1008 series GSM/GPRS wireless transceivers. The module is mainly composed of the main part of Modem (including chip, peripheral circuit, etc.), regulated power supply, and antenna. The GPRS module is connected to the motherboard through the COM port of the chip.

A method for automatic handwriting recognition, using the above-mentioned system for recognition, characterized in that the specific steps are as follows: Step 1: input of handwriting data; Step 1: preprocessing of handwriting data; Step 3: shape of handwriting data Feature extraction; step 4: handwriting data recognition.

Compared with the prior art, the present invention has the following obvious outstanding substantive features and significant advantages:

The handwriting automatic recognition method is that the recognition method is the handwriting search, and a new method of polar coordinate feature recognition with the handwriting font centroid as the reference point is established, and the recognition accuracy is extremely high; and it is convenient to find writing information, eliminating the complicated paper search process, Real-time storage of notes in the form of pictures is realized, and the storage time is long.

The present invention is different from the previous single-screen display, and adopts a dual-screen display mode, and can be set to a dual-screen independent display mode and a combined display mode in the software. In the dual-screen independent display mode, the main display can be set to a digital display, and the secondary The display is set to MID, and the screen rotation can also be set. This convertible display mode can meet the requirements of various occasions, and it is easy to carry without losing the visual effect.

The font shape feature extraction algorithm of the present invention is advanced, and is different from the traditional X-Y axis coordinate recognition method. It innovatively proposes a polar coordinate recognition method with the center of mass of the font as the center of the circle, which solves the problem of segmentation for Chinese characters with left-right structure or left-middle-right structure. The setting of the algorithm threshold and personal writing habits are a pair of contradictions that are difficult to solve, so as to improve the recognition rate of fonts.

Description of drawings

Fig. 1 is a structural block diagram of the automatic handwriting recognition system of the present invention.

Fig. 2 is a flow chart of automatic recognition of stylus in the present invention.

Fig. 3 is a flowchart of the preprocessing of the handwriting data in Fig. 2 .

FIG. 4 is a flow chart of image binarization in FIG. 3 .

FIG. 5 is a flowchart of font recognition in FIG. 1 .

Detailed ways:

Preferred implementation cases of the present invention are described in detail as follows in conjunction with accompanying drawings:

Embodiment one: referring to Fig. 1, this stylus automatic recognition system comprises an EMB-4650 main board that is made by Intel® Atom Z510P chip master control, and the sub-interface of its expansion has LVDS interface, DVI interface, USB interface and COM interface ; These interfaces are respectively connected with the AUO display screen, DVI-LVDS signal conversion board, USB controller, and GSM module; thus the DVI-LVDS signal conversion board is connected with another AUO display screen, and the USB controller is connected with the touch screen. When the user writes data on the touch screen, the data is transmitted to the EMB-4650 main board through the USB controller, and the EMB-4650 main board transmits the data to the display screen to realize real-time display; at the same time, the EMB-4650 main board performs a recognition algorithm on these data processing, and then send the processed results to the display screen to achieve the effect of simple human-computer interaction and real-time operation. If necessary, the edited font data can be sent to the GSM module through the COM port to realize real-time data transmission. The present invention will be described in further detail below in conjunction with preferred embodiments of the present invention and accompanying drawings.

Embodiment two: referring to Fig. 2, the realization method of the automatic handwriting recognition system adopts the above-mentioned system to identify, and the specific operations include the following steps:

Step 1: Input of handwriting data: According to the different working principles of the touch screen, it can be divided into surface acoustic wave touch screen, resistive touch screen, capacitive touch screen, infrared touch screen and so on. The touch screen used in the present invention is a resistive touch screen. The resistive touch screen is installed on the surface of the display, mainly composed of two transparent conductive layers, and the two conductive layers are insulated. From the perspective of electrical characteristics, each conductive layer is equivalent to a resistance element, and a corresponding relationship between the plane position and the voltage drop can be established by applying a voltage across the conductive layer. When touched, the two conductive layers are in contact at the contact, and the voltage reflecting the plane position is drawn out by means of the contact, and the contact position is obtained after measurement. The touch screen stores these contact data in the form of image data for the host to read.

Step 2: Preprocessing of stroke data, see Figure 3. The data preprocessing flow chart of the present invention is shown in FIG. 4 . The main board (1) reads the stored image data from the touch screen (7), and inputs it into the register of the main board (1).

Step 2-1: Convert 256-color bitmap to grayscale.

a) Calculate the grayscale mapping table

From the relationship between the brightness signal Y and the three primary colors R, G, B, Y=0.299R+0.587G+0.114B can be obtained bMap[i] = (BYTE)(0.299 * lpbmi->bmiColors[i].rgbRed

+0.587* lpbmi->bmiColors[i].rgbGreen

+0.114* lpbmi->bmiColors[i].rgbBlue + 0.5);

Where lpbmi is a pointer to the BITMAPINFO structure

b) Update the three primary color components of the DIB palette

lpbmi->bmiColors[i].rgbRed = i;

lpbmi->bmiColors[i].rgbGreen = i;

lpbmi->bmiColors[i].rgbBlue = i;

c) Replace the color index of each pixel (that is, replace it with a grayscale value according to the grayscale mapping table)

for(i = 0; i < lHeight; i++)

{ for(j = 0; j < lWidth; j++)

{ lpSrc = (unsigned char*)lpDIBBits + lLineBytes * (lHeight - 1 - i) + j;

*lpSrc = bMap[*lpSrc];

}

}

(lHeight and lWidth in the above program segment are the height and width of the entire image respectively)

,通过确定阈值Threshold 来确定整幅图像的黑白分布。需要强调的是至于将特征物与背景分开这一目标我们则编写了专门的去背景函数以从根本上将彩色背景与文字分成两幅“互相独立”的位图，而非单纯地将二者分成两个区域，这是我们处理方式上有别于传统二值化思想之处。 Step 2-2: Binarization: In the present invention, the method of hard threshold is used to realize binarization on the gray scale image, and the program flow chart of image binarization is shown in Fig. 4 . Set a certain threshold Threshold, and use Threshold to divide the grayscale image data into two parts: the pixel group greater than or equal to the threshold and the pixel group smaller than the threshold. For example, the input grayscale image function is f(x,y), and the output binary image function is g(x,y), then

, by determining the threshold Threshold to determine the black and white distribution of the entire image. It should be emphasized that as for the goal of separating the features from the background, we have written a special de-background function to fundamentally divide the colored background and text into two "mutually independent" bitmaps, rather than simply separating the two Divided into two areas, this is what we do differently from the traditional binarization thinking.

Step 2-3: Discrete noise removal: In the process of writing, especially when brewing a good article, people often have the habit of touching the pen to the paper from time to time, which makes the information we save in real time A certain amount of noise is mixed in, which brings additional unfavorable factors to the subsequent accurate search for handwriting, so denoising must be performed first.

The basic idea is to scan the entire image and set the pixels judged as discrete points to white. In the operation of de-discrete noise, the nesting of two functions is used, one function is void RemoveScatterNoise(HDIB hDIB), this function is an outer layer function, the main function is to judge whether the pixel is a discrete point, if so, set it to white, and call the DeleteScaterJudge() function when making a discrete judgment; the other The function is bool DeleteScaterJudge(). This function adopts the recursive idea to examine the eight directions of up, down, left, right, upper left, upper right, lower left, and lower right. If it is a black point, it will call itself for recursion. Since it is of type bool, so The return value is TRUE or FALSE, in void It can be obtained from RemoveScatterNoise (HDIB hDIB). If it is FALSE, it means that the pixel point is a discrete point, and it is set to white, thereby realizing the effect of removing discrete noise.

Step 2-4: Gradient sharpening: The purpose of image sharpening is to make the blurred image clearer. There are generally two methods for image sharpening: one is the differential method, and the other is the high-pass filter method. The gradient sharpening adopted in the present invention is a commonly used method in the differential method.

，定义

在点

处的梯度矢量为： Let the image be

,definition

at point

Gradient vector at for:

，梯度有两个重要性质：梯度的方向在函数

最大变化率方向上和梯度的幅度用

表示，且其值为：

, the gradient has two important properties: the direction of the gradient is in the function

The direction of the maximum rate of change and the magnitude of the gradient are given by

represents, and its value is:

在其最大变化率方向上的单位距离所增加的量。 From this formula, it can be concluded that the value of the gradient is

The amount of increase per unit distance in the direction of its maximum rate of change.

For discrete digital images, the above formula can be rewritten as:

For the convenience of calculation, the following approximate calculation formula (1) can also be used:

来表示图像，即令

=

，则有上述公式(1)可见：在图像变换缓慢的地方其值很小（对应于图像较暗）；而在线条轮廓等变换较快的地方的值很大。图像在经过梯度运算后变得清晰从而达到锐化的目的。 If the gradient value is directly used

to represent the image, that is,

=

, then it can be seen from the above formula (1): the value is small in places where the image changes slowly (corresponding to darker images); and the value is large in places where the line contour changes quickly. The image becomes clear after the gradient operation to achieve the purpose of sharpening.

，如果小于该阈值

，则保持原灰度值不变；如果大于或等于阈值，则赋值为：

For the unfavorable factor that the image is dark because the gradient is small in places where the image changes slowly, the solution we adopt is to give a threshold

, if less than the threshold

, keep the original gray value unchanged; if greater than or equal to the threshold , then the assignment is :

Step 2-5: Thinning: Thinning refers to the process of finding the central bone part of an image. The central bone part is one of the important features describing the geometric and topological properties of the image. Thinning the processed image helps to highlight the shape features and reduce the amount of redundant information.

区域，对各点标记名称P1，P2，…P8，其中P1位于中心。如果P1=1（即黑点），下面4个条件如果同时满足，则删除P1（P1=0）。①2

NZ(P1)

6；②

=1；③P2*P4*P8=0或者

1；④P2*P4*P6=0或者

1；对于图像中的每一个点重复这一步骤，直到所有的点都不可删除为止。 The thinning algorithm adopted in the present invention has the characteristics: one image in one image

region, label the points with names P1, P2, ... P8, where P1 is at the center. If P1=1 (that is, black dot), if the following four conditions are met at the same time, delete P1 (P1=0). ①2

NZ(P1)

6;②

=1; ③P2*P4*P8=0 or

1; ④P2*P4*P6=0 or

1; Repeat this step for each point in the image until all points cannot be deleted.

Step 3: Shape feature extraction of handwriting data: The shape features of handwritten Chinese characters include inclination, center of gravity offset, character stretching, etc. These features have clear physical meanings and reflect part of the writing style of the writer. Usually, this feature can be described by moments.

的字符图像

，其p+q阶矩定义为： for a picture

character image for

, its p+q order moment is defined as:

，各阶矩的物理意义说明如下：

表示图像在水平方向上的伸展度；

表示图像在垂直方向上的伸展度；

表示图像的倾斜度（>0表示图像向左上倾斜，

<0表示图像向右上倾斜）；

表示图像在水平方向上的重心偏移度（

>0表示重心偏左，<0表示重心偏右）；

表示图像在垂直方向上的重心偏移度（>0表示重心偏上，<0表示重心偏下）；

表示图像水平伸展的均衡程度（

>0表示图像下部的水平伸展度比图像上部大，

<0表示图像下部的水平伸展度比图像上部小）；

表示图像垂直伸展的均衡程度（

>0表示图像右部的垂直伸展度比图像左部大，

<0表示图像右部的垂直伸展度比图像左部小）。 in,

, the physical meaning of each order moment is explained as follows:

Indicates the stretch of the image in the horizontal direction;

Indicates the stretch of the image in the vertical direction;

Indicates the slope of the image ( >0 indicates that the image is tilted up to the left,

<0 means the image is tilted up to the right);

Indicates the offset of the center of gravity of the image in the horizontal direction (

>0 means the center of gravity is left, <0 means the center of gravity is to the right);

Indicates the offset of the center of gravity of the image in the vertical direction ( >0 means the center of gravity is on the upper side, <0 means the center of gravity is lower);

Indicates how evenly the image is stretched horizontally (

>0 indicates that the horizontal stretch of the lower part of the image is larger than that of the upper part of the image,

<0 means the horizontal stretch of the lower part of the image is smaller than that of the upper part of the image);

Indicates the balance of the vertical stretch of the image (

>0 indicates that the vertical stretch of the right part of the image is larger than that of the left part of the image,

<0 means the right part of the image has less vertical stretch than the left part of the image).

From the central moments calculated above, the following normalized shape features can be derived:

a) Aspect ratio:

；其中

b) Font slope:

;in

；其中

的计算式如下：

c) Pull length:

;in

The calculation formula is as follows:

；其中size表示字符点阵的大小，即 d) Stretch:

; where size represents the size of the character lattice, namely

The third-order central moment may be positive or negative because of its value, in fact, they are composed of two parts whose basis function is positive and negative.

The normalized features extracted by the third-order central moment are as follows:

①Horizontal offset:

②Vertical offset:

③Horizontal stretch balance:

④ Vertical stretch balance:

，定义的距离如下： For the corresponding feature words, find their feature vectors

, the distance is defined as follows:

；其中r为：

; where r is:

Step 4: Handwriting data recognition: take the center of mass of the text as the origin, and half of the normalized text height as the radius, establish a polar coordinate system, and divide the entire circumference into fan-shaped average blocks, and the number of blocks n is variable. Finally, judge whether the two bitmaps are similar, that is, whether the currently written Chinese characters have a certain degree of similarity with the words that appeared in the previously saved notes, and follow the same criterion: only when the two bitmaps correspond to n blocks When the pairwise differences of the total number of pixels in each area are smaller than the preset threshold, the two are considered similar. And setting n as variable is equivalent to setting the precision of the lookup as adjustable. Obviously, when n=12 we consider the lookup to be regarded as a fuzzy lookup, and when n=360, we consider it to be an exact lookup. Fig. 5 Flowchart of font data identification. After the font data recognition is completed, the computer displays the recognized fonts on the screen. After the font editing is completed, the font data can be sent to the receiving station through the wireless GSM network through software control to realize real-time transmission.

Claims (2)

1. An automatic handwriting pen body recognition system comprises an EMB-4650 mainboard (1) which is made by controlling Intel Atom Z510P chips, and is characterized in that the expanded subinterfaces of the mainboard (1) comprise an LVDS interface, a DVI interface, a USB interface and a COM interface; the interfaces are respectively connected with an AUO display screen (2), a DVI-LVDS signal conversion board (3), a USB controller (4) and a GSM module (5); the DVI-LVDS signal conversion board (3) is connected with another AUO display screen (6), and the USB controller (4) is connected with the touch screen (7); when a user writes data in the touch screen, the data are transmitted to the EMB-4650 mainboard through the USB controller, and meanwhile, the EMB-4650 mainboard transmits the data to the display screen to realize real-time display; meanwhile, the EMB-4650 mainboard performs recognition algorithm processing on the data, and then sends the processed result to a display screen for displaying, so that a simple human-computer interaction real-time operation effect is achieved; the edited font data can be sent to the GSM module through the COM port, and real-time data sending is achieved.

2. An automatic handwriting recognition method, which is recognized by the automatic handwriting recognition system of claim 1, and is characterized by comprising the following specific operation steps:

step 1: inputting the data of the pen body: when the touch screen is touched, the touch screen stores the data in the form of image data for a host to read;

step 2: preprocessing the pen body data: the main board 1 reads in stored image data from the touch screen (7) and inputs the image data into a register of the main board (1);

2-1, converting the 256-color bitmap of the data input by the touch screen (7) into a gray scale map according to a gray scale mapping table method;

step 2-2: binarization: realizing binarization on the gray level image by adopting a hard threshold value mode;

step 2-3: and (3) removing discrete noise: denoising noise mixed by touch;

step 2-4: gradient sharpening: carrying out image sharpening by adopting a differential method:

let the image be

Definition of

At the point of

Gradient vector of (2)

Comprises the following steps:

the gradient has two important properties: direction of gradient as a function

In the direction of maximum rate of change and the magnitude of the gradient

And its value is:

the following results are obtained: the value of the gradient is

The amount by which the unit distance in the direction of its maximum rate of change increases;

for discrete digital images, the above equation is rewritten as:

for convenience of calculation, the following approximate calculation formula (1) may also be employed:

if gradient values are directly adoptedTo represent an image, i.e. order

=

Then, the above formula (1) shows that: where the image transitions slowly its value is small (corresponding to the image being darker); and the values are large at places where the transformation is fast, such as line profiles; the image becomes clear after gradient operation so as to achieve the aim of sharpening;

for the adverse factor that the image is dark due to small gradient at the place where the image transformation is slow, the solution is to give a threshold value

If less than the threshold value

Keeping the original gray value unchanged; if greater than or equal to the threshold valueThen assign a value of

：

Step 2-5: thinning: the process of finding the central skeleton part of one image is specified, the central skeleton part is one of important features for describing the geometric and topological properties of the image, and the thinning of the processed image is helpful for highlighting the shape characteristics and reducing the redundant information content;

the adopted refinement algorithm has the characteristics that: one of an image

Areas, for each point, are labeled the names P1, P2, … P8, where P1 is located at the center; if P1=1, i.e., a black dot, the following 4 conditions, if satisfied simultaneously, delete P1 (P1 = 0); (ii) 2NZ(P1)

6；②= 1; ③ P2P 4P 8=0 or

1; (iv) P2 × P4 × P6=0 or

1; this step is repeated for each point in the image until all points are undeletable;

and step 3: shape feature extraction of the pen body data: the shape characteristics of the handwritten Chinese characters comprise inclination, gravity center offset and character extension, and the characteristics are described by moments;

for a frame

Character image ofThe p + q order moment is defined as:

wherein,each ofThe physical meaning of the order moments is explained below:

representing the degree of extension of the image in the horizontal direction;

representing the degree of extension of the image in the vertical direction;representing the inclination of an image

>0 means that the image is tilted up to the left,

<0 represents the image tilting up to the right;

representing the degree of shift of the center of gravity of an image in the horizontal direction

>0 indicates that the center of gravity is to the left,<0 indicates the center of gravity is to the right;

representing the degree of shift of the center of gravity of an image in the vertical direction

>0 means that the center of gravity is on the upper side,

<0 represents a center of gravity off;

indicating the degree of equalization of the horizontal stretching of the image,

>0 means that the horizontal extension in the lower part of the image is larger than in the upper part of the image,

<0 indicates that the horizontal stretch in the lower part of the image is smaller than in the upper part of the image;

indicating the degree of equalization of the vertical stretching of the image,>0 means that the vertical extension of the right part of the image is greater than the left part of the image,

<0 indicates that the vertical extension of the right part of the image is smaller than the left part of the image;

from the central moments calculated above, some normalized shape features are derived:

aspect ratio:

font gradient:

(ii) a Wherein

Drawing length:(ii) a Wherein

Is calculated as follows:

the stretching degree:

(ii) a Wherein size represents the size of the character lattice, i.e.

The central moments of the third order, since their values may be positive or negative, are in fact composed of two parts, the basis functions of which are positive and negative;

the normalized features extracted by the third-order central moment are as follows:

horizontal offset degree:

vertical offset degree:

③ degree of horizontal stretching and balancing:

vertical stretching equilibrium degree:

for corresponding feature words, a feature vector is obtained

The defined distances are as follows:

(ii) a Wherein r is:

and 4, step 4: and (3) identifying pen body data: establishing a polar coordinate system by taking the centroid of the character as an origin and half of the normalized height of the character as a radius, and performing sector average partitioning on the whole circumference, wherein the number n of blocks is variable;

finally, whether the two bitmaps are similar or not is judged, namely whether the current written Chinese character and the character appearing in the past stored notes have certain similarity or not follows the same criterion: only when the difference value of every two pixel points of each block area in n block areas corresponding to the two bitmaps is smaller than a preset threshold value, the two bitmaps are regarded as similar; freezing n to be variable is equivalent to setting the search precision to be adjustable;

obviously, we consider a search to be a fuzzy search when n =12, and a precise search when n = 360; after the font data recognition is completed, the computer displays the recognized font on the screen.

Images