CN1848137A - Method and equipment for intensifying character line image and storage medium - Google Patents

Abstract

The present invention relates to a method for intensifying row-of-character image, its equipment and storage medium. In order to raise OCR identification rate the row-of-character image can be divided into three kinds of types, including clear type whose image quality is best, stroke-broken type, which has not noise and whose stroke is broken and other type, which has not broken stroke and has noise more than that of clear type. According to the type of image, said invention can adopt different measures including removing noise, making cross expansion and smoothing image, etc. to intensify image.

Description

Strengthen method, equipment and the storage medium of character row image

Technical field

The application relates generally to for the file and picture of OCR (optical character identification) purpose and handles, and relates in particular to the method, equipment and the storage medium that are used to strengthen the character row image.

Background technology

In the OCR of any language, the row cut apart with Character segmentation be in character classification (identification) basic step before.

In different language, pictographic character is such as Chinese character and japanese character, and other similar character is such as the assumed name (being that example is illustrated the present invention with Chinese character and Chinese document after this) of Korean characters and Japanese, is very different with the expression way of western language.They have following feature:

1. Chinese character is a character independently;

2. they are square basically.

As the essential step in the general OCR system, " cutting apart " step is more crucial for Chinese file and picture, and this is because itself and the correlativity of discerning processing.Investigation result shows, in the OCR system of the prior art of Chinese file and picture, 80% OCR mistake is owing to cause incorrect cutting apart.

When picture quality was high, for the image segmentation of Chinese document, the pixel column (row) of search blank was just enough as cut-point.But when having many noises or disconnected in the image (for example because the quality of original paper is low, perhaps because the quality degradation that scanning causes), row is cut apart with Character segmentation and will be very easy to make mistakes.Even in the single character picture that segmentation obtains, the OCR discrimination also can sharply descend because of the noise that centers on these characters.In actual applications, need widely low-quality image is carried out OCR.

In the research and development of Chinese OCR, how correctly being partitioned into character (edge noise is not comprised recently, effective stroke is not removed) is a very important problem.

Propose many algorithms and cut apart Chinese file and picture.For example:

1. apply to row and cut apart direct level and vertical projection with Character segmentation.

In the partitioning algorithm of prior art, image block flatly is divided into plurality of sections arranged side by side.Blank pixel in blank pixel row in each section and each image line is listed as cut-point.Then, by some simple methods,, adjust too little or too big character such as distance calculation, further projection etc.

In document 1, obtain the horizontal histogram of image block and the vertical histogram of image line.Use the square features of these histograms and Chinese character to cut apart.

2. the row image is being carried out cutting apart after the pre-service.

In document 2, described a kind of in OCR engine especially at shadow tone (Chinese) file and picture, classification and recover the algorithm of row image.Half tone document image is compared with general file and picture has some distinguishing characteristicss, such as ground unrest little, rule, and in whole character equally distributed disconnected pen.This method uses median filter to come at first ground unrest noise reduction to shadow tone piece image.Then, cut apart capable image according to the blank pixel row.Not through on the former image that begins of medium filtering, calculate black picture element density, then mark black eight connected domains, black four connected domains, white eight connected domains and white four connected domains.Use these information, need to judge whether noise reduction, and use which kind of method noise reduction, comprise and remove single black picture element, and smoothly fall burr.Also judge three kinds of attributes (common, hollow and disconnected pen), and shadow tone row image is recovered accordingly based on half tone image.

In document 3, provided a kind of useful method, be used for assessing the quality of the file and picture of typescript, and automatically select optimum restoration methods according to black and white eight connected domains of mark, such as the burn into form near and kFill wave filter (seeing document 4) etc.

3. the capable dividing method that is used for low-quality image.

In document 5, provided a kind of being used for image block is divided into capable effective ways.This method is divided into plurality of sections level, arranged side by side, that have preset width with image block, and obtains the pixel distribution statistics of entire image piece and each section thereof.According to described pixel distribution statistics, the row in the image block can be split well.

Document 1: " Chinese Optical Character Recognition for informationextraction from Video Images ", Wing Hang Cheung etc., www.cse.cuhk.edu.hk/～king/PUB/CISST00.pdf visited on June 18th, 2004;

Document 2: " being used to strengthen method, device and the storage medium of file and picture and character recognition ", Hu Ou, Lee offers, Chinese patent application No.200310118684.7;

Document 3: " Quality Assessment and Restoration of TypewrittenDocument Images ", Michael Cannon etc., http://www.c3.lanl.gov/～tmc/quality/quarc.pdf visited on June 18th, 2004;

Document 4: " Lawrence O ' Gorman and Rangachar Kasturi, Document Image Analysis ", IEEE Computer Society Press, 1998, p13;

Document 5: character and picture branch's method and apparatus and character and picture recognition methods and device, Luo Zhaohai, Li Yi, Chinese patent application No.01140938.x;

The algorithm of above-mentioned prior art has many shortcomings:

1. be used for row and cut apart direct level and vertical projection with Character segmentation.

For high-quality Chinese image, according to direct level and vertical projection, the segmentation result that can obtain.But if many noises are arranged in file and picture, then row and character can be in contact with one another, and just can not find desired blank pixel row or row, and the partitioning algorithm of prior art will lose efficacy.

If noise is many, peak in the histogram (level or vertical projection distribution plan) and paddy just no longer have periodically, the segmentation result that the algorithm in the document 1 just can not obtain.

On the other hand, in the algorithm of prior art such as document 1, a disconnected character of Chinese is split into many easily.Even according to the square features of Chinese character, these fragments can not correctly be merged.

2. to row image cutting apart after carrying out pre-service.

In document 2, its algorithm at be half tone image, be used for common low-quality image and can not promote.

Median filter wherein only is suitable for the image block of low noise background.For many low-quality images, after the process medium filtering, row still is in contact with one another.Simultaneously, noise reduction can damage some seriously images (very thin stroke) of disconnected pen.

Because between different Chinese character (frequent nearly altogether 3000 to 5000 characters that use), the quantity of black and white connected domain and their ratio change very greatly, it is not enough to assess the quality of common Chinese file and picture.Simultaneously, between different Chinese characters, they are different to the susceptibility of noise and disconnected pen.For example, in Chinese file and picture, the connected domain information of noisy capable image may be similar to the information of capable image of high-quality and disconnected pen row image.

For common low-quality image (being different from half tone image), the algorithm in the document 2 may be judged as some noisy capable image " hollow " or " disconnected pen ", and carries out expansive working, thereby and some disconnected pen may be judged as noise and is eliminated.Test shows that this classification and recovery algorithms are not suitable for common low-quality image.

In document 3, image classification algorithms wherein is to design at western language (and the frequent special character that uses of minority).As previously mentioned, between different Chinese characters, connected domain information and be very different to the sensitivity of noise or disconnected pen.Therefore, in inferior quality Chinese file and picture, the distribution of connected domain is more complicated, is necessary to seek a kind of new sorting algorithm and relevant recovery algorithms.

3. the capable partitioning algorithm that is used for low-quality image.

Even the capable partitioning algorithm in the document 5 also can obtain extraordinary result for the image with many noises or disconnected pen, nearly all provisional capital can be split.But, need a kind of good Character segmentation algorithm, because may comprise many noises in the image of being expert at low-quality image.

Summary of the invention

Therefore, the purpose of this invention is to provide a kind of method and apparatus that is used to strengthen the character row image, cut apart and Character segmentation, thereby improve discrimination, especially for low-quality file and picture so that can more accurately go.The present invention also aims to provide a kind of storage medium that is used to realize described method.

In order to achieve the above object, a kind of method that strengthens the character row image is provided, comprise the steps: original image is categorized as a kind of at least three types, but but these three types comprise best the knowing type, almost do not have the disconnected pen type of noise stroke for disconnected pen of picture quality, and have than knowing noise that type Duo other type of almost disconnected; And, strengthen described original image according to its type, comprising: if original image is to know type then delete independent noise; Perhaps, if original image be disconnected pen type then original image is carried out cross expand; Perhaps, if original image be other type then delete independent noise, then resulting image is carried out smoothly.

In order to achieve the above object, a kind of equipment that strengthens the character row image also is provided, comprise: sorter, be used for original image is categorized as at least three types a kind of, but but these three types comprise best the knowing type, almost do not have the disconnected pen type of noise stroke for disconnected pen of picture quality, and have than knowing noise that type Duo other type of almost disconnected; And booster, be used for strengthening described original image according to the type of original image, comprising: the independent noise denoiser, being used at original image is deletion independent noise when knowing type; First expander, being used at original image is that disconnected pen type is then carried out the cross expansion to original image; And smoother, be used for when original image is other type, carrying out smooth operation for image from the output of independent noise denoiser.

The present invention also provides a kind of storage medium, it is characterized in that, has stored the program code that is used to realize said method therein.

Description of drawings

Other purpose of the present invention, feature and advantage will become more clear after the detailed description of preferred embodiments reading hereinafter.The accompanying drawing part of book as an illustration is used for the diagram embodiments of the invention, and is used from explanation principle of the present invention with instructions one.In the accompanying drawings:

Fig. 1 is the block diagram that can be used for realizing an example of computer system of the present invention;

Fig. 2 is the block diagram according to a preferred embodiment of equipment of the present invention;

Fig. 3 is the process flow diagram that is used to illustrate first embodiment of method of the present invention;

Fig. 4 is the process flow diagram of step that is used for illustrating the second embodiment classified image of method of the present invention;

Fig. 5 is the process flow diagram of step that is used for illustrating the 3rd embodiment classified image of method of the present invention;

Fig. 6 is the process flow diagram that is used to illustrate described the 3rd embodiment of method of the present invention;

Fig. 7 is the process flow diagram of step that is used for illustrating the 4th embodiment classified image of method of the present invention;

Fig. 8 is the process flow diagram that is used to illustrate a kind of application of the present invention;

Fig. 9 is the histogram that is used to illustrate the effect when the present invention is used for OCR;

Figure 10 is the synoptic diagram that is used to illustrate improved median filter used in this invention;

Figure 11 is an example knowing the image of type;

Figure 12 is an example of the image of disconnected pen type;

Figure 13 is an example of the image of other type;

Figure 14 is the process flow diagram that is used to illustrate as the step of Fig. 3 and deletion independent noise shown in Figure 6.

Embodiment

Computer system for example

Method and apparatus of the present invention can be realized in any messaging device.Described messaging device for example is the single-chip microcomputer (SCM) of personal computer (PC), notebook computer, embedding scanner, duplicating machine, facsimile recorder etc., or the like.For those of ordinary skills, be easy to realize method and apparatus of the present invention by software, hardware and/or firmware.Especially it should be noted that, it is evident that for those of ordinary skills, for any step of carrying out this method or the combination of step, the perhaps any parts of equipment of the present invention or the combination of parts may need to use input-output device, memory device and microprocessor such as CPU etc.May not be certain to mention these equipment in the explanation to method and apparatus of the present invention below, but in fact used these equipment.

As above-mentioned messaging device, the block diagram of Fig. 1 shows giving an example of a computer system, can realize method and apparatus of the present invention therein.It should be noted that the computer system that is shown in Fig. 1 just is used for explanation, does not really want to limit the scope of the invention.

From the angle of hardware, computing machine 1 comprises a CPU 6,5, RAM of hard disk (HD) 7, a ROM 8 and an input-output device 12.Input-output device can comprise input media such as keyboard, Trackpad, tracking ball and mouse etc., and output unit is such as printer and monitor, and input-output unit is such as floppy disk, CD drive and communication port.

From the angle of software, described computing machine mainly comprises operating system (OS) 9, input/output driver 11 and various application program 10.As operating system, can use any operating system that to buy on the market, such as Windows series and based on the operating system of Linux.Input/output driver is respectively applied for and drives described input-output device.Described application program can be an Any Application, such as text processor, image processing program etc., comprising can be used in this invention and aim at the present invention's application program establishment, that can call described existing program.

Like this, in the present invention, can in the hardware of described computing machine, realize method and apparatus of the present invention by operating system, application program and input/output driver.

In addition, computing machine 1 can be connected to digital device 3 and application apparatus 2.Digital device can be used as image source, can be camera, video camera, scanner or the digitizer that is used for analog image is converted to digital picture.The result that equipment of the present invention and method obtain is output to application apparatus 2, and the latter carries out suitable operation according to described result.This application apparatus also can be implemented as the Another application (combining with hardware) that realizes in computing machine 1, be used for further handling described image.

Be used to strengthen the method and apparatus of character row image

For convenience, the file and picture of setting type with level is an example.Obviously, vertically be similarly cutting apart of the image of setting type, and only direction need be revolved to turn 90 degrees, just from horizontally rotating to vertically or from vertically rotating to be level.

Hereinafter, division symbol "/" is meant division of integer, and its result of division also is an integer, and fraction part is omitted.Term " OK " is meant horizontal line, and just, its length direction is the direction of row image.

The present invention be to be expert at cut apart after correctly to the capable image classification of Chinese, automatically to carry out the recovery operation of corresponding row, the capable image of Zeng Qianging can be split into character then.As an example, the capable partitioning algorithm in the document 5 can be used for image block is divided into row before using the present invention.

Fig. 2 illustrates a preferred embodiment of equipment of the present invention.This equipment comprises a sorter 202, is used for original image is categorized as different types, and type information is offered booster 204, and next booster 204 strengthens raw information according to type information, and the image after the output enhancing.

Further describe concrete parts shown in Figure 2 below in conjunction with method of the present invention.When the explanation below reading, those of ordinary skill in the art will appreciate that parts shown in Figure 2 are not all to be absolutely necessary, and wherein some parts can omit in certain embodiments.

(first embodiment)

Fig. 3 illustrates the process flow diagram of first embodiment of method of the present invention.It is the step (S302) of knowing type, disconnected pen type and other type with image classification that this method starts from one.Know type, such as person shown in Figure 11, mean that capable image comprises noise hardly, that is to say, the row image is a best in quality.Disconnected pen type such as person shown in Figure 12, means that the quality of capable image is low, but this is not because noise causes, but because information dropout causes, just, some stroke of character has ruptured.Correspondingly, the image of other type, such as person shown in Figure 13, almost disconnected pen, but comprise the noise of considerable amount.

This step can be carried out with sorter shown in Figure 2, but also can carry out with other sorter.For example, all be the extreme case of row image owing to know type and disconnected pen type, this sort operation can easily be undertaken by manual, and at this moment, described sorter can comprise input-output unit, such as display, keyboard and/or mouse etc.

Get back to Fig. 3, simultaneously with reference to Fig. 2.After step S302, at step S304, booster 204 judges that the row image belongs to any type.If know type, then use independent noise denoiser 214 to remove independent noise (step S306) in the original images, described independent noise denoiser 214 can adopt any noise reduction technology, and will go the good quality of image and take into account.Resulting image is output, as the image after strengthening.The term here " independent noise " is meant the noise that does not link to each other with any stroke of character.

If disconnected pen type, then 220 pairs of capable images of first expander are carried out cross expansion (step S308).Resulting image is output, as the image after strengthening.

If the row image is neither know the type pen type that neither break, that is to say, if the row image has many noises in varying degrees, but almost not disconnected pen, so, at first, use 214 pairs of independent noise noise reductions of described independent noise denoiser (step S310), then, use 218 pairs of resulting images of smoother to carry out smoothly (step S312), the image that obtains at last is output, as the image after strengthening.

(second embodiment)

When carrying out classification step S302 with sorter shown in Figure 2 202, sorter 202 may further include second expander 206 and spot noise counter 208, and classification step also comprises step as shown in Figure 4.At first, the quantity A of the spot noise in 208 pairs of former images that begin of spot noise counter counting (step S402).Then, in step S408, judge that whether the A value is less than first predetermined threshold.If the type of the then former image that begins is for knowing type.If not, then 206 pairs of former images that begin of second expander are carried out cross expansive working (step S404), and count (step S406) by the quantity D of the spot noise in 208 pairs of resulting images of spot noise counter.When carrying out cross and expand, the projection by the row image distributes and finds out possible character zone, only carries out the cross expansive working in these character zones.Therefore, after expanding, the effect of the noise between each character area (it does not link to each other with character area) has been eliminated.

Obviously, it is readily appreciated by a person skilled in the art that in image processing process, any intermediate result of original image or Flame Image Process can be preserved for using in the future.Therefore, above-mentioned four step S402, S408, S404 and S406, and step subsequently can be arbitrarily such as the order of S410 (and other step in other accompanying drawing), condition is that step S404 is directly or indirectly before step S406, directly or indirectly after step S402, the determining step that relates to value D is directly or indirectly after step S406 for the determining step that relates to value A.

Get back to Fig. 4.In step S410, whether judgment value A and value D satisfy first predetermined relationship (R1).If satisfy, the then former image that begins belongs to disconnected pen type, otherwise the former image that begins belongs to other type.

Here, first predetermined threshold (TH1) and first predetermined relationship can change with language (Chinese, Japanese or Korean etc.).For Chinese, for example, TH1 can be that 5, the first predetermined relationship R1 can be expressed as:

D×4＜A (R1)

For the modification or the modification of first predetermined threshold and first predetermined relationship, the back has detailed discussion.

In this embodiment, aforementioned first expander 220 is carried out identical cross expansion with second expander 206.Therefore, first expander and second expander can be merged into one, and the cross expansive working only need be carried out once just much of that.From another angle, can consider like this: during classification step, expander is carried out cross and is expanded, and is strengthening step, if the row image belongs to disconnected pen type, the image after then expanding with cross replaces the former image that begins.

(the 3rd embodiment)

In first and second embodiment, the former image that begins can be classified as one of three types, comprises knowing type, disconnected pen type and other type.

In more preferred the 3rd embodiment, described other type can further be classified.Particularly, as shown in Figure 5, after the step S410 of Fig. 4, can be further at step S512 judgment value A whether less than second predetermined threshold (TH2).If, then whether satisfy second predetermined relationship (R2) according to A and D, the former image classification that begins is disconnected pen type or spot noise type (step S514); Otherwise the former image classification that begins is other type.Here, " spot noise type " is meant that the noise ratio that image comprises knows that type is many, but noise is point-like basically.

Equally, second predetermined threshold (TH2) and second predetermined relationship can change with language (Chinese, Japanese or Korean etc.).For Chinese, for example, TH2 can be that 10, the second predetermined relationship R2 can be expressed as:

D×2＜A (R2)

For the modification or the modification of second predetermined threshold and second predetermined relationship, the back has detailed discussion.

Owing to another image type occurred, after the step S304 of Fig. 3, carry out another judgement (step S614), judge whether original image belongs to the spot noise type.If treatment step so subsequently is similar to shown in the S310 and S312 among Fig. 3.If not, then between the rapid S620 of the peaceful sliding steps of step (S616) of deletion independent noise, add an intermediate value noise reduction step (S618).In this case, booster 204 also comprises the improved median filter 216 of a conventional art.As shown in figure 10, improved median filter uses 3 * 3 templates.If P is a black, at least five is white in its eight neighbors (A, B, C, D, E, F, G and H), and B and G are not black, and D and E be not black, then P is set at white.

(the 4th embodiment)

In this embodiment, sorter 202 may further include the character counter 212 of the quantity C of the possible character that is used for the count line image, so that sorter 202 can more accurately be classified to image.Here, obtain the vertical projection of capable image, seeking possible character zone, and count their quantity C.

Among the embodiment in front, only consider A and D, because for the capable image of knowing type and disconnected pen type, the quantity of spot noise is considerably less.But for other type with a considerable amount of noises, the quantity of noise changes along with the length (the possible number of characters C in the just relevant row image) of row image.Therefore, in order to carry out more sophisticated category, just need to consider C to having a considerable amount of capable images.

For example, as shown in Figure 7, after the step S512 of Fig. 5, can further judge whether to satisfy the 3rd predetermined relationship (R3) (step S716) between value A and the value C.If satisfy, be the spot noise type then, otherwise be categorized as other type with this row image classification.

For example, for Chinese, the 3rd predetermined relationship R3 can be expressed as:

A＜C (R3)

For modification or the modification of the 3rd predetermined relationship R3, the back has detailed discussion.

As described in the present embodiment and second embodiment, first and second expanders (they can merge into) and character counter all need to obtain possible character zone, for example pass through the vertical projection of row image.Obviously, obtaining aspect the possible character zone, expander can utilize the result of character counter, and vice versa.

(the 5th embodiment)

Further, according to the relation between A and the C, above-mentioned other type can further be categorized as a kind of at least two kinds thinner type.As previously described in the embodiment, C can influence the value of A.Therefore, for example, the classification of the former image that begins can be followed following principle:

Do not concern that R3 still satisfies the 4th and concerns R4 if A and C do not satisfy above-mentioned the 3rd, then the former image classification that begins can be few noise type, the meaning is that amount of noise is more than the spot noise type.Do not concern that R4 still satisfies the 5th and concerns R5 if A and C do not satisfy the above-mentioned the 4th, then original image can be classified as middle noise type, and its amount of noise is more than few noise type.Do not concern R5 if A and C do not satisfy the above-mentioned the 5th, the then former image that begins can be classified as many noise types, and its amount of noise is more than middle noise type.

For Chinese, for example, the 4th concerns that R4 and the 5th concerns that R5 can be expressed as:

A＜2×C (R4)

A＜4×C (R5)

Concern that for the 4th R4 and the 5th concerns the variation of R5, the back has detailed discussion.

But under extreme case, possible C=0 this means that relevant image line may not comprise any character.But this image line still may comprise some characters really.Therefore, even C=0 still needs this row image classification.In this case, the classification of the former image that begins can be followed following principle:

Less than the 3rd threshold value (TH3), then the former image classification that begins can be few noise type if A is not less than above-mentioned second predetermined threshold, the meaning is that amount of noise is more than the spot noise type.Less than the 4th predetermined threshold (TH4), then original image can be classified as middle noise type if A is not less than the 3rd predetermined threshold TH3, and its amount of noise is more than few noise type.If A is not less than the 4th predetermined threshold TH4, the then former image that begins can be classified as many noise types, and its amount of noise is more than middle noise type.

For Chinese, for example, TH3 can be 20, and TH4 can be 50.For their variation, the back has detailed discussion.

(the 6th embodiment)

As another preferred embodiment, sorter 202 may further include a block noise counter 210, is used for the quantity B counting to the block noise of the former image that begins.

Obviously, when the picture quality variation, spot noise quantity and block noise quantity can rise.When the non-constant of picture quality, a large amount of spot noises can be overlapped, thereby be interconnected, and forms more block noise.That is to say that when picture quality is good (just when A hour), the value of B is inessential; When poor image quality (just, when A is big), the value of B becomes even more important, and therefore, based on the value of B, sorter can further be categorized as one of at least two kinds of meticulousr types with many noise types.

For example, if C=0 then can follow following principle many noise types further are categorized as thinner type:

If B less than the 5th predetermined threshold (TH5), is many noises L1 type with the former image classification that begins then; Less than the 6th predetermined threshold (TH6), is many noises L2 type with the former image classification that begins if B is not less than TH5 then; If B is not less than TH6, be many noises L3 type then with the former image classification that begins, wherein, the amount of noise in many noises L2 type still is less than many noises L3 type more than many noises L1 type.

Equally, for Chinese, for example, TH5 can be 10, and TH6 can be 20.For their variation, the back has detailed discussion.

Again for example, if C is not equal to 0, then the classification of the former image that begins of many noise types can be followed following table:

Table 1

Type	Condition 1	Condition 2	Condition 3	Condition 4
					Many noises L1	A＞30 and C * 10 〉=A＞C * 6 and C 〉=B * 2	30 〉=A 〉=10 and A＞C * 4 and B * 2≤C * 3 and B≤C	C * 6 〉=A 〉=C * 4 and A 〉=10 and B * 2≤C * 3 and B≤C
Many noises L2	A＞30 and A＞C * 6 and B * 2＞C 〉=B	A＞30 and A＞C * 10 and C 〉=B * 2	30 〉=A 〉=10 and A＞C * 4 and B * 2≤C * 3 and B＞C	C * 6 〉=A 〉=C * 4 and A 〉=10 and B * 2≤C * 3 and B＞C
					Many noises L3	A＞30 and A＞C * 6 and B＞C	30 〉=A 〉=10 and A＞C * 4 and B * 2＞C * 3	C * 6 〉=A＞C * 4 and A 〉=10 and B * 2＞C * 3

Equally, the threshold value that comprised of last table or relation are just at the example of Chinese.For their variation, the back has detailed discussion.

(the 7th embodiment)

The quantity D of the spot noise in the image after the quantity B of the block noise in the quantity A of the spot noise in the former image that begins, the former image that begins and cross expand can in all sorts of ways and be determined.A kind of method is the connected domain in the capable image of mark, comprises four connected domains and eight connected domains.

In this case, as shown in Figure 2, sorter may further include connected component labeling device 222, is used for black eight connected domains of the former image that begins of mark, and is used for black four connected domains of mark from the image of second expander, 206 outputs.Then, the specific connected domain that spot noise counter 208 and block noise counter 210 will satisfy preassigned is considered as noise, and counts its quantity.

The notion of connected domain is known, is illustrated below.For example, " black eight connected domains " are one eight zones that is communicated with.The meaning of " deceiving " is that connected domain is determined by black pixel.That is to say that " black eight connected domains " are one eight black pixels that is communicated with.The conceptual description of " connectivity of pixels " relation between two or the more a plurality of pixel.For two pixels, become and be interconnected, they must meet some requirements aspect pixel intensity and the space adjacency.At first, for two pixels, become and be interconnected, its pixel value must be all in same sets of pixel values V.For grayscale image, V can be any grey level range, V={22 for example, and 23 ... 40}.For bianry image, can be V={1}.In order to provide the formula that is used for connective adjacency standard, at first to introduce the notion of " neighborhood ".For have coordinate (x, pixel p y), collection of pixels:

N4(p)＝{(x+1，y)，(x-1，y)，(x，y+1)，(x，y-1)} (E1)

Be called as its 4-neighborhood.Its 8-neighborhood is following set:

N8(p)＝N4(p)∪{(x+1，y+1)，(x+1，y-1)，(x-1，y+1)，(x-1，y-1)}(E2)

Can obtain the definition that four connected sums eight are communicated with thus:

For two pixel p and q, its pixel value all belongs to set V, so, if it belongs to set N4 (p), then is four connections, if it belongs to set N8 (p), then is eight connections.

Obviously, the stroke of character is bigger connected domain.Noise is less connected domain.According to the average character duration in the row image, the little connected domain that can be defined as noise is divided into two classes: spot noise connected domain and block noise connected domain.Which is similar to the connected domain of effective stroke, and just the block noise connected domain need be noted more.

Consider that in the capable image of Chinese the relation between picture quality and the black and white connected domain quantity ratio is very difficult to determine, at step S402 and step S406, only deceives connected component labeling (perhaps statistics).At step S402, carry out black eight connected domains statistics.As mentioned above, at step S404, image is carried out cross expand (for example 3 * 3).In the capable image after expansion, disconnected pen is reconnected, and their connectedness normally four is communicated with.Therefore, at step S406, carry out black four connected domains statistics.Therefore, definite system of A and B is based on the result of black eight connected domains statistics, and definite system of D is based on the statistics of black four connected domains of the image after cross is expanded.

Normal image with noise always comprises many spot noise connected domains.Its quantity can be used to judge the little connected domain block noise whether that is similar to effective stroke.The block noise connected domain is many more, and row picture quality is low more, just needs stronger noise-reduction method.

Based on black eight connected component labelings, calculation level noise connected domain quantity (A) and block noise connected domain quantity (B).

An example of deterministic process is as described below, and wherein H and W are the height and the width of connected domain.

dotSize ＝3 (averWidth＜32)

＝4 (48＞averWidth≥32)

＝averWidth/8-1 (averWidth≥40)

blockSize＝dotSize+2

Wherein, dotSize and blockSize are used to judge that a connected domain is the threshold value of spot noise connected domain or block noise connected domain.AverWidth is the average character duration in the row image.

Table 2

Noise type	Condition
		The spot noise connected domain	H＜dotSize and W＜dotSize
The block noise connected domain	(dotSize≤H＜blockSize and W＜blockSize) or, (H＜blockSize and dotSize≤W＜blockSize) or, (H＜dotSize and dotSize≤W≤dotSize * 2) or, (dotSize≤H≤dotSize * 2 and W＜dotSize)

(the 8th embodiment)

In addition, except the noise of each character zone inside, also have the upper and lower edge noise in the row image, these edge noises might influence the accuracy that row is cut apart.

Therefore, the method according to this invention can also comprise the step of the edge noise up and down that is used to eliminate capable image, so that the up-and-down boundary of row image is more accurate.Correspondingly, equipment of the present invention can also comprise edge noise denoiser 222 (see figure 2)s, is used to eliminate edge noise.In Fig. 2, the edge noise denoiser is illustrated as parts of booster 204.But enhancing operation of describing among the embodiment in front and the operation of the edge noise noise reduction in the present embodiment are separate, and any one in them can constitute an invention.Therefore, in Fig. 2, described equipment can comprise described sorter 202 and described booster 204 (not having edge noise denoiser 222), perhaps comprises sorter 202 and edge noise denoiser, perhaps comprises sorter 202 and booster 204 and edge noise denoiser.Under in the end a kind of situation, the edge noise denoiser can be to receive the output of sorter 202 and provide the parts of input to booster 204, perhaps can be the parts that receive the output of booster 204.

For method, eliminate the step of edge noise and can carry out in any stage after the step S302 among Fig. 3 and Fig. 6, perhaps, as another invention, this step can replace among Fig. 3 and Fig. 6 the institute after the step S302 in steps.

In the capable cutting procedure before key step of the present invention, because picture quality also is not classified, the noise between the row may be integrated among the capable image, to keep row information as much as possible.In addition, the many noises between the row have coupled together with effective row image.

Obviously, the blank between the row is much larger than the blank between the character.Therefore, the noise between the row is bigger to the influence of segmentation result.

The step of edge noise about being used for eliminating (noise between the row just) and described edge noise denoiser system distribute based on the projection of the horizontal direction of row image.In this step,, adopt two kinds of different projecting methods respectively for knowing type (and disconnected pen type) and other picture quality type.

When the row image be when knowing type or disconnected pen type, the segmenting device of edge noise denoiser is configured to count the black pixel in each pixel column.Then, this segmenting device is found out the blank pixel row, is the plurality of vertical ground capable image of child arranged side by side with the row image segmentation.Then, the denoising device of edge noise denoiser judge according to the average character duration of the height of the capable image of child and row and eliminates about edge noise.

As an example, as the capable height of fruit is H1, the height of the remainder of row image is H2, so, if H1 * 6＜=averWidth and H2 * 3＞averWidth * 2 (wherein, averWidth is the average character duration of row), then should the child capable noise that is considered as is capable and eliminate, and adjusts corresponding row image boundary.

For other image quality level, described segmenting device is to each pixel column and next adjacent lines of pixels actuating logic and (AND) operation.In resulting pixel column, described segmenting device is counted the black pixel in the collection of pixels with at least two continuously black pixels.Such histogram can suppress The noise.

Provide a preliminary threshold value by histogrammic average height.Recomputate threshold value according to row picture quality then:

basicThresh＝max(histAver/20，1) (E3)

Wherein, basicThresh is preliminary threshold value, and histAver is the histogrammic average height of row image, and the histogrammic area that equals the row image is divided by line height.

To recomputating of preliminary threshold value (wherein: " Thresh " represents threshold value) as shown in the table:

Table 3

Spot noise, few noise	Thresh＝basicThresh
		Middle noise	Thresh＝3×basicThresh/2
Many noises L1	Thresh＝2×basicThresh
		Many noises L2	Thresh＝4×basicThresh
Many noises L3	Thresh＝10×basicThresh

Be similar to the image of knowing type, described segmenting device is sought so-called " blank " pixel column, and just the standoff height in histogram is less than the row of threshold value (the as above Thresh shown in the table).Segmenting device will be gone image segmentation and will be side by side plurality of sub row image vertically.Then, denoising device is judged according to the average character duration of the histogrammic average height of the histogrammic average height of the height of the capable image of child, whole capable image, the capable image of each height and row and is eliminated edge noise up and down.

For example, as the capable height of fruit is H1, the height of the remainder of row image is H2, the histogrammic average height of son row be histAver1 and the histogrammic area that equals the son row divided by sub-line height, so, if H1 * 8＞averWidt and H1 * 3＜averwidth and histAver1＜=histAver, if perhaps H1 * 8＜=averWidth and H2 * 3＞averWidth * 2, then should child capable to be regarded as noise capable, it can be removed, thereby correspondingly adjust the border of row image.

Operation in the present embodiment has the remarkable noise-reducing effect to noisy Chinese file and picture, and does not lose (removal) effectively capable information.This step is not only served Character segmentation process afterwards, and is used for adjusting capable the preceding segmentation result.

(independent noise noise reduction)

Among the embodiment in front, described booster 204 and can comprise the independent noise denoiser, and, for the capable image (except that resolving pen type) of each type, can carry out the step (S306, S310 and S616) of a deletion independent noise.As previously mentioned, for the deletion of independent noise, there are many technology to adopt.In a preferred embodiment, the inventor provides a kind of new method that is used to remove independent noise, and is as described below.

In order to remove the independent noise in the capable image,,, provide the scope of the size of the connected domain that should be regarded as independent noise according to the average character duration and the quality grade thereof of row image for the image of each type.According to the scope of described size, some connected domains are regarded as independent noise, thereby are directly removed.For other connected domain, calculate black picture element density or black continuously picture element density, and, other connected domain is judged according to result calculated, see which connected domain is noise and need be removed.

The bound of the scope of described size is as follows:

Limit=1 (clear)

=max (dotSize/3,1) (spot noise)

=max (dotSize/3+1,2) (other picture quality)

DotS_m=2 (clear)

=max ((dotSize * 2+1)/3,1) (spot noise)

=dotSize (other picture quality)

DotS_2x=dotS_m (clear)

=2 * dotS_m (other picture quality)

dotS_3x ＝3×dotS_m

dotS_4x ＝4×dotS_m

Here, in above-mentioned the 7th embodiment, obtained dotSize.

In three kinds of situations of following table (step 1402 of Figure 15), connected domain is judged as little noise, with deleted (step 1404 among Figure 14).H and W are the height and the width of connected domain.

Table 4

Situation 1	H＜dotS_m and W＜dotS_m
		Situation 2	H≤limit and W≤dotS_2x
Situation 3	H≤dotS_2x and W≤limit

For other size, if H≤dotS_2x and W≤dotS_2x (step 1406 among Figure 14) then calculate black picture element density (step 1408 among Figure 14).When 3 * blackCount＜H * W (step 1410 among Figure 14), then delete this connected domain (step 1412 among Figure 14), wherein blackCount is the quantity that comprises the black pixel among the rectangle H * W of described connected domain.

Described else if size satisfies any situation (step 1414 among Figure 14) in the following table, then calculates black picture element density and continuously black picture element density (density that has the black pixel in the pixel groups of at least four continuously black pixels in a pixel column) (step S1416 among Figure 14 and S1418).If 3 * blackCount＜H * W and black continuously picture element density are then deleted this connected domain (step 1422 among Figure 14) less than 30% (step 1420 among Figure 14).

Table 5

Situation 1	H＜dotS_4x and W≤dotS_3x
		Situation 2	H≤dotS_3x and W＜dotS_4x

As mentioned above, the connected domain that satisfies preassigned is regarded as noise and is removed.Here, connected domain can be the result that the connected component labeling device 222 in the sorter produces.Described independent noise denoiser of it would, of course, also be possible to 214 itself or described booster 204 can comprise that another marker (not shown) is used for the mark connected domain.

(smooth operation)

Among the embodiment in front, described booster 204 and can comprise a smoother 218, and, for the type of knowing outside type and the disconnected pen type, carry out a level and smooth step (S312 and S620).

Level and smooth noise reduction is in order to eliminate the burr around stroke, is a kind of weak noise-reduction method.By many tests and the analysis to test pattern, the inventor recognizes, for the image of different quality, and the different number of times that smooth operation need be circulated, as shown in the table:

Table 6

Picture quality	Cycle index
		Spot noise, few noise, middle noise	2
Many noises L1	3
		Many noises L2, many noises L3	4

(span of threshold value and the quantity of type)

Many threshold values and relational expression have been mentioned in the explanation in front.As previously mentioned, some top object lessons are at Chinese.If the language difference, when being Korean or Japanese, described threshold value and relational expression can correspondingly change.Even for Chinese, they also can change.

At first, when relating to the scope of value, when perhaps relating to two comparisons between the value, end value both can belong to the scope greater than this end value, also can belong to the scope less than this end value, as long as this end value does not belong to described two scopes simultaneously.For example, if mentioned two condition X＜Y and X＞=Y, they also can be understood that or be modified to X＜=Y and X＞Y.

Secondly, name and the quantity thereof to type just is used for illustrative purposes above, should not be interpreted as determinate.Aforementioned type is such as " knowing type ", " disconnected pen type ", " spot noise type ", " few noise type ", " middle noise type " or " many noise types " (comprising many noises L1, many noises L2 and many noises L3), the name that also can be named as other arrives L8 such as L1.But as previously mentioned, under the enlightenment of this instructions, these types can be merged into thicker type still less, perhaps are broken down into more thinner type, and perhaps the various combination by merging and decomposing is different types with their reorganizations.

Go through the theoretical foundation of aforesaid threshold values and relational expression below, and they are provided more example.

1.dotSize and blockSize

Obviously, the value of A, B and D is relevant with the value of dotSize and blockSize.If adjusted the latter, the former can change a lot, thereby correspondingly, the criterion in each step (such as S408, S410, S512, S514, some steps among S716 and the 5th and the 6th embodiment) can change.Concrete value can obtain by test.

For example, dotSize and blokSize can be adjusted to:

DotSize=2 are to 3 (averWidth＜32)

=2 to 4 (48＞averWidth 〉=32)

=2 to averWidth/8-1 (averWidth 〉=40)

BlockSize=dotSize+ (1 to 3)

2. similarly, in table 2, when definite spot noise connected domain and block noise connected domain, its big or small scope can narrow or relax 1 or 2 pixel.Certainly, corresponding with the adjustment of magnitude range, the value of A, B and D, thereby the criterion in each step have quite big degree change.Occurrence can obtain by test.

Additional disclosure when the quantity D of timing point noise after cross expands, can amplify 2 with dotSize, because after cross expanded, connected domain can enlarge 2 pixels.

3. when to the row image classification, general, criterion can be more conservative.That is to say that from knowing that type (perhaps disconnected pen type) to the spot noise type, until many noises L3 type, can adjust greatlyyer by the threshold value of upper limit type, the threshold value of lower limit type can be adjusted forr a short time.Along with criterion becomes conservative, noise reduction of the present invention (thereby figure image intensifying effect) can be lowered.Extreme case is that any capable image all is judged as and knows type.For example, if the first predetermined threshold TH1 is a very large value, then the present invention can almost not have noise reduction (figure image intensifying) effect, but can not produce minus effect yet.

But if criterion becomes too radical (threshold value of upper limit type is fallen too lowly), then some capable image with high image quality can be judged as and have inferior quality, and accept noise reduction and operate, thereby cause minus effect, just, the degree of accuracy of cutting apart can descend.

Guarantee that simultaneously the present invention can not produce secondary effect if adjust described threshold value, must follow following principle to adjustment so as the threshold value of the upper limit in radical direction:

TH1 that uses when type and disconnected pen type are known in judgement and TH2 can be lowered and be no more than 20%.For example, TH1 can be 4, preferably is not less than 4.The upper limit of A can descend and is no more than 30% when being used for judging point noise type (step S716, the 3rd predetermined relationship R3).The upper limit of the A that uses when judging few noise type and middle noise type (TH3, TH4, R4 and R5) can descend and be no more than 50%.The A that uses when judging many noises L1 type and the upper limit of B can descend and be no more than 80%.For many noises L2 and many noises L3 type, the lower limit of A and B can improve 100%, even surpasses 100%.Picture quality is poor more, because excessively the secondary effect of noise reduction generation is more little, just, secondary effect can not offset the plus effect of noise reduction.

4. in the step of eliminating edge noise, basicThresh and other threshold value can be more conservative, even are 0.This moment, any capable image all was regarded as knowing type.The increase of described threshold value can not surpass 20%, otherwise, because excessively noise reduction can produce negative effects.

5. in the step of deletion independent noise, the criterion that relates to H and W can be more conservative.The upper limit even can be 0, this moment, in fact this step did not carry out any noise reduction operation.In radical direction, the described upper limit can increase one or two pixels.When considering black picture element density, the upper limit can be reduced to 0 (in conservative direction), and perhaps rising is no more than 30% and do not have a negative effects.When considering continuously black picture element density, the upper limit can drop to 0 (in conservative direction), and perhaps rising is no more than 20% and do not have a negative effects.

6. in smooth operation, the round-robin number of times can be 0, and this moment, in fact this step did not carry out any noise reduction operation.In radical direction, for spot noise type and few noise type, cycle index can increase by 1 time, and centering noise type and many noises L1 type can increase by 2 times, can increase by 4 times for many noises L2 and many noises L3, and not have negative effects.

Applicating example of the present invention and effect thereof

As an example, the present invention can be used for discerning the type fount OCR engine of the Chinese file and picture of various quality.Also can be used to cut apart other Chinese character (2 byte) file and picture, for example Japanese, Korean etc.

The process flow diagram of Fig. 8 is used for the key step of diagram OCR.The present invention can be applied to row and strengthen step.Use this algorithm, can significantly improve the Character segmentation degree of accuracy of the capable image in the inferior quality Chinese file and picture, thereby improve the performance of whole OCR engine.

Fig. 9 illustrates with using OCR engine of the present invention to cut apart and discern the example that result that Chinese file and picture obtains compares with prior art (baseline results).Fig. 9 illustrates the Character segmentation accuracy (Segment_high) of high quality graphic, the Character segmentation accuracy (Segment_low) of low-quality image, the discrimination (Recog_high) of high quality graphic and the discrimination (Recog_low) of low-quality image.

Sample is the simplified Chinese characters file and pictures.Its quantity and distribution are as follows:

High-quality: 94900 characters;

Inferior quality: 97210 characters.

Can see that the present invention is better than the algorithm of prior art to the recognition result of inferior quality document, the discrimination to high quality graphic does not descend simultaneously, even also better.

Storage medium

Described purpose of the present invention can also be by realizing with program of operation or batch processing on any messaging device that described image source is communicated by letter with subsequent processing device aforesaid.Described messaging device, image source and subsequent processing device are known common apparatus.Therefore, described purpose of the present invention also can be only by providing the program code of realizing described method or equipment to realize.That is to say that the storage medium that stores the program code of realizing described method or equipment constitutes the present invention.

To those skilled in the art, can realize described method with any program language programming easily.Therefore, omitted detailed description at this to described program code.

Obviously, described storage medium can be well known by persons skilled in the art, and perhaps therefore the storage medium of any kind that is developed in the future also there is no need at this various storage mediums to be enumerated one by one.

Although in conjunction with concrete steps and structrual description the present invention, the present invention is not limited to details as described herein.The application should cover all variation, modification and modification without departing from the spirit and scope of the present invention.

Claims (25)

1. a method that strengthens the character row image comprises the steps:

Original image is categorized as a kind of at least three types, but but these three types comprise best the knowing type, almost do not have the disconnected pen type of noise stroke for disconnected pen of picture quality, and have than knowing noise that type Duo other type of almost disconnected; And,

Strengthen described original image according to its type, comprising:

If original image is to know type then delete independent noise;

Perhaps, if original image be disconnected pen type then original image is carried out cross expand;

Perhaps, if original image be other type then delete independent noise, then resulting image is carried out smoothly.

2. the method for claim 1, wherein:

Described classification step comprises following substep:

A) the low noise quantity A in the counting original image knows type if A is categorized as less than first predetermined threshold then with original image;

B) if A is not less than described first predetermined threshold, then each character zone in the original image being carried out cross expands, count the spot noise quantity D in the resulting image then, if A and D satisfy first predetermined relationship then original image is categorized as disconnected pen type.

3. method as claimed in claim 2, wherein, the spot noise in the original image is black eight connected domains of size less than a predetermined threshold, the spot noise in the image after cross expands is black four connected domains of size less than a predetermined threshold.

4. method as claimed in claim 3, wherein:

Described substep b) also comprise:

If A is not less than described first predetermined threshold still less than second predetermined threshold

And if satisfied second predetermined relationship, then original image would be categorized as disconnected pen type;

Otherwise original image is classified as the second-best spot noise type of picture quality, and,

Described enhancing step also comprises: for the original image of knowing the type outside type, spot noise type, the disconnected pen type, the deletion independent noise is carried out the intermediate value noise reduction to resulting image, level and smooth then resulting image then.

5. method as claimed in claim 4, wherein, described classification step also is included in described substep b) before a sub-steps a '): the possible number of characters C in the counting original image, and described substep b) also comprise: satisfy the predetermined relationship between A and the C if A is not less than described second predetermined threshold, then original image is categorized as the spot noise type.

6. method as claimed in claim 5, wherein:

According to the value of A and/or the relation between A value and the C value, know that the original image of the type outside type, disconnected pen type and the spot noise type further is categorized as one of at least two kinds of types with different images quality; And,

Resulting image is repeated described level and smooth step pre-determined number, and this number of times depends on the type of original image, makes that the picture quality of original image is high more, and level and smooth step is repeated fewly more.

7. method as claimed in claim 6, wherein:

Described classification step also comprises the block noise quantity B in the counting original image; And,

When original image being classified as when one of described at least two types, consider the value of B and/or the relation between B value and the C value.

8. method as claimed in claim 7, wherein, the block noise in the original image be size be not less than predetermined threshold but less than black eight connected domains of another predetermined threshold.

9. as the described method of one of claim 1 to 8, wherein, in the step of deletion independent noise, the magnitude range of independent noise is that the type according to the average character duration of row image and row image provides.

10. method as claimed in claim 9, wherein, in the step of deletion independent noise, undersized connected domain is regarded as noise and is directly deleted, and depends on its black picture element density or black continuously picture element density greater than the deletion or the reservation of described undersized connected domain.

11. as the described method of one of claim 1 to 8, comprise also being used to eliminate the step of noise in the ranks that it comprises following substep:

To go image division for sub capable according to capable image projection histogram in the horizontal direction; And

Judge and the elimination edge noise according to following standard:

For the capable image of knowing type or disconnected pen type, according to the height of the capable image of child and the average character duration of row image;

For the capable image of other type, according to the histogrammic average height of the height of the capable image of child, whole capable image, the histogrammic average height of the capable image of each height and the average character duration of row image.

12. method as claimed in claim 11, wherein:

For the capable image of knowing type or disconnected pen type, form projection histogram by the black pixel of counting in each pixel column,

Capable image for other type, form projection histogram by following manner: to each pixel column and next adjacent pixels row actuating logic AND-operation thereof, then in resulting pixel column to the black pixel counts in the pixel groups with at least two continuously black pixels.

13. an equipment that strengthens the character row image comprises:

Sorter, be used for original image is categorized as at least three types a kind of, but but these three types comprise best the knowing type, almost do not have the disconnected pen type of noise stroke for disconnected pen of picture quality, and have than knowing noise that type Duo other type of almost disconnected; And

Booster is used for strengthening described original image according to the type of original image, comprising:

The independent noise denoiser, being used at original image is deletion independent noise when knowing type or other type;

First expander, being used at original image is that disconnected pen type is then carried out the cross expansion to original image; And

Smoother is used for carrying out smooth operation for the image from the output of independent noise denoiser when original image is other type.

14. equipment as claimed in claim 13, wherein:

Described sorter comprises:

Second expander is used for that each character zone of original image is carried out cross and expands;

The spot noise counter is used for counting the spot noise quantity A of original image, and is used for counting the spot noise quantity D from the image of second expander output; And

Described sorter be further configured into: know type if A less than first predetermined threshold, then is categorized as original image; Otherwise, if A and D satisfy first predetermined relationship then original image are categorized as disconnected pen type.

15. equipment as claimed in claim 14, wherein, described sorter also comprises the connected component labeling device, black eight connected domains that are used for the former image that begins of mark, and be used for black four connected domains of mark from the image of described second expander output, and described spot noise counter further is configured to size is considered as spot noise less than described black eight connected domains of a predetermined threshold and size less than described black four connected domains of a predetermined threshold.

16. equipment as claimed in claim 15, wherein:

Described sorter be further configured into:

If A is not less than described first predetermined threshold still less than second predetermined threshold

And if satisfied second predetermined relationship, then original image would be categorized as disconnected pen type;

Otherwise original image is classified as the second-best spot noise type of picture quality, and,

Described booster also comprises median filter, be used for: for the original image of knowing the type outside type, spot noise type, the disconnected pen type, image from described independent noise denoiser output is carried out the intermediate value noise reduction, and resulting image is output to described smoother.

17. equipment as claimed in claim 16, wherein, described sorter also comprises character counter, be used for counting the possible number of characters C of original image, and described sorter be further configured for: satisfy predetermined relationship between A and the C if A is not less than described second predetermined threshold, then original image be categorized as the spot noise type.

18. equipment as claimed in claim 17, wherein:

Described sorter is configured to: according to the value of A and/or the relation between A value and the C value, the original image of knowing the type outside type, disconnected pen type and the spot noise type further is categorized as one of at least two kinds of types with different images quality; And,

Described booster be further configured for: control described smoother, resulting image is repeated the smooth operation pre-determined number, this number of times depends on the type of original image, makes that the picture quality of original image is high more, and smooth operation is repeated fewly more.

19. equipment as claimed in claim 18, wherein:

Described sorter also comprises the block noise counter, is used for counting the block noise quantity B of original image; And

Described sorter further is configured to: when original image being categorized as when one of described at least two types, consider the value of B and/or the relation between B value and the C value.

20. equipment as claimed in claim 19, wherein, but described block noise counter is further configured and is not less than predetermined threshold for: size is regarded as block noise less than black eight connected domains of another predetermined threshold.

21. as the described equipment of one of claim 13 to 20, wherein, described independent noise denoiser is configured to: the magnitude range of independent noise be according to the row image average character duration and the row image type provide.

22. equipment as claimed in claim 21, wherein, described independent noise denoiser is configured to: undersized connected domain is regarded as noise and is directly deleted, and depends on its black picture element density or black continuously picture element density greater than the deletion or the reservation of described undersized connected domain.

23. as the described equipment of one of claim 13 to 20, comprise also being used to eliminate the edge noise denoiser of noise in the ranks that it comprises:

Segmenting device is used for will going image division for sub capable according to capable image projection histogram in the horizontal direction; And

Denoising device is used for judging and the elimination edge noise according to following standard:

For the capable image of knowing type or disconnected pen type, according to the height of the capable image of child and the average character duration of row image;

For the capable image of other type, according to the histogrammic average height of the height of the capable image of child, whole capable image, the histogrammic average height of the capable image of each height and the average character duration of row image.

24. equipment as claimed in claim 23, wherein: described segmenting device be further configured into:

For the capable image of knowing type or disconnected pen type, form projection histogram by the black pixel of counting in each pixel column,

Capable image for other type, form projection histogram by following manner: to each pixel column and next adjacent pixels row actuating logic AND-operation thereof, then in resulting pixel column to the black pixel counts in the pixel groups with at least two continuously black pixels.

25. a storage medium is characterized in that, has stored the program code that is used to realize as the described method of one of claim 1 to 12 therein.

Images