CN102087741B - Method and system for processing image by using regional architecture - Google Patents
Method and system for processing image by using regional architecture Download PDFInfo
- Publication number
- CN102087741B CN102087741B CN 200910252678 CN200910252678A CN102087741B CN 102087741 B CN102087741 B CN 102087741B CN 200910252678 CN200910252678 CN 200910252678 CN 200910252678 A CN200910252678 A CN 200910252678A CN 102087741 B CN102087741 B CN 102087741B
- Authority
- CN
- China
- Prior art keywords
- clustering
- zone
- image processing
- pixel
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Abstract
The invention discloses a method and a system for processing an image by using a regional architecture. The method comprises the following steps of: classifying a plurality of neighboring pixels with at least one similar characteristic in an image into a cluster region with the same characteristic; giving a unique cluster label to each cluster region and describing an edge of each cluster region as a regional chain code, so that subsequent image processing programs can be only operated aiming at the neighboring pixels with the same cluster label around the processed pixel; and finally differentiating the cluster regions according to the cluster label and the regional chain code of each cluster region to synchronously operate image processing in the cluster regions.
Description
Technical field
The disclosure relates to a kind of image processing method and system that adopts regional architecture.
Background technology
Image processing techniques is now used many " image processing base units ", such as level and smooth (smooth) processing unit, removal noise (de-noise) processing unit, edge (edge) detecting unit, corner (corner) detecting unit, straight line (straight line) detecting unit, and curve (curve line) detecting unit, to optimize original image or to obtain effective characteristics of image.Yet the computing of these base units all is take " some framework " as the basis, and utilizes " neighborhood territory pixel (neighborhood pixels) " to finish the computing demand that various images are processed.When implementing specific base unit, each pixel all can produce corresponding result according to its neighborhood territory pixel, and the operation result of each pixel is compiled at every turn, namely is the result after processing.
Fig. 1 (a) and Fig. 1 (b) are the example of tradition " some framework level and smooth/Denoising disposal ".Wherein, Fig. 1 (a) illustrates when the image of original image 100 being implemented level and smooth/denoising is processed, and can adopt the mode of " from top to bottom " and " by left and right " at the shade 102 of the mobile 3*3 of image.And for the calculating of each pixel, then be to add again stack up after first the pixel value of the pixel of relative position in each numerical value on the shade 102 and the original image 100 being multiplied each other, this is convolution (convolution) computing of mathematics, and then with the summation of this convolution divided by numerical value on the shade 102, can obtain at last through smoothly/image 110 after the denoising computing.For example, pixel 104 on the original image 100 is being carried out smoothly/during Denoising disposal, namely be to add up after itself and eight adjacent pixels are multiplied by respectively numerical value on the shade 102, and then divided by the summation of numerical value on the shade 102, i.e. (1*65+1*66+1*65+1*67+1*90+1*68+1*67+1*66+1*65)/9, the last pixel value that obtains 69 are the result (as Fig. 1 (b) shown in) of pixel 104 behind level and smooth/Denoising disposal.Yet after implementing level and smooth computing, the feature at edge will be by obfuscation in the image.
Fig. 2 (a)～Fig. 2 (e) is the example of tradition " rim detection of some framework ".Wherein, Fig. 2 (a) is the original image 200 of 8*8, and Fig. 2 (b) and Fig. 2 (c) are respectively the shade 202 and 204 in order to detection level and vertical edge.Fig. 2 (d) and Fig. 2 (e) then are respectively original image 200 carrying out horizontal of Fig. 2 (a) and the result of vertical detection.Wherein, traditional edge detection method is when judging edge pixel (edge pixel), also be to utilize mobile and horizontal detection shade 202 and vertical detection shade 204 on original image 200, and obtain the convolution algorithm result with Fig. 2 (e) such as Fig. 2 (d), judge the edge according to the gradient (gradient) of the pixel in the image after the computing more at last.Although can successfully detect level and vertical marginal point in the image by above-mentioned edge detection method, why can't learn the corresponding zone of each marginal point.
Fig. 3 (a) and Fig. 3 (b) are the example of tradition " straight-line detection of some framework ".Wherein, Fig. 3 (a) is illustrated in the point (x, y) in the two-dimensional space, and by following Formula of Coordinate System Transformation, point (x, y) can be converted to corresponding w value:
w＝xcos(φ)+ysin(φ) (1)
Wherein, because x, y are datum, therefore for each different parameter φ (by 0 ° to 180 °), can calculate corresponding w value, and obtain the cumulant matrix shown in Fig. 3 (b).Each point in the xy space all can be converted to a curve in w φ space, and the curve number on the intersection point of maximum curve processes is arranged in this w φ space, namely represents the straight line number in the xy space.Although can learn straight line number in the image by above-mentioned line detection method, why can't further learn every corresponding zone of straight line.
Therefore, several major defects below the image processing techniques of tradition take " some framework " as the basis has.
Therefore first: can't learn the image property of pixel, when carrying out computing, must carry out computing in the hope of the result actual to all pixels.Yet, if can before computing, learn in advance the characteristic (for example certain pixel is to belong to noise spot or marginal point) of each pixel, just edge or corner detection limit only can implemented for these pixels, and then reduce the computing cost.
Therefore second: can't learn that pixel is adjacent the nature difference of pixel, when carrying out computing, must use all neighborhood territory pixels actual.Yet, be adjacent the nature difference (for example neighbor is noise) between pixel if can before computing, learn in advance each pixel, then when carrying out smoothing processing, rim detection or corner detection, can ignore neighbor of different nature, and then reduce wrong generation.
The the 3rd: need extra step to revise the characteristics of image because of basic computing was changed.For example, if the straight line in the detected image, traditional image processing techniques must use first " smoothing processing unit " and " removing the noise processed unit " with noise removal.Thus, just can cause the increase of edge pixel in the image.In other words, when implementing " edge detection unit ", can detect thicker line.Then need to implement extra " graph thinning (Thinning) processing unit " this moment, uses the backbone (Skeleton) who takes out thick line, could use afterwards " straight-line detection base unit " to come the straight line in the detected image.
The the 4th: can't effectively integrate, and the multiple image of parallel running is processed base unit.For example, if when image done three kinds of images such as " straight line ", " curve " and " corner detection " and processes base unit, the Pixel Information of the information that can share after " smoothing processing " and " rim detection ", can only carry out separately computing.In other words, can't be by parallel processing to reduce unnecessary computing.
The the 5th: through the image after the base unit computing, and can't provide extra information for referencial use to follow-up " image is processed advanced unit ".For example, " edge detection unit " only can detect the edge pixel in the image, but can't know which edge pixel is to belong to same zone (object), therefore also can't provide area information to process advanced unit to follow-up " image segmentation " or images such as " pattern recognitions ", to process more efficiently.
From the above, still there is shortcomings in the image processing techniques of tradition take " some framework " as the basis, and how can carry out synchronously various " image processing base units " to reduce unnecessary computing, and provide more information to use to follow-up " image is processed advanced unit ", become very important problem in the image processing field.
Summary of the invention
The disclosure provides a kind of image processing method that adopts regional architecture, a plurality of neighbors of similar characteristic is classified as the clustering zone with same characteristic features, and can process these a little clusterings zone synchronous operation images.
The disclosure provides a kind of image processing system that adopts regional architecture, utilizes a plurality of clusterings zone in the regional chain code Description Image, uses and can provide more information to process advanced unit to follow-up image.
The disclosure proposes a kind of image processing method that adopts regional architecture, it comprises that a plurality of neighbors that will have at least one similar characteristic in the image classify as the clustering zone with same characteristic features, and given each unique clustering mark in clustering zone, and be regional chain code with the edge-description in each clustering zone, then distinguish the clustering zone according to clustering mark and the regional chain code in each clustering zone, process in these clustering zones with at least a image of synchronous operation.
The disclosure proposes a kind of image processing system that adopts regional architecture, and it comprises bunching of picture element pretreatment unit, zone marker pretreatment unit, region description pretreatment unit and image processing base unit.Wherein, the bunching of picture element pretreatment unit classifies as the clustering zone with same characteristic features in order to a plurality of neighbors that will have at least one similar characteristic in the image.The zone marker pretreatment unit is in order to given each unique clustering mark in clustering zone.The region description pretreatment unit is regional chain code in order to the edge-description with each clustering zone.Image processing base unit then is to pick out each clustering zone according to the clustering mark in each clustering zone and regional chain code, processes with at least a image of synchronous operation on these clustering zones.
Based on above-mentioned, the image processing method of employing regional architecture of the present disclosure and system are that a plurality of neighbors that will have similar characteristic classify as the clustering zone with same characteristic features, and for each given unique clustering mark in clustering zone, and the regional chain code that can describe its edge, so when carrying out follow-up image processing, can be for each clustering zone synchronous operation handling procedure.
For above-mentioned feature and advantage of the present disclosure can be become apparent, exemplary embodiment cited below particularly, and cooperate accompanying drawing to be described in detail below.Present disclosure specification provides different exemplary embodiment that the technical characterictic of the different execution modes of the disclosure is described.Wherein, the usefulness that is configured to explanation of each element in the exemplary embodiment is not to limit the disclosure.
Description of drawings
Fig. 1 (a) and Fig. 1 (b) are the example of tradition " some framework level and smooth/Denoising disposal ".
Fig. 2 (a)～Fig. 2 (e) is the example of tradition " rim detection of some framework ".
Fig. 3 (a) and Fig. 3 (b) are the example of tradition " straight-line detection of some framework ".
Fig. 4 is the calcspar of the image processing system of the employing regional architecture that illustrates according to the disclosure one exemplary embodiment.
Fig. 5 then is the flow chart of the image processing method of the employing regional architecture that illustrates according to the disclosure one exemplary embodiment.
Fig. 6 (a) and Fig. 6 (b) are the examples of processing according to the image quantization that the disclosure one exemplary embodiment illustrates.
Fig. 7 (a) and Fig. 7 (b) are the examples of the given clustering mark that illustrates according to the disclosure one exemplary embodiment.
Fig. 8 (a), Fig. 8 (b), Fig. 8 (c) and Fig. 8 (d) are the chain code that illustrates according to the disclosure one exemplary embodiment and the example of provincial characteristics describing method.
Fig. 9 is the calcspar of processing base unit according to the image that the disclosure one exemplary embodiment illustrates.
Figure 10 A and Figure 10 B are the examples of the image processing method of the employing regional architecture that illustrates according to the disclosure one exemplary embodiment.
[main element symbol description]
100,200: original image
102,202,204: shade
104: pixel
110: the image after the computing
400: image processing system
410: the bunching of picture element pretreatment unit
420: the zone marker pretreatment unit
430: the region description pretreatment unit
440: image is processed base unit
442: the smoothing processing base unit
444: angle Calculating Foundation unit
446: the straight-line detection base unit
448: the curve detection base unit
600,1010: original image
610: quantized result
620: clustering mark result
1000: the clustering zone
1020: clustering mark result
A, B, C, D, E, F: clustering mark
S510～S540: each step of the image processing method of the employing regional architecture of the disclosure one exemplary embodiment
Embodiment
The disclosure proposes a kind of brand-new image and processes framework, it defines three image-region pretreatment units, with the combination of original image by " pixel ", be converted to various " zone combinations ", and then the image processing is promoted to " regional architecture " processing by traditional " some framework " processing.Therefore, the image processing techniques of new architecture can be in the situation that do not affect characteristics of image, and systematic integration is also carried out the computing that " smoothing processing ", " rim detection ", " corner detection ", " straight-line detection " reach various " image processing base units " such as " curve detection " synchronously.In addition, image processing techniques of the present disclosure also can provide more information (area data structure) to use to " image is processed advanced unit ".Adopt the image processing method of regional architecture and the detailed implementation content of system with the next act exemplary embodiment explanation disclosure.
Fig. 4 is the calcspar of the image processing system of the employing regional architecture that illustrates according to the disclosure one exemplary embodiment, and Fig. 5 then is the flow chart of the image processing method of the employing regional architecture that illustrates according to the disclosure one exemplary embodiment.Please be simultaneously with reference to Fig. 4 and Fig. 5, the framework of the image processing system 400 of this exemplary embodiment is image to be processed be divided into " compartmentalization pre-treatment " and " image processing " two stages, process base unit 440 comprising bunching of picture element pretreatment unit 410, zone marker pretreatment unit 420, region description pretreatment unit 430 and image, its function is described below:
In " compartmentalization pretreatment stage ", at first by the similarity criterion of bunching of picture element pretreatment unit 410 according to its definition, a plurality of neighbors that have at least one similar characteristic in the image are classified as the clustering zone (step S510) with same characteristic features.Wherein, bunching of picture element pretreatment unit 410 for example is image to be quantized (Quantization) process, with the pixel value with a plurality of pixels in the image be quantified as respectively a plurality of quantized values one of them, will have again afterwards identical quantized value and a plurality of pixels adjacent one another are and classify as the have same characteristic features clustering zone of (or color).
For instance, Fig. 6 (a) and Fig. 6 (b) are the examples of processing according to the image quantization that the disclosure one exemplary embodiment illustrates.Wherein, Fig. 6 (a) is the pixel value that illustrates each pixel in the original image 600, the pixel value scope of this exemplary embodiment hypothesis original image 600 is 1～100, and when carrying out quantification treatment, then be that this pixel value scope is divided into 4 sections (1～25,26～50,51～75,76～100) to give respectively quantized value.Wherein, when pixel value falls within 1～25, be about to this pixel value and be quantified as 1; When pixel value falls within 26～50, be about to this pixel value and be quantified as 2; When pixel value falls within 51～75, be about to this pixel value and be quantified as 3; When pixel value falls within 76～100, be about to this pixel value and be quantified as 4.And according to above-mentioned rule original image 600 is carried out can obtaining the quantized result 610 shown in Fig. 6 (b) after the quantification treatment.
Then, each given unique clustering mark (step S520) in clustering zone of being sorted out for bunching of picture element pretreatment unit 410 by zone marker pretreatment unit 420.Wherein, zone marker pretreatment unit 420 for example is to utilize connected component labeling method (connected componentslabeling) to come given clustering mark.
For instance, Fig. 7 (a) and Fig. 7 (b) are the examples of the given clustering mark that illustrates according to the disclosure one exemplary embodiment.Wherein, Fig. 7 (a) continues to use the quantized result 610 that Fig. 6 (b) illustrates, and this exemplary embodiment is namely for the regional given unique clustering mark of the middle quantized value of Fig. 6 (a) clustering identical and that communicate with each other.For example, be the given clustering mark A in 0 clustering zone for quantized value; Be the given clustering mark B in 1 clustering zone for quantized value; Be the given clustering mark D in 2 clustering zone for quantized value.It should be noted that, have 3 although quantized value is 3 clustering zone, these clustering zones do not link to each other each other, so this exemplary embodiment is about to it and is considered as independently zone, and distinguish given clustering mark C, E, F, then obtain at last the clustering mark result 620 shown in Fig. 7 (b).
Except given clustering mark, the image processing system 400 of this exemplary embodiment also comprises by region description pretreatment unit 430 describing mode by chain code (chain code), is regional chain code (step S530) with the edge-description in each clustering zone.In detail, region description pretreatment unit 430 can be for the different chain code of different directions definition, and when describing the edge, then for example be from the pixel in the upper left corner, clustering zone, sequentially find out the in twos relative direction between neighbor with counter clockwise direction or clockwise direction along the edge in clustering zone, and the corresponding chain code of these relative directions is recorded as regional chain code.
For instance, Fig. 8 (a), Fig. 8 (b), Fig. 8 (c) and Fig. 8 (d) are the chain code that illustrates according to the disclosure one exemplary embodiment and the example of provincial characteristics describing method.Wherein, Fig. 8 (a) continues to use the clustering mark result 620 that Fig. 7 (b) illustrates, and this exemplary embodiment utilizes the mode of chain code to describe its edge namely respectively for each clustering zone among Fig. 8 (a); Fig. 8 (b) illustrates the corresponding chain code of a plurality of different directions, that is right-hand, upper right side, top, upper left side, left, lower left, below and lower right correspond respectively to chain code 0～7; Fig. 8 (c) row illustrate the regional chain code that obtains after each clustering zone is described via chain code; Fig. 8 (d) then is listed as the area attribute that each clustering zone is shown, and comprises the size, mean flow rate, starting point, girth in this zone etc.Wherein, take clustering zone A as example, the chain code describing method of this exemplary embodiment is by pixel (0, the 0) beginning in its upper left corner, sequentially find out the in twos relative direction between neighbor along its edge with counter clockwise direction, and record chain code corresponding to this relative direction.Wherein, pixel (0,1) is the below with respect to the direction of pixel (0,0), therefore can record the corresponding chain code 6 in below.In like manner, ensuing pixel (0,2) ..., (0,7) all be the below with respect to the direction of front pixel, therefore can record 6 chain codes 6.Afterwards, the edge of clustering zone A then turns to right-hand, and pixel (1,7) is right-hand with respect to the direction of pixel (0,7), therefore can record right-hand corresponding chain code 0.By that analogy, treat that pixel (0,0) is got back in the description of chain code, can obtain the regional chain code of clustering zone A.In like manner, the regional chain code of its correspondence also can be obtained through method thus in other clustering zones.
Behind the mark of finishing each clustering zone and edge-description, namely finish " compartmentalization pretreatment stage ", and enter " processing stage of image ".At this moment, image is processed base unit 440 and can be utilized the clustering mark in each clustering zone and regional chain code to distinguish these clustering zones according to the needs of its running, uses synchronous operation image processing (step S540) on these clustering zones.Wherein, described image is processed base image such as comprising smoothing processing, corner detection, straight-line detection, curve detection and is processed, image processing base unit 440 then can configure respectively smoothing processing base unit 442, angle Calculating Foundation unit 444, straight-line detection base unit 446 and curve detection base unit 448 (as shown in Figure 9) and carry out these base image processing, with the next detailed practice of introducing respectively these base image processing.
About smoothing processing: smoothing processing base unit 442 can be known by the regional chain code in clustering zone the marginal information in clustering zone, therefore actual when carrying out smoothing processing, the pixel that also only belongs to same clustering zone on every side for the processing pixel is carried out computing.In other words, smoothing processing base unit 442 employed homogenizing shades only can act in the clustering zone under the pixel of processing, and the extra-regional pixel of clustering can not included in the level and smooth computing in the lump.Accordingly, the smoothing processing of this exemplary embodiment can avoid other area pixel to add the result who takes advantage of the edge blurry that causes.
Detect about the corner: any three neighbor (x on its edge can be found out by the regional chain code in each clustering zone in angle Calculating Foundation unit 444
0, y
0), (x
1, y
1), (x
2, y
2) and relative direction therebetween, and utilize following cosine formula to ask for this three formed angle theta of neighbor:
Wherein, according to angle theta, then can judge pixel (x
0, y
0) the corner characteristic properties.For example θ=90 represent that then this point is the right angle corner pixels.
About straight-line detection: straight-line detection base unit 446 can be by analyzing the angle of each pixel of obtaining above-mentioned angle Calculating Foundation unit 444, and judge whether there is straight line between these pixels.Wherein, find to have the angle of a plurality of neighbors when straight-line detection base unit 446 and be 180 when spending, can judge and have straight line between the neighbor, and its length equals the number that these angles are the neighbors of 180 degree.
For instance, following table 1 row illustrate angle value corresponding to a plurality of pixels among the C of clustering zone.Wherein, be 180 degree as can be known by pixel C (2) to the angle value of C (7): pixel C (2) is straight line to C (7), and the length of this straight line is 6.
| Pixel | C(1) | C(2) | C(3) | C(4) | C(5) | C(6) | C(7) | C(8) | … |
| |
90 | 180 | 180 | 180 | 180 | 180 | 180 | 90 | … |
Table 1
About curve detection: curve detection base unit 448 can be by analyzing the angle of each pixel of obtaining above-mentioned angle Calculating Foundation unit 444, and judge whether there is curve between these pixels.Wherein, find to have the angle of a plurality of neighbors identical and when being not equal to the angle of 180 degree when curve detection base unit 448, can judge and have curve between the neighbor, and its length equals the number of these neighbors.
For instance, following table 2 row illustrate angle value corresponding to a plurality of pixels among the D of clustering zone.Wherein, be 145 degree as can be known by pixel D (2) to the angle value of D (7): pixel D (2) be curve to D (7), and the angle of this curve is 145 to spend, and length is 6.
| Pixel | C(1) | C(2) | C(3) | C(4) | C(5) | C(6) | C(7) | C(8) | … |
| |
90 | 145 | 145 | 145 | 145 | 145 | 145 | 90 | … |
Table 2
From the above, use the clustering mark of " compartmentalization pretreatment stage " gained and the information of regional chain code, image processing system of the present disclosure can be on these clustering zones in " processing stage of image " the synchronous operation image processing program.
It is worth mentioning that, before carrying out the image processing, image processing system 400 also can utilize the next area attributes (shown in Fig. 8 (d)) such as size, girth or mean flow rate according to each clustering zone of sequencing unit (not illustrating), is sorted in these clustering zones.And when carrying out image processing program, then can consider the computing load of entire system actual, process in advance in the preceding part clustering zone of only selecting to sort.
The feature of processing in order to be illustrated more clearly in the multiple image of above-mentioned disclosure synchronous operation illustrates that with the next exemplary embodiment of lifting again the disclosure carries out the program of the synchronous computing that multiple image processes by pixel ground for the clustering edges of regions.
Figure 10 A and Figure 10 B are the examples of the image processing method of the employing regional architecture that illustrates according to the disclosure one exemplary embodiment.Please be simultaneously with reference to Figure 10 A and Figure 10 B, clustering is labeled as clustering zone original image 1010 and the clustering mark result 1020 on every side of D among the clustering mark result that Figure 10 A acquisition Fig. 7 of system (b) of this exemplary embodiment illustrates, and each pixel of clustering zone 1000 (thick black line frame plays the zone) is carried out synchronously the image processing computing of smoothing processing, rim detection, angle calculation and straight line and curve detection.Each representative hereinafter according to regional chain code and the edges of regions pixel of sequentially processing in stage.In addition, step below uses the four neighbor definition that are communicated with (4-connectivity) to be used as the computing basis that image is processed for the purpose of simplifying the description.
In (1) stage, image processing system is processed for the pixel in the 1000 upper left corners, clustering zone.Wherein, when carrying out smoothing processing, only take belong to the neighbor (being right-hand and lower pixel) of the same area and itself carry out convolution algorithm around the pixel of processing, and be averaged to obtain the result of this pixel, that is (65*1+66*1+67*1)/3=66; When carrying out rim detection, then can know that by the regional chain code in clustering zone 1000 this pixel is edge pixel, therefore be labeled as E; When carrying out angle calculation, then can calculate the pixel of processing and be adjacent the formed angle of pixel (being right-hand and lower pixel), and obtain 90 operation results of spending; When carrying out straight line and curve detection, then owing to only processing a pixel at present, the method for there is no is confirmed whether whether it belongs to the part of straight line or curve.
In (2) stage, image processing system is processed for clustering zone 1000 second left pixels.Wherein, when carrying out smoothing processing, only take the pixel of processing and neighbor thereof (i.e. top and lower pixel) carry out convolution algorithm, and be averaged to obtain the result of this pixel, that is (65*1+67*1+67*1)/3=66; When carrying out rim detection, then can know that by the regional chain code in clustering zone 1000 this pixel is edge pixel, therefore be labeled as E; When carrying out angle calculation, then can calculate the pixel of processing and be adjacent the formed angle of pixel (i.e. top and lower pixel), and obtain the operation result of 180 degree; When carrying out straight line and curve detection, then owing to only obtaining the pixel of one 90 degree and one 180 degree at present, the method for there is no is confirmed whether whether it belongs to the part of straight line or curve.
In (3) stage, image processing system is processed for clustering zone 1000 third left pixels.Wherein, when carrying out smoothing processing, only take the pixel of processing and neighbor thereof (i.e. top, right-hand and lower pixel) carry out convolution algorithm, and be averaged to obtain the result of this pixel, that is (67*1+66*1+67*1+67*1)/4=66; When carrying out rim detection, then can know that by the regional chain code in clustering zone 1000 this pixel is edge pixel, therefore be labeled as E; When carrying out angle calculation, then can calculate the pixel of processing and be adjacent the formed angle of pixel (i.e. top and lower pixel), and obtain the operation result of 180 degree; When carrying out straight line and curve detection, then owing to having obtained the pixel of two 180 degree at present, therefore can confirm to be straight line between these two pixels, and the length of this straight line is 2.
In (4) stage, image processing system is processed for clustering zone 1000 fourth left pixels.Wherein, when carrying out smoothing processing, only take the pixel of processing and neighbor thereof (i.e. top and right-hand pixel) carry out convolution algorithm, and be averaged to obtain the result of this pixel, that is (67*1+66*1+67*1)/3=67; When carrying out rim detection, then can know that by the regional chain code in clustering zone 1000 this pixel is edge pixel, therefore be labeled as E; When carrying out angle calculation, then can calculate the pixel of processing and be adjacent the formed angle of pixel (i.e. top and lower right pixel), and obtain the operation result of 135 degree; When carrying out straight line and curve detection, then owing to 135 of at present acquisition is not to belong to straight line, therefore still keep the straight line one of previous judgement.
In (5) stage, image processing system is processed for lower pixel in the middle of the clustering zone 1000.Wherein, when carrying out smoothing processing, only take the pixel of processing and neighbor thereof (i.e. top pixel) carry out convolution algorithm, and be averaged to obtain the result of this pixel, that is (66*1+69*1)/2=67; When carrying out rim detection, then can know that by the regional chain code in clustering zone 1000 this pixel is edge pixel, therefore be labeled as E; When carrying out angle calculation, then can calculate the pixel of processing and be adjacent the formed angle of pixel (being upper left side and top pixel), and obtain 45 operation results of spending; When carrying out straight line and curve detection, then owing to 45 of at present acquisition is not to belong to straight line, therefore still keep the straight line one of previous judgement.
According to above-mentioned way, get final product class and release the operation result in (11) stage in (6) stage to the, and by this exemplary embodiment as can be known, image processing system is by clustering mark and the regional chain code information of " compartmentalization pretreatment stage " gained, can be in the situation that do not affect the original image feature, to the multiple image processing program of each clustering zone synchronous operation in the image.
In sum, the image processing method of employing regional architecture of the present disclosure and device are with the combination of original image by " pixel ", be converted to the combination in " zone ", its advantage not only can be so that follow-up " image processing base unit " does computing take " same area in the neighbor " as the basis, and can carry out synchronously various " image processing base units " in the situation that do not affect the original image feature.Even in some applications, can also skip the computing of base unit, and utilize the area data (provincial characteristics) after " compartmentalization pre-treatment " computing directly to carry out " image is processed advanced unit ".
Although the disclosure with exemplary embodiment openly as above; so it is not to limit the disclosure, and those skilled in the art are not within breaking away from spirit and scope of the present disclosure; when doing a little change and retouching, therefore protection range of the present disclosure is as the criterion when looking the appended claims person of defining.
Claims (23)
1. image processing method that adopts regional architecture comprises:
The a plurality of neighbors that have at least one similar characteristic in one image are classified as the clustering zone with same characteristic features, comprising:
This image is carried out a quantification treatment, be quantified as a plurality of quantized values with the pixel value with a plurality of pixels in this image; And
A plurality of neighbors that will have identical quantized value classify as the clustering zone with same characteristic features;
Given each these unique clustering mark in clustering zone;
An edge of describing each these clustering zone is a regional chain code, comprising:
Definition is corresponding to a plurality of chain codes of different directions; And
With one counterclockwise or a clockwise direction sequentially find out a relative direction between two adjacent pixels along this edge in each these clustering zone, and record chain code corresponding to these relative directions to generate this zone chain code; And
This clustering mark and this zone chain code according to each these clustering zone are distinguished these clustering zones, process in these clustering zones with at least one image of synchronous operation.
2. the image processing method of employing regional architecture as claimed in claim 1, wherein the step of unique this clustering mark in given each these clusterings zone comprises the clustering mark that utilizes given each these clustering zone of a connected component labeling method.
3. the image processing method of employing regional architecture as claimed in claim 1, wherein sequentially find out this relative direction between two adjacent pixels along this edge in each these clustering zone with this counter clockwise direction or this clockwise direction, and record chain code corresponding to these relative directions and comprise that with the step that generates this zone chain code the top left corner pixel by each these clustering zone begins.
4. the image processing method of employing regional architecture as claimed in claim 1, wherein these directions comprise right-hand, upper right side, top, upper left side, left, lower left, below and lower right.
5. the image processing method of employing regional architecture as claimed in claim 1, wherein distinguish these clustering zones at this clustering mark and this zone chain code according to each these clustering zone, processed before the step in these clustering zones with this at least one image of synchronous operation, also comprise:
According to the area attribute in each these clustering zone these clusterings zones of sorting; And
According to the ordering in these clustering zones, select these clustering zones of part to process to carry out this follow-up at least one image.
6. the image processing method of employing regional architecture as claimed in claim 5, wherein this area attribute comprises size, girth or the mean flow rate in this clustering zone.
7. the image processing method of employing regional architecture as claimed in claim 1, wherein distinguish these clusterings zones according to this clustering mark in each these clustering zone and this zone chain code, process in the step in these clustering zones with this at least one image of synchronous operation and comprise:
Each pixel in this clustering zone is respectively carried out a level and smooth computing, wherein only carry out computing for the pixel that belongs to same clustering zone around this pixel.
8. the image processing method of employing regional architecture as claimed in claim 1, wherein distinguish these clusterings zones according to this clustering mark in each these clustering zone and this zone chain code, process in the step in these clustering zones with this at least one image of synchronous operation and comprise:
Find out three neighbors at this edge that is positioned at this clustering zone and relative direction therebetween according to this zone chain code in each these clustering zone, and in order to ask for the formed angle of these three neighbors.
9. the image processing method of employing regional architecture as claimed in claim 8, the step of wherein asking for these three formed these angles of neighbor comprises utilizes a cosine formula.
10. the image processing method of employing regional architecture as claimed in claim 8, wherein in three neighbors finding out this edge that is positioned at this clustering zone according to this zone chain code in each these clustering zone and relative direction therebetween, and in order to after the step of asking for these three formed these angles of neighbor, also comprise:
Analyze this angle of each these pixel in each these clustering zone, detect in line and a curve detection with complete.
11. the image processing method of employing regional architecture as claimed in claim 10 is wherein analyzed this angles of each these pixel in each these clustering zone, comprises with the step of finishing this straight-line detection:
When this angle that a plurality of neighbors are arranged is 180 when spending, judge to have the straight line that a length equals the number of these neighbors between these neighbors.
12. the image processing method of employing regional architecture as claimed in claim 10 is wherein analyzed this angles of each these pixel in each these clustering zone, comprises with the step of finishing this curve detection:
When this angle that a plurality of neighbors are arranged identical and for be not equal to 180 the degree an angle time, judge to have the curve that a length equals the number of these neighbors between these neighbors.
13. an image processing system that adopts regional architecture comprises:
One bunching of picture element pretreatment unit, the a plurality of neighbors that have at least one similar characteristic in one image are classified as the clustering zone with same characteristic features, wherein this bunching of picture element pretreatment unit carries out a quantification treatment to this image, be quantified as a plurality of quantized values with the pixel value with a plurality of pixels in this image, and a plurality of neighbors that will have an identical quantized value classify as the clustering zone with same characteristic features;
One zone marker pretreatment unit, given each these unique clustering mark in clustering zone;
One region description pretreatment unit, an edge of describing each these clustering zone is a regional chain code, wherein this region description pretreatment unit definition is corresponding to a plurality of chain codes of different directions, and with one counterclockwise or a clockwise direction sequentially find out a relative direction between two adjacent pixels along this edge in each these clustering zone, and record chain code corresponding to these relative directions to generate this zone chain code; And
One image is processed base unit, distinguishes these clustering zones according to this clustering mark and this zone chain code in each these clustering zone, processes in these clustering zones with at least one image of synchronous operation.
14. the image processing system of employing regional architecture as claimed in claim 13, wherein this zone marker pretreatment unit comprises and utilizes given each these these the unique clustering marks of clusterings zone of a connected component labeling method.
15. the image processing system of employing regional architecture as claimed in claim 13, wherein this region description pretreatment unit comprises that top left corner pixel by each these clustering zone begins sequentially to find out this relative direction between two adjacent pixels along this edge in each these clustering zone with this counter clockwise direction or this clockwise direction, and records chain code corresponding to these relative directions to generate this zone chain code.
16. the image processing system of employing regional architecture as claimed in claim 13, wherein these directions comprise right-hand, upper right side, top, upper left side, left, lower left, below and lower right.
17. the image processing system of employing regional architecture as claimed in claim 13 also comprises:
One sequencing unit according to the area attribute in each these clustering zone these clusterings zones of sorting, and according to the ordering in these clustering zones, is selected these clustering zones of part to offer this image and is processed base unit and process to carry out this follow-up image.
18. the image processing system of employing regional architecture as claimed in claim 17, wherein this area attribute comprises size, girth or the mean flow rate in this clustering zone.
19. the image processing system of employing regional architecture as claimed in claim 13, wherein this image processing base unit comprises:
One smoothing processing base unit carries out a level and smooth computing to each pixel in this clustering zone respectively, wherein only carries out computing for the pixel that belongs to same clustering zone around this pixel.
20. the image processing system of employing regional architecture as claimed in claim 13, wherein this image processing base unit comprises:
Three neighbors at this edge that is positioned at this clustering zone and relative direction are therebetween found out according to this zone chain code in each these clustering zone in one angle Calculating Foundation unit, and in order to ask for the formed angle of these three neighbors.
21. the image processing system of employing regional architecture as claimed in claim 20, wherein this angle Calculating Foundation unit comprises and utilizes a cosine formula to ask for formed this angle of these three neighbors.
22. the image processing system of employing regional architecture as claimed in claim 20, wherein this image processing base unit also comprises:
One straight line detects base unit, analyzes this angle of each these pixel in each these clustering zone, and is 180 when spending when this angle that a plurality of neighbors are arranged, and judges to have the straight line that a length equals the number of these neighbors between these neighbors.
23. the image processing system of employing regional architecture as claimed in claim 20, wherein this image processing base unit also comprises:
One curve detection base unit, analyze this angle of each these pixel in each these clustering zone, and identical and for being not equal to one of 180 degree during angle when this angle that a plurality of neighbors are arranged, judge to have the curve that a length equals the number of these neighbors between these neighbors.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200910252678 CN102087741B (en) | 2009-12-03 | 2009-12-03 | Method and system for processing image by using regional architecture |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200910252678 CN102087741B (en) | 2009-12-03 | 2009-12-03 | Method and system for processing image by using regional architecture |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102087741A CN102087741A (en) | 2011-06-08 |
| CN102087741B true CN102087741B (en) | 2013-01-02 |
Family
ID=44099530
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200910252678 Active CN102087741B (en) | 2009-12-03 | 2009-12-03 | Method and system for processing image by using regional architecture |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102087741B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2019021991A (en) * | 2017-07-12 | 2019-02-07 | オリンパス株式会社 | Imaging element, imaging apparatus, imaging program and imaging method |
| CN110825080B (en) * | 2019-10-31 | 2020-11-13 | 华南农业大学 | Orchard path visual navigation method, system and medium based on fuzzy control algorithm |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1340273A (en) * | 1999-04-23 | 2002-03-13 | 三星电子株式会社 | Color image segmentation method |
| CN1568479A (en) * | 2001-10-11 | 2005-01-19 | 皇家飞利浦电子股份有限公司 | Method and apparatus for discriminating between different regions of an image |
| CN101520894A (en) * | 2009-02-18 | 2009-09-02 | 上海大学 | Method for extracting significant object based on region significance |
-
2009
- 2009-12-03 CN CN 200910252678 patent/CN102087741B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1340273A (en) * | 1999-04-23 | 2002-03-13 | 三星电子株式会社 | Color image segmentation method |
| CN1568479A (en) * | 2001-10-11 | 2005-01-19 | 皇家飞利浦电子股份有限公司 | Method and apparatus for discriminating between different regions of an image |
| CN101520894A (en) * | 2009-02-18 | 2009-09-02 | 上海大学 | Method for extracting significant object based on region significance |
Non-Patent Citations (1)
| Title |
|---|
| JP特开2003-173442A 2003.06.20 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102087741A (en) | 2011-06-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109840521B (en) | Integrated license plate recognition method based on deep learning | |
| Liang et al. | Delta-density based clustering with a divide-and-conquer strategy: 3DC clustering | |
| CN102609686B (en) | Pedestrian detection method | |
| KR101403876B1 (en) | Method and Apparatus for Vehicle License Plate Recognition | |
| CN102292745B (en) | Image signature extraction device | |
| CN109002824B (en) | OpenCV-based building drawing label information detection method | |
| CN103955660B (en) | Method for recognizing batch two-dimension code images | |
| US9224207B2 (en) | Segmentation co-clustering | |
| CN104463795A (en) | Processing method and device for dot matrix type data matrix (DM) two-dimension code images | |
| CN101122953A (en) | Picture words segmentation method | |
| CN107506765B (en) | License plate inclination correction method based on neural network | |
| CN105453153A (en) | Traffic light detection | |
| CN103473545B (en) | A kind of text image method for measuring similarity based on multiple features | |
| CN103593695A (en) | Method for positioning DPM two-dimension code area | |
| CN104239909A (en) | Method and device for recognizing images | |
| CN103106265A (en) | Method and system of classifying similar images | |
| CN109858438B (en) | Lane line detection method based on model fitting | |
| CN112200186B (en) | Vehicle logo identification method based on improved YOLO_V3 model | |
| CN102194102A (en) | Method and device for classifying a traffic sign | |
| CN113343989B (en) | Target detection method and system based on self-adaption of foreground selection domain | |
| CN104408728A (en) | Method for detecting forged images based on noise estimation | |
| CN108229247A (en) | A kind of mobile vehicle detection method | |
| CN113435240A (en) | End-to-end table detection and structure identification method and system | |
| CN104966047A (en) | Method and device for identifying vehicle license | |
| KR101574874B1 (en) | Method and apparatus for character segmentation of license plate |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |