CN104182966A - Automatic splicing method of regular shredded paper - Google Patents

Abstract

The invention belongs to image processing technology, and in particular relates to a method for automatic splicing of regular shredded paper. The technical scheme of the present invention is realized through six steps: (1) prepare the image data set and perform preprocessing; (2) classify the shredded paper according to Chinese and English, single and double sides; (3) extract the local area feature of each image , such as the position and gray value of the boundary pixel of the shredded paper, and the height of the upper (lower) boundary; the extraction range of the features of the English shredded paper is expanded, and the additional features include: the line height of the English shredded paper, the English old paper (4) Reclassify the fragments according to the feature values extracted in step (3); (5) Partially match the fragments, row matching and column matching; (6) Matched images are restored. The method provided by the invention can splice a large amount of shredded paper more accurately.

Description

A method for automatic splicing of regular shredded paper

technical field

The invention belongs to the application field of image processing technology, and in particular relates to an automatic splicing method of regular shredded paper. the

Background technique

Shredded paper splicing technology is an important research branch of digital image processing technology. It is to spatially match and align a group of shredded papers that overlap with each other, so as to seamlessly stitch together to obtain a complete image of a wide-view scene. the

Shredded paper automatic splicing and restoration technology has important applications in the fields of judicial evidence restoration, historical document restoration, and military intelligence acquisition. In recent years, with the publication of the German Stasi Document Restoration Project, the research on shredded document recovery technology has attracted widespread attention. the

The key to the splicing of shredded paper is the matching technology of the fragments. The splicing of traditional broken files is more about using the edge shape of the fragments to extract their contour curves and using computer algorithms for splicing. Nowadays, with the widespread application of paper shredders, in more and more splicing problems of shredded paper, the edge shapes of shredded paper are roughly the same, and edge shape splicing is no longer applicable. For regular-shaped shredded paper, according to the text content contained in the edge of the paper, the parameters of the shredded paper boundary are determined through image registration operations, and the shreds are matched to finally achieve seamless splicing. However, in practical applications, the larger the number of paper sheets to be spliced, the larger the number of edges of paper sheets with similar text information, and the higher the similarity. However, the resolution of digital images formed by computer scanning has certain limitations, so a certain amount of wrong splicing will occur during the splicing process. The effect that the ideal splicing technology wants to achieve is "zero error". As far as the current technical situation is concerned, most of the existing splicing methods for shredded paper are aimed at irregular shapes, and methods that can be effectively applied to splicing large-scale wide and regular pieces of paper are relatively rare. the

The technical key to improving the quality of automatic splicing of shredded paper lies in how to obtain text or image information on shredded paper with high quality. Generally speaking, the smaller the amount of information on the fragments, the greater the chance of splicing errors or even failure to splice. Therefore, so far, in this technical field, the process of automatically splicing shredded paper images in the hope of obtaining the final high-quality wide-width shredded paper splicing has relatively large technical difficulties. the

Contents of the invention

The purpose of the present invention is to provide a method for automatically splicing regular shredded paper, which can more accurately splice a large number of shredded paper. the

The present invention is achieved through the following technical solutions, mainly comprising the following six steps:

1. The specific steps of image data set preparation and preprocessing include:

1.1 Number the scraps of paper from left to right and from top to bottom, and record them as 1, 2, 3...n; if you need to distinguish the front and back, record the front as a1, a2, a3...an; Recorded as b1, b2, b3...bn;

1.2 Digitize the image, take the pixel as the smallest unit, and extract the gray value and location of each pixel, and establish a function matrix;

1.3 Value the image: the point with the gray value "0" is a black point, the point with the gray value "255" is a white point, and the point between "0" and "255" is a gray point;

1.4 Denoising: Since the original information is a continuous analog signal, the digitally processed image should also be a discontinuous point image with a continuous trend. For the situation that the same color point completely surrounds the different color point, the color of the different color point is assimilated into the color of the surrounding points;

2. Classify the shredded paper as a whole, and divide it into 4 situations according to Chinese and English, single and double-sided: Chinese single-sided, Chinese double-sided, English single-sided, English double-sided;

3. Extract the features of the local area of each image respectively, these features include: the position and gray value of the pixel points on the boundary of the shredded paper, and the height of the upper (lower) boundary; expand the extraction range of the English shredded paper features, add Features include: row height of English scraps, horizontal position of English scraps, line spacing of English scraps;

The method of feature extraction is as follows:

i) The position and gray value of the outermost pixel of the shredded paper:

Define a column of pixels at the leftmost (right) end of the shredded paper as the left (right) boundary, and a row of pixels at the top (bottom) end as the upper (lower) boundary, and extract the position and gray value of each boundary pixel;

ii) Upper (lower) boundary height:

According to whether the upper and lower borders of each fragment are completely white, they are divided into two categories: white border height and black border height. The specific classification methods are as follows:

The bottom of the fragment is the x-axis, the left side of the fragment is perpendicular to the x-axis and the y-axis is the y-axis, and the intersection point of the x-axis and the y-axis is the origin to establish a coordinate system, and each pixel on the picture is projected to the y-axis. As shown in Figure 1. The projection of a black or gray point is recorded as a valid projection, and the number of projections is increased by 1, while the projection of a white point is invalid, and the number of projections does not change. Record the number of projections f(h) on the projection point whose distance from the origin is h pixels. the

$\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 10</span>}\\ \mathrm{<span\; class="notranslate">\; 255</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 10</span>}\end{array}\right.$

When the number of projections f(h) is less than 1/10 of the total pixel points n of the row, the gray value g(h) of point h on the y-axis is recorded as "0"; when the number of projections f(h) is greater than or equal to When it is 1/10 of the total pixel point n in this row, the gray value g(h) of point h is recorded as "255". the

On the projected axis, count down from the upper boundary of the shards until points with different colors appear. The height of this section is the height of the upper boundary, and the height of the lower boundary is also the same. the

iii) Row height of English shreds:

The height of English letters and the height of the position occupied in the same line are roughly the same, therefore, according to step i) for projection, the interval with a gray value of "1" is the effective interval of letters, and the height of the effective interval is defined as row high;

iv) Horizontal position of English fragments:

After step i) projection, the upper and lower boundaries of the effective projection interval of the letter, the distance from the top of the shredded paper is called the horizontal position of the row of letters, which is used to determine the position of the row of letters on the shredded paper;

v) Line spacing:

Extract the vertical distance between two horizontal positions as the line spacing;

4. Classify the fragments according to the feature set extracted in step 3:

Specific steps are as follows:

i) According to whether there is text stroke information on the edge of the paper, the shredded paper is divided into three categories: upper and lower border fragments, left and right border fragments and middle fragments;

ii) According to the characteristics of line spacing, further classify the above three types of shredded paper, and the same line spacing is divided into one category;

iii) According to the height of the upper (lower) boundary, classify the three types of fragment sets formed in step i), and the fragments with the same or similar height of the upper (lower) boundary are divided into the same fragment set:

Classification needs to follow certain conditions:

(1) The number of fragments of each type must be equal to or slightly less than the number of longitudinal cuts of the paper;

(2) If the number of categories separated from other heights is less than 1/5 of the number of fragments of each category, it is not an independent category;

(3) Several categories that are highly continuous with each other are classified into the same category;

(4) The final total number of categories is the number of cross-cuts of the paper;

(5) If the category still cannot be determined, use the same method to make an auxiliary judgment on the height of the bottom. the

iv) Utilize the horizontal position to further classify each fragment set formed in step ii), and the fragments at the same horizontal position are divided into one class;

5. Specific steps for matching fragments:

5.1 Partial matching of fragments, that is, matching between two fragments, the following takes left and right matching as an example:

i. Define X _ij as the gray value of the jth row of pixels on the right boundary of the i'th fragment, and define Y _i'j as the gray value of the jth row of pixels on the left boundary of the i'th fragment (i≠i '). The key to judging whether it matches or not lies in the degree of matching between X _ij and Y _i′j . The feature set extracted in step 3 is based on the right boundary feature, and the judgment standard is defined as:

X _ij is white, Y _i′j-1 , Y _i′j , Y _i′j+1 have three colors of gray, white and black and not all black, which is normal and can be matched;

X _ij is gray, any color of Y _i′j-1 , Y _i′j , Y _i′j+1 is normal and can be matched;

X _ij is black, Y _i′j-1 , Y _i′j , Y _i′j+1 are not all white, it is normal and can be matched;

The rest of the cases are abnormal and cannot be matched. the

The relationship between X _ij and Y _i′j-1 , Y _i′j , and Y _i′j+1 is shown in FIG. 2 .

Among them: X _ij : the gray value of the pixel point in the jth row of the leftmost column of the i-th note;

Y _i′j : the gray value of the pixel point in the jth row of the rightmost column of the i′th paper strip;

The specific process of the boundary tracking algorithm is as follows:

(1) Select fragments i and i′

(2) Assume that fragments i and i′ match each other;

(3) Read the gray value of the pixel point X _ij of the row j of the right boundary of the fragment i;

(4) Scan the pixels Y i _′(j-1) , Y _i′j , Y i′(j+1 _{) of the j-1, j, j+1th} row of the left boundary of the fragment i′, and judge its Is it all white;

(5) If all are white and exceed the line range, return to (3) after j=j+1;

(6) If it is not all white, then j=j+1, read the next line, and judge whether X _ij is white;

(7) If it is white, return to (5);

(8) If it is not white, judge whether Y _i′(j-1) , Y _i′j , Y _i′(j+1) are all white;

(9) If all are white, then return to (5);

(10) If it is not white, then j=j+1 reads the next line, and judges the color of X _ij ;

(11) If it is white, return to (5);

(12) If it is gray, return to (5);

(13) If it is black, judge whether Y _i′(j-1) , Y _i′j , Y _i′(j+1) are all white;

(14) If it is not white, return to (5);

(15) If all are white, the matching process of fragments i and i' ends, and fragments i and i' do not match;

(16) If j+1 exceeds the row range, the matching process of fragment i and i' ends, and fragment i and i' match. the

ii. According to the judgment standard of step i, determine the mathematical model of the image matching index, specifically:

${\mathrm{<span\; class="notranslate">\; S</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span>$

in:

S _ii′ : the matching index between the i-th fragment and the i′-th fragment;

N: the total number of pixels on the vertical height of the shredded paper;

X _ij : the gray value of the pixel points in row j on the right boundary of fragment i;

T _i′ (X _ij ): judging the matching index of the right boundary feature of the i-th shredded paper row j and the left boundary feature of the i′-th paper strip in the corresponding row;

The matching index is specifically expressed as:

${\mathrm{<span\; class="notranslate">\; T</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\\ {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>& \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; 255</span>}<span\; class="notranslate">\; .</span>\end{array}\right.$

in, T ₂ (X _ij )=0.

If and only when the S _ii′ index is 0, the two fragments are regarded as matching; if it is not 0, the matching cannot be performed, and the larger the value, the worse the matching degree.

5.2 Step 5.1 has completed the local matching process between the two fragments. The shreds that meet the matching conditions obtained in step 5.1 are formed into small fragment sets, and row matching and column matching are performed on the fragment sets. i. Specific process:

i) Based on one of the fragments, if the partial matching of the two fragments is successful, the two fragments will be merged into one fragment and put into a new fragment set; if the partial matching is not successful, the reference fragment will be kept and the local match. When none of the fragments in the original fragmentation set can be partially matched successfully, they are put into the new fragmentation set;

ii) For new pieces, repeat the above steps until all pieces of paper are spliced into a complete line of pieces. the

ii. According to the above process, determine the mathematical model of the image row matching index, specifically:

Objective function:

W=min∑S _ii′

Restrictions:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; S</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span>\\ {\mathrm{<span\; class="notranslate">\; T</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\\ {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>& \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; 255</span>}<span\; class="notranslate">\; .</span>\end{array}\right.\\ \mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; m</span>}\end{array}\right.$

Wherein, a is the number of S _ii′ ;

The minimum value of W is 0;

M is the number of longitudinal cuts of shredded paper

iii will form a fragmented row through the fragmented set matched by the row, transpose the fragmented row matrix, and then perform column matching in the same way;

6. Restore the image after matching in step 5. the

Effective benefits of the present invention are:

It can splice and process a relatively large number of shredded paper at one time, and proposes a corresponding optimization solution for the matching splicing process, which is mainly reflected in: 

(1) For Chinese fragments, only feature processing is performed on the boundary, and a mathematical model is established on the boundary feature. Therefore, when the present invention updates the image sample database, the scope of the scanned database is greatly reduced. In the case of a large number of shredded paper to be spliced, Has a time advantage. the

(2) The boundary tracking algorithm designed by the present invention can ensure the uniqueness in the shredded paper matching process, further improving the effectiveness and operability of the present invention. the

Description of drawings

Fig. 1 is the text projection diagram of shredded paper of the embodiment of the present invention;

Fig. 2 is the local matching figure of the embodiment of the present invention;

Fig. 3 is a flow chart of the boundary tracking algorithm of the embodiment of the present invention. the

Detailed ways

Taking Chinese single-sided as an example, the execution process of the present invention will be briefly described below. In this example, a total of 209 shredded paper images are selected, and these 209 shredded pieces are cut 10 times crosswise and 18 lengthwise on an A4 paper. The specific execution steps are as follows:

(1) Pretreatment

(a) Prepare and preprocess the image data set, including image digitization, denoising, and binarization;

(b) Arrange the scraps of paper into a matrix of 11 rows and 19 columns, numbered from 1 to 209 in order from left to right and from top to bottom;

(2) Extract the feature set of the debris matrix, these features include: the position and gray value of the pixel points of the outermost layer of the debris, and the height of the upper boundary;

(3) Use eigenvalues to classify fragments:

① According to whether there is text and stroke information on the edge of the paper, the shredded paper is divided into three categories: 11 left and right border fragments, 19 upper and lower border fragments, and 149 middle fragments;

② Classify the fragments according to the extracted upper boundary height:

Generally, the height of the white upper border or the height of the black upper border of the same row of fragments is roughly the same. Calculate the boundary height of each fragment and classify the fragments with the same boundary height into one category, and count the number of such fragments. The statistical results are shown in Table 1. the

Table 1 Number of Fragments with the Same Upper Boundary Height

It can be seen from Table 1 that there are 43 groups of white tops and black tops, and the picture is only cut into 11 rows. Therefore, further processing of the divided categories is required. the

Classification needs to follow certain conditions:

(6) The number of fragments of each category must be equal to or slightly less than 19;

(7) The category separated from other heights, if the number is less than 10, is not an independent category (for example, height 3);

(8) Several categories that are highly continuous with each other are classified into the same category;

(9) The final total number of categories is 11 categories;

(10) If the category still cannot be determined, an auxiliary judgment will be made based on the height of the bottom. the

The classification after further processing is shown in Table 2:

Table 2 Classification of the height of the upper boundary and the corresponding number of debris

(4) Row and column matching of fragments based on boundary tracking algorithm;

(5) Display the spliced image:

Table 3 Fragment number table after splicing

049 054 065 143 186 002 057 192 178 118 190 095 011 022 129 028 091 188 141 061 019 078 067 069 099 162 096 131 079 063 116 163 072 006 177 020 052 036 168 100 076 062 142 030 041 023 147 191 050 179 120 086 195 026 001 087 018 038 148 046 161 024 035 081 189 122 103 130 193 088 167 025 008 009 105 074 071 156 083 132 200 017 080 033 202 198 015 133 170 205 085 152 165 027 060 014 128 003 159 082 199 135 012 073 160 203 169 134 039 031 051 107 115 176 094 034 084 183 090 047 121 042 124 144 077 112 149 097 136 164 127 058 043 125 013 182 109 197 016 184 110 187 066 106 150 021 173 157 181 204 139 145 029 064 111 201 005 092 180 048 037 075 055 044 206 010 104 098 172 171 059 007 208 138 158 126 068 175 045 174 000 137 053 056 093 153 070 166 032 196 089 146 102 154 114 040 151 207 155 140 185 108 117 004 101 113 194 119 114

Claims (10)

1. A method for automatic splicing of regular shredded paper, characterized in that: the method mainly comprises the following six steps: (1) Preparation and preprocessing of image data sets, including numbering shredded paper, image digitization, image value, and denoising; (2) Classify the shredded paper as a whole, and divide it into 4 situations according to Chinese and English, single and double sides: Chinese single side, Chinese double sided, English single side, English double sided; (3) Extract the features of the local area of each image respectively. For Chinese, these features include: the position and gray value of the boundary pixel point of the shredded paper, and the height of the upper and lower boundaries; expand the extraction range of the English shredded paper features, and add The features also include: row height of English scraps, horizontal position of English scraps, line spacing of English scraps; (4) According to the feature set extracted in step 3, the fragments are classified: firstly, they are divided into three types: upper and lower boundary fragments, left and right boundary fragments, and middle fragments; Fragments of features are grouped into one class; (5)a: Carry out local matching on the fragments according to the boundary tracking algorithm, and determine the matching index of the boundary features of the fragments; b: Use the boundary tracking algorithm to perform row matching and column matching on the shreds that meet the matching conditions obtained in step a; (6) Restore the image after step 5 matching. 2. The method for automatically splicing a piece of regular shredded paper according to claim 1, characterized in that: the numbering method described in step (1) is: number the shredded paper pieces sequentially from left to right and from top to bottom, It is recorded as 1, 2, 3...n; if it is necessary to distinguish the front and back, the front is recorded as a1, a2, a3...an; the reverse is recorded as b1, b2, b3...bn; step (1) The image digitization described above refers to: taking pixels as the minimum unit, and extracting the gray value and location of each pixel, and establishing a function matrix; the image value described in step (1) refers to: the gray value is "0 " is a black point, a point with a grayscale value of "255" is a white point, and a point between "0" and "255" is a gray point; the denoising point described in step (1) refers to: for the same color point Completely surround the heterochromatic point, and assimilate the color of the heterochromatic point to the color of the surrounding points. 3. The method for automatic splicing of a piece of regular shredded paper according to claim 1, characterized in that: the position and gray value feature and extraction method of the boundary pixel points of the shredded paper pieces described in step (3) are: define shredded paper The leftmost or rightmost column of pixels on the paper is the left or right boundary, and the topmost or bottommost row of pixels is the upper and lower boundaries, and the position and gray value of each boundary pixel are extracted; The upper and lower boundary height features described in step (3) and the extraction method are: according to whether the upper and lower boundaries of each fragment are completely white, they are divided into two categories: white boundary height and black boundary height; the specific classification methods are as follows: Take the bottom of the fragment as the x-axis, the left side of the fragment perpendicular to the x-axis as the y-axis, and the intersection point of the x-axis and the y-axis as the origin to establish a coordinate system, and horizontally project each pixel point on the picture to the y-axis; The projection of a black or gray point is recorded as a valid projection, and the number of projections is increased by 1, while the projection of a white point is invalid, and the number of projections does not change; record the number of projections f(h) on a projection point that is h pixels away from the origin; $\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 10</span>}\\ \mathrm{<span\; class="notranslate">\; 255</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 10</span>}\end{array}\right.$ When the number of projections f(h) is less than 1/10 of the total number of pixels n in the row, record the gray value g(h) of point h on the y-axis as "0"; when the number of projections f(h) is greater than or equal to When 1/10 of the total number of pixels n in this row, the gray value g(h) of point h is recorded as "255"; On the projection axis, count down from the upper boundary of the fragments until points with different colors appear; the height of this section is the height of the upper boundary, and the height of the lower boundary is the same; The line height feature and extraction method of the English fragments described in step (3) are: the height of the English letters and the height of the positions occupied in the same row are roughly the same, therefore, after the English fragments are horizontally projected, the gray value The interval of "1" is the effective interval of the letter, and the height of the effective interval is defined as the row height, and the row height is extracted; The horizontal position feature and extraction method of the English fragment described in step (3) are: after the English fragment is carried out lateral projection, the upper and lower boundaries of the effective projection interval of the letter, the distance from the top of the fragment is called the horizontal position of the row letter. position, to extract horizontal position features; The line spacing feature and extraction method of the English shreds described in step (3) are: the vertical distance between two horizontal positions is defined as the line spacing, and the line spacing feature is extracted. 4. The method for automatic splicing of a piece of rule shredded paper according to claim 1, characterized in that: according to the feature set extracted in step 3 described in step (4), the fragments are classified: the specific steps are: i) According to whether there is text stroke information on the border of the paper sheet, the shredded paper is divided into three categories: upper and lower border fragments, left and right border fragments and middle fragments; ii) According to the characteristics of line spacing, further classify the above three types of shredded paper, and the same line spacing is divided into one category; iii) Classify the three types of fragment sets formed in step i according to the height of the upper and lower boundaries, and the fragments with the same or similar heights as the upper and lower boundaries are divided into the same fragment set; iv) Use the horizontal position feature to further classify each fragment set formed in step ii, and divide the fragments at the same horizontal position into one category. 5. The method for automatic splicing of a piece of regular shredded paper according to claim 1 or 4, characterized in that: in the described step iii), the conditions for classifying are to be observed: (1) The number of fragments of each type must be equal to or slightly less than the number of longitudinal cuts of the paper; (2) If the number of categories separated from other heights is less than 1/5 of the number of fragments of each category, it is not an independent category; (3) Several categories that are highly continuous with each other are classified into the same category; (4) The final total number of categories is the number of cross-cuts of the paper; (5) If the category still cannot be determined, an auxiliary judgment will be made based on the height of the bottom. 6. The method for automatic splicing of a piece of regular shredded paper according to claim 1, characterized in that: the step of matching the fragments described in step (5) comprises: a. Partially match the fragments according to the boundary tracking algorithm, that is, match between two fragments, and determine the matching index of the boundary characteristics of the fragments; b. The shredded pieces of paper obtained in step a that meet the matching conditions are formed into small fragment sets, and the boundary tracking algorithm is used to perform row matching and column matching on the fragment sets. 7. The method for automatic splicing of a piece of ruled shredded paper according to claim 6, characterized in that: the criterion for partial matching described in step a is: define X _ij as the jth row of pixels on the right boundary of the ith fragment The gray value of the point, define Y _i'j as the gray value of the pixel point in the jth row on the left boundary of the i'th fragment (i≠i'); the key to determine whether it matches or not lies in _Xij and Y _i'j The degree of matching between, the feature set extracted in step 3 is based on the right boundary feature, and the judgment standard is defined as: X _ij is white, Y _i′j-1 , Y _i′j , Y _i′j+1 have three colors of gray, white and black and not all black, which is normal and can be matched; X _ij is gray, any color of Y _i′j-1 , Y _i′j , Y _i′j+1 is normal and can be matched; X _ij is black, Y _i′j-1 , Y _i′j , Y _i′j+1 are not all white, it is normal and can be matched; The rest of the cases are abnormal and cannot be matched; The relationship between X _ij and Y _i′j-1 , Y _i′j , Y _i′j+1 is shown in Table 1; Among them: X _ij : the gray value of the pixel point in the jth row of the leftmost column of the i-th note; Y _i′j : the gray value of the pixel point in the jth row of the rightmost column of the i′th paper strip. 8. A kind of regular shredded paper automatic mosaic method according to claim 7, is characterized in that: according to the judging standard of described partial match, determine the mathematical model of image matching index, be specifically: ${\mathrm{<span\; class="notranslate">\; S</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span>$ in: S _ii′ : the total matching index between the i-th fragment and the i′-th fragment; N: the total number of pixels on the vertical height of the shredded paper; X _ij : the gray value of the pixel points in row j on the right boundary of fragment i; T _i' (X _ij ): judge the matching index of the data of the right boundary feature of the i-th shredded paper row j and the data of the left boundary feature of the i'th paper strip in the corresponding row; The matching index is specifically expressed as: ${\mathrm{<span\; class="notranslate">\; T</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\\ {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>& \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; 255</span>}<span\; class="notranslate">\; .</span>\end{array}\right.$ in, T ₂ (X _ij )=0 If and only when the S _ii′ index is 0, the two fragments are regarded as matching; if it is not 0, the matching cannot be performed, and the larger the value, the worse the matching degree. 9. The method for automatic splicing of a ruled shredded paper according to claim 6, characterized in that: the specific flow of the boundary tracking algorithm described in step a is as follows: 1) Select fragments i and i'; 2) Assume fragments i and i' match each other; 3) Read the gray value of the pixel point X _ij in row j of the right boundary of fragment i; 4) Scan the pixel points Y i'(j-1) , Y i'j _, Y _i'(j+1) of the j-1, j, j+1 rows adjacent to X _ij on the left border of the fragment i _' , judge whether it is all white; 5) If all are white and exceed the line range, return 3 after j+1); 6) If it is not all white, then j+1 reads the next line, and judges whether X _ij is white; 7) If it is white, return to 5); 8) If it is not white, judge whether Y _i′(j-1) , Y _i′j , Y _i′(j+1) are all white; 9) If all are white, return to 5); 10) If it is not white, then j+1 reads the next line to judge the color of X _ij ; 11) If it is white, return to 5); 12) If it is gray, return to 5); 13) If it is black, judge whether Y _i′(j-1) , Y _i′j , Y _i′(j+1) are all white; 14) If it is not white, return to 5); 15) If all are white, the matching process of fragments i and i' ends, and fragments i and i' do not match; 16) If j+1 exceeds the row range, the matching process of fragment i and i' ends, and fragment i and i' match. 10. The method for automatic splicing of a piece of regular shredded paper according to claim 6, characterized in that: the specific process of performing row matching and column matching described in step b is: (1) The specific process of line matching: i) Based on one of the fragments, if the partial matching of the two fragments is successful, the two fragments will be merged into one fragment and put into a new fragment set; if the partial matching is not successful, the reference fragment will be kept and the local Matching; when the fragments in the original fragmentation set cannot be partially matched successfully, they are put into the new fragmentation set; ii) For new pieces, repeat the above steps until all the pieces of paper are spliced into a whole row; (2) According to the above process, determine the mathematical model of the image row matching index, specifically: Objective function: W=min∑S _ii′ Restrictions: $\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; S</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span>\\ {\mathrm{<span\; class="notranslate">\; T</span>}}_{{\mathrm{<span\; class="notranslate">\; i</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\\ {\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>& \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; 255</span>}<span\; class="notranslate">\; .</span>\end{array}\right.\\ \mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; m</span>}\end{array}\right.$ Wherein, a is the number of S _ii′ ; The minimum value of W is 0; M is the number of times the shredded paper is cut horizontally; (3) Form a fragment row from the fragment set matched by row, transpose the fragment row matrix, and then perform column matching. The specific process of column matching is the same as that of row matching.