CN101576958A - Extraction method of rectangle seal graphics with border based on form feature - Google Patents

Abstract

The invention discloses an extraction method of rectangle seal graphics with border based on form feature. The method comprises the following steps: (1) screening connected components of a target image foreground layer, and removing small connected components dissatisfying seal dimension specification in the target image; (2) screening connected components of a target image background layer and removing the connected components which are far larger than or smaller than the seal dimension specification; (3) adopting watershed algorithm in image segmentation to extract skeleton of closed graphics in the target image, using the longest tour length of skeleton point in the seal to judge rectilinear direction and remove flash; (4) extracting the seal according to the seal dimension specification provided by users. The method of the invention can fast and effectively detect a plurality of rectangle seals with unfixed number in the image, and has the advantages of high detection precision, fast speed, strong anti-interference and the like.

Description

A kind of band edge frame rectangle seal graphics extracting method based on shape facility

Technical field:

The invention belongs to the text image process field, relate to a kind of extracting method of the band edge frame seal graphics based on shape facility, especially a kind of extraction and recognition methods regular figure.

Background technology:

In recent years, the part scholar has carried out successively about the automatic detection of seal and the research of extraction.Notice that the seal graphics in the bill is generally red, these researchs bias toward the extraction of carrying out seal graphics by color characteristic.By image transitions is arrived certain color space,, adopt the Threshold Segmentation technology to extract seal then as the HSI space.Because seal also may be subjected to the interference of objects such as red writing line, therefore, also need to carry out the removal work of red writing line etc. usually.The major defect of this method is: image is after scanning, and its color tends to drift about, and, because seal is added a cover usually on paper document, unavoidably can overlap with other objects.Because be subjected to the influence of ink paste diffusional effect in paper document, saltus step usually can take place in the color of seal lap.Therefore phenomenons such as incompleteness, breakage can appear in the seal graphics that adopts color characteristic to extract.In addition, in processes such as bill fax, duplicating, colouring information can be dropped or destroy usually.So it is inadequate often only to adopt color characteristic to carry out the extraction of seal.

The shape facility of seal is another very important feature.The HOUGH conversion usually is used for detecting the seal of certain designated shape in image.Yet the common speed of this method is slower, is difficult to carry out real-time detection, and when seal number in the image not fixedly the time, the HOUGH transform method can occur many flase drops usually and survey.This mainly is because when determining local extremum in the array that adds up of HOUGH conversion, threshold value is difficult to effectively choose.We propose a kind of not based on the fast detecting and the extracting method of the seal graphics of color characteristic.This method is primarily aimed at the detection of frame seal, comes rectangle seal graphics in the location scanning image by analyzing its frame shape facility.This method belongs to a kind of general seal and detects and extracting method, can detect and extract the seal graphics of right angle rectangle fast.

Summary of the invention:

The objective of the invention is to overcome the shortcoming of above-mentioned prior art, a kind of extracting method of the band edge frame rectangle seal graphics based on shape facility is provided, according to following steps:

(1) target image foreground layer connected component is screened, remove the little connected component of discontented footmark chapter dimensions in the target image;

(2) destination image background layer connected component screened, reject destination image background layer connected component much larger than or much smaller than the connected component of seal dimensions

(3) watershed algorithm in the employing image segmentation extracts the skeleton of closed figure in the target image, utilizes the longest brigade commander of skeleton point in seal inside to judge rectilinear direction and removal burr;

(4) the seal dimensions that provides according to the user is carried out the seal extraction.

Described step (1) is according to following steps: earlier bianry image I is carried out connected component labeling, obtain all connected components in the image; Again each connected component among the bianry image I is carried out inverse and handle, then inverse figure is carried out connected component labeling, with all obtain the connected region that outer boundary surrounded of connected component among the former figure greater than 1 mark value zero clearing in the signature; Secondly the connected region that obtains is carried out the morphological erosion operation, structural elements adopts circular configuration unit, and radius is made as r ₁Last area of detection is greater than 1 connected component on the image after the corrosion, and the connected component in the pairing original image of these connected components satisfies the screening requirement; All connected components that do not meet the demands are removed from former figure.

Described step (2) is according to following steps: earlier bianry image I is carried out negate, obtain inverse figure I; Again inverse figure I is carried out connected component labeling, and analyze the connected region that obtains one by one; Secondly meet the demands for all and need the connected region C of reservation _i, note $\stackrel{\‾}{U}=\underset{i}{\∪}{\stackrel{\‾}{C}}_i,$ U is carried out the morphology expansive working, then it is covered on the I as template; Be changed to the image background look for the outer pixel of image cope match-plate pattern then, then remain unchanged for the pixel in the image cope match-plate pattern; At last image is carried out inverse and handle, obtain the result after final connected component screens.

Described step (3) is to utilize watershed algorithm in the image segmentation to extract the skeleton of closed figure in the target image earlier, carry out the black brigade commander calculating of all directions and the longest black brigade commander's of record length for each pixel on the skeleton that extracts, setting threshold is removed the skeleton point that the longest black brigade commander length is lower than threshold value then.

Extraction is meant according to the rectangle seal in the described step (4): at first by satisfying the straight-line segment of seal size dimension constraint in the HOUGH change detection connected component, judge by the mutual corresponding relation of analyzing between these straight-line segments whether they constitute rectangle then, at last the seal dimensions that provides according to the user is determined the scope at the scope of scope, minor face on the long limit of the rectangle seal that extracted and seal pitch angle and is extracted the rectangle seal that meets the demands.

The present invention has adopted the thinking of local coarse positioning in conjunction with the back checking, and there are not many pieces of rectangle seals of fixed number in this method in the detected image fast and effectively.This method has advantages such as accuracy of detection height, speed are fast, strong interference immunity, is a kind of practical technique that seal graphics detects and extracts automatically.

Description of drawings:

The rectangle seal graphics of multiple shape in Fig. 1 file and picture of the present invention

Fig. 2 the present invention contains the bianry image of many pieces of seal graphics

Fig. 3 connected component of the present invention outer boundary area surrounded

Fig. 4 the present invention is based on the connected component after the foreground layer Analysis and Screening

Fig. 5 the present invention is based on the connected component after the background layer Analysis and Screening

Burr phenomena in Fig. 6 connected component of the present invention

The skeleton diagram of Fig. 7 traditional method for extracting of the present invention

The skeleton diagram that Fig. 8 watershed algorithm of the present invention extracts

Fig. 9 skeleton diagram of the present invention

The longest black brigade commander of Figure 10 rectangular edges internal point of the present invention;

The longest black brigade commander of Figure 11 burr internal point of the present invention;

Figure 12 skeleton diagram of the present invention;

Figure 13 the present invention removes figure behind the burr according to the longest black brigade commander;

Figure 14 skeleton diagram of the present invention;

Figure 15 the present invention is based on the straight-line detection result of black brigade commander HOUGH conversion;

The calculating of parallel relation between Figure 16 straight-line segment of the present invention;

The calculating of vertical relation between Figure 17 straight-line segment of the present invention;

Rectangle seal in Figure 18 file and picture of the present invention detects.

Embodiment:

Below in conjunction with accompanying drawing the present invention is done and to describe in further detail:

Referring to Fig. 1-17, a kind of extracting method of the band edge frame rectangle seal graphics based on shape facility, according to following steps:

(1) target image foreground layer connected component is screened, remove the little connected component of discontented footmark chapter dimensions in the target image;

(2) destination image background layer connected component screened, reject destination image background layer connected component much larger than or much smaller than the connected component of seal dimensions

(3) watershed algorithm in the employing image segmentation extracts the skeleton of closed figure in the target image, utilizes the longest brigade commander of skeleton point in seal inside to judge rectilinear direction and removal burr;

(4) the seal dimensions that provides according to the user is carried out the seal extraction.

Described step (1) is according to following steps: earlier bianry image I is carried out connected component labeling, obtain all connected components in the image; Again each connected component among the bianry image I is carried out inverse and handle, then inverse figure is carried out connected component labeling, with all obtain the connected region that outer boundary surrounded of connected component among the former figure greater than 1 mark value zero clearing in the signature; Secondly the connected region that obtains is carried out the morphological erosion operation, structural elements adopts circular configuration unit, and radius is made as r ₁Last area of detection is greater than 1 connected component on the image after the corrosion, and the connected component in the pairing original image of these connected components satisfies the screening requirement; All connected components that do not meet the demands are removed from former figure.In bianry image I,, remember that the zone that its outer boundary surrounds is S for its certain C of connected component.If the radius r of the maximum inscribed circle of S _i＜b ₁/ 2, then must not comprise the designated seal figure among the C.

Described step (2) is according to following steps: earlier bianry image I is carried out negate, obtain inverse figure I; Again inverse figure I is carried out connected component labeling, and analyze the connected region that obtains one by one; Secondly meet the demands for all and need the connected region C of reservation _i, note $\stackrel{\‾}{U}=\underset{i}{\∪}{\stackrel{\‾}{C}}_i,$ U is carried out the morphology expansive working, then it is covered on the I as template; Be changed to the image background look for the outer pixel of image cope match-plate pattern then, then remain unchanged for the pixel in the image cope match-plate pattern; At last image is carried out inverse and handle, obtain the result after final connected component screens.In bianry image I, for its certain C of connected component, if the circumradius r of C ₀＞a ₂/ 2, then C must not be contained among the designated seal figure.For the white space that the image lastblock is communicated with, if its size is too big, can judge that then it must not belong to the interior zone of specifying seal graphics, therefore, comprise the further division of these regional connected component's needs.After connected component's screening, the connected component that does not contain seal graphics in the image has obtained effective removing.The screening process that it is pointed out that connected component is equivalent to former figure is divided into a plurality of less image blocks, may include seal graphics to be detected in each image block.When containing the connected component of a large amount of discontented footmark chapter dimension constraint conditions in the image, this step can effectively be dwindled the seal graphics sensing range usually greatly, thereby can improve the detection efficiency of seal graphics significantly.Next, we then need each connected component that filters out is analyzed one by one, verify whether it comprises real seal graphics.

Described step (3) is to utilize watershed algorithm in the image segmentation to extract the skeleton of closed figure in the target image earlier, carry out the black brigade commander calculating of all directions and the longest black brigade commander's of record length for each pixel on the skeleton that extracts, setting threshold is removed the skeleton point that the longest black brigade commander length is lower than threshold value then.For each connected component that filters out, need analyze its boundary shape feature one by one and verify whether it comprises the designated seal figure.Yet; because seal graphics usually covers with other objects such as literal; usually many burrs can have been attached in the connected component that filters out; when these connected components are extracted skeleton; near the burr skeleton tends to take place more serious distortion, thereby the analysis of boundary shape is exerted a certain influence.Therefore, need at first connected component to be carried out some processing.

Closure hypothesis according to the seal graphics border is understood that, the inevitable part that does not belong to the seal border of the non-sealing lines in the connected component.Therefore, when extracting the skeleton of connected component, can remove these redundant line.Here, we adopt watershed algorithm classical in the image segmentation to extract the skeleton of closed figure.Compare with traditional framework extraction method, do not have burr on the skeletal graph that this method obtains, and the distortion that the skeleton that extracts produces is also less usually.Provided the effect comparison that each connected component among Fig. 3 is adopted the skeleton diagram of thinning algorithm and watershed algorithm extraction respectively among Fig. 4.After adopting watershed algorithm to extract skeleton, we have further left out the little connected component separated from one another of discontented full size cun constraint.

Extraction is meant according to the rectangle seal in the described step (4): at first by satisfying the straight-line segment of seal size dimension constraint in the HOUGH change detection connected component, judge by the mutual corresponding relation of analyzing between these straight-line segments whether they constitute rectangle then, at last the seal dimensions that provides according to the user is determined the scope at the scope of scope, minor face on the long limit of the rectangle seal that extracted and seal pitch angle and is extracted the rectangle seal that meets the demands.For the detection of right angle rectangle seal, detection of straight lines in image at first then then by analyzing the relation between the straight line, as parallel, vertical etc., judges whether they constitute rectangle.

The detection method of right angle rectangle seal is as follows in the connected component: at first detect the straight-line segment that satisfies the constraint of seal size dimension in the connected component, by analyzing the mutual corresponding relation between these straight-line segments, judge as parallel, vertical, the crossing grade whether they constitute rectangle then.

For making things convenient for the judgement of straight-line segment relation, need at first set up the syntople matrix of a straight-line segment.This syntople matrix is the matrix of n * n, is designated as { R _{I, j}} _{1≤i, j≤n}, wherein n is the number of straight-line segment.Each element R in the matrix _{I, j}Rower or row marks corresponding straight line section respectively.Stored parallel relation between the straight-line segment and the vertical relation between the straight-line segment in the last diagonal matrix of this matrix and the following diagonal matrix respectively, the size of element value has then reflected parallel or vertical mutual spatial relation between two straight-line segments in the matrix.

The calculating of straight-line segment relational matrix as shown in Figure 8.For two straight-line segment L that are parallel to each other _i(ρ _i, θ _iA _i, B _i) and L _j(ρ _j, θ _jA _j, B _j), θ wherein _i=θ _j=θ and i＞j, then R _{I, j}Be the length of two straight-line segments projection lap on parallel lines.R _{I, j}Size reflected whether these two straight-line segments may constitute two opposite side of rectangle.(ρ, θ) (x, y), this point is at straight line L (ρ, θ) the subpoint P on for Wai 1 P for straight line L ^*(x ^*, y ^*) can calculate by following formula:

$\left\{\begin{array}{c}{x}^{*}=x+(x_0-x)\cos \mathrm{\θ}+(y_0-y)\sin \mathrm{\θ}\\ y^{*}=y+(x_0-x)\cos \mathrm{\θ}+(y_0-y)\sin \mathrm{\θ}\end{array}\right.,$

In the formula:

(x ₀, y ₀)---straight line L (ρ, θ) on more arbitrarily.

If || A _iB _i|| 〉=|| A _jB _j||, so, if note A _jAnd B _jAt straight-line segment L _iProjection be respectively A _j ^*And B _j ^*, R then _{I, j}Can be explicit be calculated as follows:

$R_{i,j}=\left\{\begin{array}{c}0,\mathrm{if}{A}_j^{*}\∉\stackrel{\‾}{A_i{B}_i}\mathrm{and}{B}_j^{*}\∉\stackrel{\‾}{A_i{B}_i}\\ \left|\right|A_j^{*}-B_i\left|\right|,\mathrm{if}{A}_j^{*}\∈\stackrel{\‾}{A_i{B}_i}\mathrm{and}{B}_j^{*}\∉\stackrel{\‾}{A_i{B}_i}\\ \left|\right|A_i-B_j^{*}\left|\right|,\mathrm{if}{A}_j^{*}\∉\stackrel{\‾}{A_i{B}_i}\mathrm{and}{B}_j^{*}\∈\stackrel{\‾}{A_i{B}_i}\\ \left|\right|A_j^{*}-B_j^{*}\left|\right|,\mathrm{if}{A}_j^{*}\∈\stackrel{\‾}{A_i{B}_i}\mathrm{and}{B}_j^{*}\∈\stackrel{\‾}{A_i{B}_i}\end{array}.\right.$

For two orthogonal straight-line segment L _i(ρ _i, θ _iA _i, B _i) and L _j(ρ _j, θ _jA _j, B _j), wherein ${\mathrm{\θ}}_i={\mathrm{\θ}}_j+\frac{\mathrm{\π}}{2}$ And i＜j, R _{I, j}Value can just can bear.If two straight-line segments truly intersect, then R _{I, j}For just, otherwise for negative.Its size has reflected the intersection point of two straight vertical line segments and the distance relation of straight-line segment end points.For two orthogonal straight line L _i(ρ _i, θ _i) and L _j(ρ _j, θ _j), its intersection point C (x ₀, y ₀) can calculate by following formula:

$\left\{\begin{array}{c}{\mathrm{<figure-callout\; id="0"\; label="x"\; filenames="A20091015081600161.png"\; state="\{\{state\}\}">x</figure-callout>}}_0={\mathrm{\&rho;}}_i\cos {\mathrm{\&theta;}}_i+{\mathrm{\&rho;}}_j\sin {\mathrm{\&theta;}}_i\\ y_0={\mathrm{\&rho;}}_i\sin {\mathrm{\&theta;}}_i+{\mathrm{\&rho;}}_j\cos {\mathrm{\&theta;}}_i\end{array}\right.,$

So, R _{I, j}Can be explicit be calculated as follows:

$R_{i,j}=\left\{\begin{array}{cc}-\max \left\{\min \right\{\left|\right|C-A_i\left|\right|,\left|\right|C-B_i\left|\right|\},\min \{\left|\right|C-A_j\left|\right|,\left|\right|C-B_j\left|\right|\left\}\right\},& \\ & \mathrm{ifC}\&NotElement;\stackrel{\&OverBar;}{A_i{B}_i}\mathrm{andC}\&NotElement;\stackrel{\&OverBar;}{A_j{B}_j}\\ -\min \left\{\right||C-A_j||,||C-B_j|\left|\right\},& \\ & \mathrm{ifC}\&Element;\stackrel{\&OverBar;}{A_i{B}_i}\mathrm{andC}\&NotElement;\stackrel{\&OverBar;}{A_j{B}_j}\\ -\min \left\{\right||C-A_i||,||C-B_i|\left|\right\},& \\ & \mathrm{ifC}\&NotElement;\stackrel{\&OverBar;}{A_i{B}_i}\mathrm{andC}\&Element;\stackrel{\&OverBar;}{A_j{B}_j}\\ \min \left\{\min \right\{\left|\right|C-A_i\left|\right|,\left|\right|C-B_i\left|\right|\},\min \{\left|\right|C-A_j\left|\right|,\left|\right|C-B_j\left\}\right\},& \\ & \mathrm{ifC}\&Element;\stackrel{\&OverBar;}{A_i{B}_i}\mathrm{andC}\&Element;\stackrel{\&OverBar;}{A_j{B}_j}\end{array}\right..$

After obtaining the syntople matrix of straight-line segment, the detection problem of right angle rectangle seal changes into the subgraph search problem, promptly searches for the subgraph that satisfies given restriction relation in a figure.

The subgraph search algorithmic method that the right angle rectangle detects is as follows:

Step 1. detection of straight lines sections, and the syntople matrix of structure straight-line segment, obtaining with the straight-line segment is the figure on summit.The search on all summits is labeled as 0 among the initialization figure;

Each search among 2. couples of figure of Step is labeled as 0 summit, and alternately search is corresponding with current summit and satisfy the straight vertical line segment or the Line Segment of seal dimension constraint condition;

Step 3. then continues the next corresponding summit of search if this summit search mark is not 0;

If this summit search is labeled as 0,

Then the straight-line segment of judging this summit correspondence by the syntople matrix whether with current straight-line segment true vertical or parallel:

If not, continue the next corresponding summit of search;

If judge whether the vertex sequence of current search can constitute a rectangle:

If, then store this rectangle and also the search mark on this summit is changed to 1, return Step 2 and continue the search on next summit;

If not, the search mark on this summit is changed to 1, returns Step 2 and continue the search on next summit;

Step 4. repeats Step 2 and Step 3, and all summits have been searched and finished in figure.

When a plurality of rectangles are common edge with a long straight-line segment, if at first begin subgraph search from this long straight-line segment, then may the holiday rectangle.For avoiding this situation, when carrying out subgraph search, can at first begin search from summit with low number of degrees.Perhaps, when mark has been searched in the summit in to figure, when the searching times on and if only if this summit reaches this degree of vertex, again it has been searched for mark.Because our handled object is mainly the binary and grayscale image, therefore the seal that meets the demands that we mainly adopt red-label to detect for result's demonstration shows.

Above content is to further describing that the present invention did in conjunction with concrete preferred implementation; can not assert that the specific embodiment of the present invention only limits to this; for the general technical staff of the technical field of the invention; without departing from the inventive concept of the premise; can also make some simple deduction or replace, all should be considered as belonging to the present invention and determine scope of patent protection by claims of being submitted to.

Claims (5)

1, a kind of extracting method of the band edge frame rectangle seal graphics based on shape facility is characterized in that, according to following steps:

(1) target image foreground layer connected component is screened, remove the little connected component of discontented footmark chapter dimensions in the target image;

(2) destination image background layer connected component screened, reject destination image background layer connected component much larger than or much smaller than the connected component of seal dimensions;

(3) watershed algorithm in the employing image segmentation extracts the skeleton of closed figure in the target image, utilizes the longest brigade commander of skeleton point in seal inside to judge rectilinear direction and removal burr;

(4) the seal dimensions that provides according to the user is carried out the seal extraction.

2, according to the extracting method of the described a kind of band edge frame rectangle seal graphics based on shape facility of claim 1, it is characterized in that, described step (1) is according to following steps: earlier bianry image I is carried out connected component labeling, obtain all connected components in the image; Again each connected component among the bianry image I is carried out inverse and handle, then inverse figure is carried out connected component labeling, with all obtain the connected region that outer boundary surrounded of connected component among the former figure greater than 1 mark value zero clearing in the signature; Secondly the connected region that obtains is carried out the morphological erosion operation, structural elements adopts circular configuration unit, and radius is made as r ₁Last area of detection is greater than 1 connected component on the image after the corrosion, and the connected component in the pairing original image of these connected components satisfies the screening requirement; All connected components that do not meet the demands are removed from former figure.

3, according to the extracting method of the described a kind of band edge frame rectangle seal graphics based on shape facility of claim 1, it is characterized in that described step (2) is according to following steps: earlier bianry image I is carried out negate, obtain inverse figure I; Again inverse figure I is carried out connected component labeling, and analyze the connected region that obtains one by one; Secondly meet the demands for all and need the connected region C of reservation _i, note $\stackrel{\&OverBar;}{U}=\underset{i}{\&cup;}{\stackrel{\&OverBar;}{C}}_i,$ U is carried out the morphology expansive working, then it is covered on the I as template; Be changed to the image background look for the outer pixel of image cope match-plate pattern then, then remain unchanged for the pixel in the image cope match-plate pattern; At last image is carried out inverse and handle, obtain the result after final connected component screens.

4, according to the extracting method of the described a kind of band edge frame rectangle seal graphics based on shape facility of claim 1, it is characterized in that: described step (3) is to utilize watershed algorithm in the image segmentation to extract the skeleton of closed figure in the target image earlier, carry out the black brigade commander calculating of all directions and the longest black brigade commander's of record length for each pixel on the skeleton that extracts, setting threshold is removed the skeleton point that the longest black brigade commander length is lower than threshold value then.

5, extracting method according to the described a kind of band edge frame rectangle seal graphics based on shape facility of claim 1, it is characterized in that: extraction is meant according to the rectangle seal in the described step (4): at first by satisfying the straight-line segment of seal size dimension constraint in the HOUGH change detection connected component, judge by the mutual corresponding relation of analyzing between these straight-line segments whether they constitute rectangle then, at last the seal dimensions that provides according to the user is determined the scope on the long limit of the rectangle seal that extracted, the scope at the scope of minor face and seal pitch angle also extracts the rectangle seal that meets the demands.

Images