CN102708371A - Method for recognizing and automatically sequencing comic frames according to segmenting lines - Google Patents

Abstract

The invention provides a scheme for automatically recognizing, extracting and sequencing comic squares according to the comic constituent grammar so as to reduce a large amount of manual segmenting work. The segmented comic squares can be put on the mobile phone or PDA with a small screen to read, therefore, the cartoons can be read conveniently, easily and vividly. According to the method for recognizing and automatically sequencing comic frames provided by the invention, the straight lines in the pictures of the cartoons are detected to provide an iterative segmentation method according to the comic frames, the comic segmenting lines and the distribution characteristics of the comic frames to recognize and sequence the comic frames, and the segmented comic frames are stored by using a binary tree structure. The experiments prove that the method can be used for recognizing and segmenting the comic frames (comic squares) of most cartons and sequencing the comic frames (comic squares) in sequence according to the pot of the story.

Description

Comic frame recognition and automatic sorting method based on dividing line

technical field

The invention designs a comic frame recognition and automatic sorting method based on dividing lines, and belongs to the field of image data processing. the

Background technique

Like e-books, electronic comics not only have the advantages of convenience, economy and automatic sorting, but also have a sense of dynamism. Now they are gradually developing and becoming popular, especially in the youth consumer market. While reading electronic comics on mobile phones or PDAs, the effect of animation comics is produced by flipping the up and down keys, which is favored by readers. But reading comics is different from e-books, because of the limitation of the screen size of mobile phones and PDAs, and in order to make the storylines have a sequence when reading on mobile phones, each comic needs to be divided into smaller pictures without affecting reading. the

There have been many electronic comic services, but most of these works are done manually to identify the comic frames in the segmented comic pages. At present, there are relatively few researches on electronic comics at home and abroad. Masashi YAMADA et al. proposed a method of sorting manga frames and dialog boxes in manga frames by angle, focusing on the correctness of the sorting results, and did not mention how to effectively deal with manga frames with incomplete dividing lines. Takamasa Tanaka, Kenji Shoji and Fubito Toyama proposed a tree structure to represent manga frames, and analyzed the layout of manga frames by identifying straight lines with gradient density functions. This method can solve the problem that the split line is incomplete, that is, the dialog box cuts off the split line, but it cannot be applied in real time. There are also some other aspects of comic picture processing work that have been studied. For example, Syeda-Mahmood et al. proposed a method for text recognition in comic pictures, so that the text in the dialog boxes in the comics can be read separately, which is very important for electronic comics. It is also very important, but it needs further research work to be able to get practical application. the

Contents of the invention

The technical problem to be solved in the present invention is to identify the cartoon frame by judging the characteristics of the dividing line between the cartoon frames according to the characteristics of the general cartoon dividing line, and according to the arrangement characteristics of the cartoon frame, a combination of a binary tree and a cartoon frame is proposed. The sorting method is simple and easy to understand, and has high execution efficiency. the

In order to solve the above-mentioned technical problems, the invention provides a kind of caricature frame recognition and automatic sorting method based on dividing line, it is characterized in that, make following definition:

a. Comic frame: the smallest picture unit that cannot be divided in each page of comics;

b. Comic segment: a quadrilateral composed of at least two connected comic frames;

The method includes the following steps:

A. Preprocess the given manga page, including:

a. Binarize the comic page to obtain a binarized image;

b. Perform contour detection on the binarized image to obtain the foreground image of the comic page;

B. Perform straight line detection on the foreground image to obtain the straight line contained in the foreground image;

C. Define the quadrilateral structure obtained after straight line detection as coma, coma is a comic frame or comic segment, and comic segmentation starts from the largest coma. If coma satisfies one of the following conditions, it is determined that coma is indivisible and the picture corresponding to coma is a comic frame, execute step G, otherwise execute step D;

a. There is no straight line inside coma;

b. The ratio of the largest foreground image area inside the coma to the area of the coma is less than the threshold m, where 0<m<0.25;

c. The ratio of the number of pixels in the largest foreground image area inside the coma to the number of all pixels inside the coma is less than the threshold n, where 0<n<0.2;

d. The ratio of the length and width of the largest foreground image area inside the coma to the coma is less than the threshold p, 0.08<p<0.2;

D. Select a dividing line from the straight line obtained by the straight line detection, and the dividing line must meet the following conditions:

a. The dividing line only intersects the two opposite sides of the coma, and has no intersection with other sides;

b. The intersection of the dividing line and the coma is on the edge line of the coma;

E. Screen the dividing lines one by one in descending order of their weights, and select a dividing line whose number of intersections with other straight lines does not exceed 3, and define it as a standard dividing line;

F, within the width range of each HW pixel point on both sides of the standard dividing line, select a straight line that does not intersect with other straight lines, or only has intersection points with two straight lines at most, and then selects a straight line that is intersecting with the foreground image from these straight lines The straight line with the least number of intersection points is the most segmented position of the coma, which is defined as the best segmented line; if the best segmented line that meets the conditions cannot be selected, return to step E, and select the next best segmented line that meets the requirements according to the weight of the segmented line Dividing line;

G. Use the binary tree data structure to store the sub-processed coma according to the reading order of the comics;

H. After the comic page is divided for the first time according to the optimal dividing line, continue to divide and store the newly generated coma in the same way to the comic segment produced by the division, until there is no divisible coma, and one page of comic The split is complete. the

A non-parameter and non-supervised image segmentation threshold selection method is used to binarize the comic page, and a threshold value is selected to set the pixel value smaller than the threshold value to 0, and the pixel value greater than the threshold value to be set to 1. the

For the manga page foreground image obtained after contour detection, thinning is performed before straight line detection, and the lines in the foreground image with a line width greater than 1 pixel are thinned to only one pixel wide. the

The method of judging whether there is a straight line inside the coma is: if the amount of pixels that the straight line falls into the coma is less than half of the total number of pixels of the straight line, then the straight line is considered not to be inside the coma; otherwise, the straight line is considered to be inside the coma. the

Use the Kernel-based Hough transform method for line detection and determine the weight of the dividing line. the

The comic coma method for storing right-to-left reading order in a binary tree is as follows:

Define a binary tree, the relationship between the binary tree and the comic coma is that a coma is a node of the binary tree, the root node is a page of comics, the middle node is a comic segment, and the leaf node is a comic frame; if the optimal dividing line divides the coma into left and right two part, the coma on the right side of the best dividing line is taken as the left subtree, and the coma on the left side of the best dividing line is taken as the right subtree; The coma of the coma is used as the left subtree, and the coma below the best dividing line is used as the right subtree; if a coma does not have the best dividing line, the image corresponding to the coma is a comic frame, and there are no child nodes, so it is no longer divided; according to In the above method, each coma is stored in a binary tree, and all the leaves of the obtained binary tree are all comic frames, and the reading order of the comic frames is the arrangement order of the corresponding leaves from left to right. the

The comic coma method of storing the reading order from left to right in binary tree is as follows:

Define a binary tree, the relationship between the binary tree and the comic coma is that a coma is a node of the binary tree, the root node is a page of comics, the middle node is a comic segment, and the leaf node is a comic frame; if the optimal dividing line divides the coma into left and right two part, the coma on the left side of the best dividing line is taken as the left subtree, and the coma on the right side of the best dividing line is taken as the right subtree; if the best dividing line divides the coma into upper and lower parts, then the above The coma of the coma is used as the left subtree, and the coma below the best dividing line is used as the right subtree; if a coma does not have the best dividing line, the image corresponding to the coma is a comic frame, and there is no child node, so it is no longer divided; according to In the above method, each coma is stored in a binary tree, and all the leaves of the obtained binary tree are all comic frames, and the reading order of the comic frames is the arrangement order of the corresponding leaves from left to right. the

In the preprocessing of manga, binarization, contour detection, image thinning and then line detection make line detection faster; in the segmentation process, the segmentation line is selected by weight and the best segmentation line is selected, which improves the accuracy of segmentation line selection High efficiency; using the binary tree data structure to store and sort each frame that constitutes a cartoon, the operation is convenient and fast, and the practicability is strong; the whole scheme is fast, accurate and efficient. the

Description of drawings

Fig. 1 is a flowchart of the present invention;

Figure 2(a) is a schematic diagram of cartoon segmentation;

Fig. 2 (b) is the schematic diagram of the comic frame stored in the binary tree;

Figure 3 is a schematic diagram of the selection of the best dividing line;

Figure 4 is a statistical table of test results. the

Detailed ways

The present invention will be further described below in conjunction with accompanying drawing. the

As shown in Figure 1, it is a flowchart of the present invention, comprising the following steps:

A. Preprocess the given manga page, including:

a. Binarize the comic page to obtain a binarized image;

b. Perform contour detection on the binarized image to obtain the foreground image of the comic page;

B. Perform straight line detection on the foreground image to obtain the straight line contained in the foreground image;

C. Define the quadrilateral structure obtained after straight line detection as coma, coma is a comic frame or comic segment, and comic segmentation starts from the largest coma. If coma satisfies one of the following conditions, it is determined that coma is indivisible and the picture corresponding to coma is a comic frame, execute step G, otherwise execute step D;

a. There is no straight line inside coma;

b. The ratio of the largest foreground image area inside the coma to the area of the coma is less than the threshold m, where 0<m<0.25;

c. The ratio of the number of pixels in the largest foreground image area inside the coma to the number of all pixels inside the coma is less than the threshold n, where 0<n<0.2;

d. The ratio of the length and width of the largest foreground image area inside the coma to the coma is less than the threshold p, 0.08<p<0.2;

D. Select a dividing line from the straight line obtained by the straight line detection, and the dividing line must meet the following conditions:

a. The dividing line only intersects the two opposite sides of the coma, and has no intersection with other sides;

b. The intersection of the dividing line and the coma is on the edge line of the coma;

E, screen the dividing lines one by one in descending order of their weights, select the dividing line with no more than 3 intersection points with other straight lines, and define it as a standard dividing line;

F, within the width range of each HW pixel point on both sides of the standard dividing line, select a straight line that does not intersect with other straight lines, or only has intersection points with two straight lines at most, and then selects a straight line that is intersecting with the foreground image from these straight lines The straight line with the least number of intersection points is the best split position of coma, and is defined as the best split line; if the best split line that meets the conditions cannot be selected, return to step E, and select the next best split line that meets the requirements according to the weight of the split line. good dividing line;

G. Use the binary tree data structure to store the split coma according to the reading order of the comics. The coma here includes both the split coma, the split coma, and the indivisible coma;

H. After the comic page is divided for the first time according to the optimal dividing line, continue to divide and store the newly generated coma in the same way to the comic segment produced by the division, until there is no divisible coma, and one page of comic The split is complete. Among them, the definition is as follows, manga frame: the smallest picture unit that cannot be further divided in each page of manga. Comic Segment: A quadrilateral consisting of at least two connected comic frames. the

Using a binary tree to store the frame structure of comics has the same principle as humans judging the sorting of comics. A binary tree corresponds to a comic page. The root node is used to store a whole page of comics. First, the comics are divided into two parts and placed in the left and right subtrees of the root node. Continue to divide the part of each subtree until all the divisions are finally completed. The final result, that is, the divided manga frames are stored in the leaf nodes of the binary tree. the

As shown in Figure 2(a), the straight lines a, b, c, and d are the four best dividing lines to divide the comic into (1), (2), (3), (4), and (5) five comic frames , the binary tree shown in Figure 2(b) stores the above five comic frames as follows: first, the whole comic is regarded as the root of the tree, which is denoted as A, and A is composed of two parts, that is, the part above the optimal dividing line a And the following parts are represented by B and C respectively. Using a recursive method, B is divided into left and right parts E and D by the best dividing line c, C is divided into upper and lower parts F and G by the best dividing line d, and F and G cannot be further divided into leaves. Finally, the best dividing line c divides E into the last upper and lower M and N leaves. We can use brackets to represent a binary tree, namely (A(B(D, E(M, N)), C(F, G)))

We use a binary tree data structure to store each frame of the comics according to the sorting of the comics, and the comics can be sorted directly. Here we use the Japanese manga layout (reading order is from right to left, from top to bottom) as an example, we use a binary tree to store each frame according to the following rules:

(a), if the best dividing line is the upper and lower lines: the part above the best dividing line needs to be used as the left subtree, for example, the B, F, and M parts in Figure 2 (b) are all in the left subtree part; the most The part below the optimal dividing line is used as the right subtree, for example, the C, G, and N parts in Figure 2(b) are all in the right subtree. the

(b), if the best dividing line is the left and right lines: the part on the right side of the best dividing line should be used as the left subtree, such as part D in Figure 2 (b); the part below the best dividing line should be used as the right subtree. Subtree, such as part E in Fig. 2(b). the

All the leaves of the binary tree stored according to the above rules are the final segmentation results, and have been sorted from left to right. All the leaf nodes of the binary tree in Figure 2(b) are the final segmentation results, the reading order is D, M, N, F, G, and the order of the frames is exactly (1)(2)(3)(4)( 5). If the reading habit of comics is from right to left (change from left to right), such as domestic comics, the same method can be used to divide and sort, and only need to exchange the storage positions of the comic frames on both sides of the best left and right dividing line. the

As shown in Figure 3, the straight line x is the standard dividing line we found, but in order to avoid damaging the content of the comic, we divide it at the straight line y in the blank space on the right, and y is the best dividing line. the

Selection of the best dividing line position: within the range of the current straight line left and right (or up and down) HW, among the straight line groups parallel to the current standard dividing line, filter out the straight lines with more than 2 intersection points with other straight lines, so that the remaining A straight line is a line that does not intersect with other straight lines, or only intersects with one or two straight lines. Among these straight lines, select a straight line with the least number of intersecting points with foreground pixels as the best dividing line. The thresholds of the split ratios in the three cases are respectively specified for the candidate split lines, where the split ratio refers to the ratio of the actual length of the split line to the theoretical length of the split line. As shown in Figure 3, since the straight lines m and n are not cut off in the middle, their actual length is consistent with the theoretical length, and the division ratio is 1. If the split line is cut off in the middle, the split ratio is less than 1. Define the split ratio of the alternative split line, the split ratio of the split line that has no intersection with other straight lines is defined as RATIO, the split ratio of the split line with only one intersection point with other straight lines is defined as RATIO1, and the split ratio of the split line with two intersection points with other straight lines Defined as RATIO2. If the candidate splitting line has no intersection with other straight lines, then we choose a splitting line that intersects the least with the foreground and the splitting ratio exceeds RATIO as the best splitting line. If the alternative dividing line has only one intersection point with other straight lines, then we choose a dividing line with the least intersection with the foreground and the dividing ratio exceeding RATIO1 as the best dividing line. If the candidate segmentation line has two intersections with other straight lines, then we choose a segmentation line with the least intersection with the foreground and a segmentation ratio exceeding RATIO2 as the optimal segmentation line. The range of RATIO is 0.45-0.55, the range of RATIO1 is 0.7-0.8, and the range of RATIO2 is 0.5-0.6. After the specific position of the optimal dividing line is determined, we can divide according to this position. If there is no division position that meets the above requirements, then go back and continue to select the standard division line. The selected dividing line and the excluded dividing line are marked as used, and the list of available dividing lines is updated. the

The specific method of comic image preprocessing is as follows:

Binarization: All the pictures used in the present invention are pictures after binarization, and the binarization of the image is conducive to the further processing of the image, making the image simple, and the amount of data is reduced, which can highlight the interesting The silhouette of the target. To process and analyze the binary image, firstly, the grayscale image should be binarized to obtain the binarized image. All images used in the present invention are binarized caricature pictures. The present invention uses the method of extracting the optimal threshold from the gray histogram according to certain rules to binarize the image, the method is as follows: the present invention uses a non-parameter and non-supervised image proposed by NOBUYUKI OTSU for the binarization of the image Segmentation threshold selection method ^[9] . By selecting a threshold, the pixel value smaller than the threshold is set to 0, and the pixel value greater than this threshold is set to 1, which is the binarization of a picture. The method we use can maximize the separation of gray-level classification results, and an optimal threshold can be obtained through this criterion. The process is very simple, only the 0-order and 1-order cumulative moments of the gray histogram are used. Since the binarization adopts the existing technology, the specific method will not be repeated.

Contour detection: In image preprocessing, it is often necessary to track the edge of the target, also called contour tracking. As the name implies, contour tracking is to track the boundary by sequentially finding edge points. In the selection of the dividing line, it is necessary to judge the number of intersections between the dividing line and the foreground image. The present invention performs contour detection on the binarized cartoon image to detect the foreground image in the cartoon. When judging the intersection of the foreground image and the dividing line, It plays an important role in the selection of the dividing line. More importantly, after the contour detection, the line detection is performed from the contour, which saves the steps of detecting all points, and only needs to calculate the pixels on the contour, which reduces a lot of calculations for line detection. Because what adopt is prior art, no longer describe in detail. the

Image thinning: Image thinning generally appears as an image preprocessing technique, the purpose is to extract the skeleton of the source image, that is, to thin the lines in the original image with a line width greater than 1 pixel to only one pixel wide, forming a "skeleton" , after forming the skeleton, it is easier to analyze the image, such as extracting the features of the image. The basic idea of thinning is "layer-by-layer stripping", that is, stripping from the edge of the line layer by layer until there is only one pixel left in the line. Image thinning greatly compresses the data volume of the original image and keeps the basic topology of its shape unchanged, thus laying the foundation for image feature extraction and other applications. The present invention refines the image after contour extraction so that we can greatly reduce the number of intersection points between the foreground and the dividing line. At the same time, because the intersection of all dividing lines and the foreground image is proportionally reduced, it will not affect our calculation of the dividing line. case of judgment. the

Straight line detection: the following technology is used, because it is an existing technology, here is only a brief introduction: the dividing line in the comic is the key to determine the segmentation of the comic. The segmentation of comics needs to detect the straight line first, and then judge whether it is a dividing line and how to divide it according to the characteristics of the straight line. If there are too few detected straight lines, the cartoon will be "missed". If there are too many or inaccurate detected straight lines, the segmentation process will do a lot of judgment work, which will easily lead to "misclassification". Therefore, finding a suitable line detection method to correctly detect the dividing line will make the dividing process greatly simplified and accurate. Hough transform is one of the basic methods for recognizing geometric shapes from images in image processing, that is, it can detect objects of known shapes and is less affected by noise and curve discontinuities. The Hough line detection technique is widely used because of its robustness against noise and erroneous data, but the computational cost is high, so it cannot be performed in real time for larger images. The invention adopts an improved Hough straight line detection method. The Kernel-based Hough transform (KHT) method proposed by Leandro A.F.Fernandes and Manuel M.Oliveira to detect straight lines. Because this method first forms clusters of approximately collinear pixels, and then selects the most consistent straight line, it is much less expensive than the Hough line detection method, but the accuracy rate is higher. The straight line detection in the present invention is performed on the basis of performing contour detection on the image after image binarization, and then thinning the contour, so the straight line detection speed in the present invention will be faster. the

For some irregular pictures, the present invention designs a human-assisted method to make the picture segmentation more accurate. The artificial straight line is firstly detected to judge the manually added straight line. the

Test results: The experimental results are shown in Figure 4, 184 pieces were successful, 16 pieces were failed, 2 pieces were over-scored, 13 pieces were under-scored, and 1 piece was mis-scored, reaching a success rate of 92%. The computer takes 926 seconds, that is, each takes 4.633 seconds. The efficiency is 0.215 sheets/second. the

Of course, the present invention can also have other various embodiments, and those skilled in the art can make various corresponding changes and deformations according to the present invention without departing from the spirit and essence of the present invention, but these corresponding Changes and deformations should belong to the scope of protection of the appended claims of the present invention. the

Claims (10)

1. the caricature frame based on cut-off rule is discerned and auto ordering method, it is characterized in that this method may further comprise the steps:

A, given caricature page or leaf is done pre-service, comprising:

A, to said caricature page or leaf binaryzation, obtain binary image;

B, said binary image is made profile detect, obtain the foreground image of caricature page or leaf;

B, foreground image is made straight-line detection, obtain the straight line that comprises in the foreground image;

The quadrilateral structure body that C, definition are obtained after the straight-line detection is coma, and coma is caricature frame or caricature section, and wherein the caricature frame refers to minimal graph blade unit in every page of caricature, and the caricature section refers to the quadrilateral be made up of at least two continuous caricature frames; Caricature is cut apart from the coma of maximum and is begun, if coma satisfies one of following condition, judges that then the picture that coma is indivisible and coma is corresponding is the caricature frame, execution in step G, otherwise execution in step D, and wherein the caricature frame refers to minimal graph blade unit in every page of caricature;

A, coma inside do not have straight line;

B, the inner maximum foreground image area of coma and coma area ratio are less than threshold value m;

The ratio of c, the inner maximum foreground image areas pixel quantity of coma and inner all pixel quantities of coma is less than threshold value n;

The inner maximum foreground image areas of d, the coma length and width ratio corresponding with coma is less than threshold value p;

D, from the straight line that straight-line detection is obtained, select cut-off rule, said cut-off rule must meet the following conditions:

A, said cut-off rule only with coma in two relative limits intersect, do not have intersection point with other limits;

The intersection point of b, said cut-off rule and coma is on the edge line of coma;

E, the descending of said cut-off rule by its weight screened one by one, select with other straight-line intersection numbers smaller or equal to 3 cut-off rule, be defined as the standard cut-off rule;

F, in the width range of each HW pixel of said standard cut-off rule both sides; Select non-intersect or maximum straight lines that intersection point is only arranged with two straight lines with other straight line; From these straight lines, select again one with the optimal segmentation position of the minimum straight line of foreground image number of hits as coma, be defined as the optimal segmentation line; If can not select qualified optimal segmentation line, return step e, select the satisfactory optimal segmentation line of next bar according to the weight of said cut-off rule;

G, employing binary tree data structure store coma according to the reading order of caricature, and the coma had here both comprised the coma of cutting apart, and also comprised cutting apart the coma that the back produces, and also comprised indivisible coma;

H, according to the optimal segmentation line caricature page or leaf is done to cut apart for the first time after, the caricature section that produces cutting apart of using the same method continues to cut apart and stores the coma of new generation, till not having alienable coma, one page caricature cut apart completion.

2. identification of caricature frame and auto ordering method based on cut-off rule according to claim 1; It is characterized in that: the image segmentation threshold value system of selection of adopting printenv and non-supervision is to said caricature page or leaf binaryzation; Through selecting a threshold value to come to be set to 0, be set to 1 greater than the pixel value of this threshold value less than the pixel value of threshold value.

3. identification of caricature frame and auto ordering method based on cut-off rule according to claim 1; It is characterized in that: the caricature page or leaf foreground image that obtains after profile is detected; Make micronization processes before the straight-line detection, line thickness in the foreground image is refined into greater than the lines of 1 pixel has only a pixel wide.

4. identification of caricature frame and auto ordering method based on cut-off rule according to claim 1, it is characterized in that: the scope of said threshold value m is: 0＜m＜0.25.

5. identification of caricature frame and the auto ordering method based on cut-off rule according to claim 1, it is characterized in that: the scope of said threshold value n is: 0＜n＜0.2 wherein.

6. identification of caricature frame and auto ordering method based on cut-off rule according to claim 1, it is characterized in that: the scope of said threshold value p is: 0.08＜p＜0.2.

7. identification of caricature frame and auto ordering method based on cut-off rule according to claim 1; It is characterized in that: judge whether coma inside has the method for straight line to be: if straight line falls into the inner amount of pixels of coma half the less than straight line pixel total amount; Think that then this straight line is not inner at coma, otherwise think that this straight line is inner at coma.

8. identification of caricature frame and auto ordering method based on cut-off rule according to claim 1 is characterized in that: adopt Kernel-based Hough transform method to make the weight of straight-line detection, also definite cut-off rule.

9. identification of caricature frame and auto ordering method based on cut-off rule according to claim 1, it is characterized in that: the method for reading order coma is to adopt binary tree data structure to store from right to left:

Define a binary tree, the relation of binary tree and caricature coma is that a coma is the node of binary tree, and root node is one page caricature, and intermediate node is the caricature section, and leaf node is the caricature frame; Two parts about if optimal segmentation bundle of lines coma is divided into, then with the coma on the optimal segmentation line right side as left subtree, the coma on the optimal segmentation line left side is as right subtree; If optimal segmentation bundle of lines coma is divided into up and down two parts, then the coma above the optimal segmentation line as left subtree, the coma below the optimal segmentation line is as right subtree; If certain coma does not have the optimal segmentation line, the image that then this coma is corresponding is the caricature frame, does not have child node, no longer it is cut apart; According to the method described above each coma all is stored in the binary tree, all leaves of the binary tree that obtains are all caricature frames, and the reading order of caricature frame is putting in order of corresponding leaf from left to right.

10. identification of caricature frame and auto ordering method based on cut-off rule according to claim 1, it is characterized in that: the caricature coma method of reading order is to adopt binary tree data structure to store from left to right:

Define a binary tree, the relation of binary tree and caricature coma is that a coma is the node of binary tree, and root node is one page caricature, and intermediate node is the caricature section, and leaf node is the caricature frame; Two parts about if optimal segmentation bundle of lines coma is divided into, then with the coma on the optimal segmentation line left side as left subtree, the coma on the optimal segmentation line right side is as right subtree; If optimal segmentation bundle of lines coma is divided into up and down two parts, then the coma above the optimal segmentation line as left subtree, the coma below the optimal segmentation line is as right subtree; If certain coma does not have the optimal segmentation line, the image that then this coma is corresponding is the caricature frame, does not have child node, no longer it is cut apart; According to the method described above each coma all is stored in the binary tree, all leaves of the binary tree that obtains are all caricature frames, and the reading order of caricature frame is putting in order of corresponding leaf from left to right.

Images