CN103593666A - An image identification method, a filtering method and relative apparatuses - Google Patents

Abstract

Embodiments of the invention disclose an image identification method, a filtering method and relative apparatuses. The identification method comprises the following steps: extracting a character region block on a target image; extracting a strong edge region of the target image by utilizing an image edge algorithm; judging whether the strong edge region is a background region of the character region block; if so, extending the background region for the character region block, and identifying the extended region as an auxiliary information region; if not, identifying the character region block as the auxiliary information region; and after identification, determining the type of an image in the auxiliary information region based on any one of or a plurality of region information and content information of the auxiliary information region. According to the image identification method, the auxiliary information region can be identified; and abnormal images can be filtered according to the auxiliary information region, so that the harm, caused by abnormal images, to consumers or managers is prevented.

Description

A kind of image-recognizing method, filter method and relevant apparatus

Technical field

The application relates to technical field of image processing, particularly relates to a kind of image-recognizing method, filter method and relevant apparatus.

Background technology

The image that is used for displaying merchandise in shopping website Shang， businessman generally comprises two parts content: the supplementary region of some supplementarys of the commodity image of displaying merchandise state and the displaying merchandise increasing at commodity image.These supplementary regions are generally used for the brand of sign commodity, the feature of introducing commodity or publicity sales promotion, information of discount etc., generally include word (comprising Chinese character, numeral and English alphabet), trade mark and pattern etc. in supplementary region.

But the content in some supplementary regions probably includes the misleading information such as advertisement unreal or exaggeration, consumer can bring loss economically because of wrong these misleading information of letter.In addition, also have some supplementary regions often with the ways of presentation " a presumptuous guest usurps the role of the host " of various exaggerations, become the focus of vision, affect the bandwagon effect of commodity image itself.And businessman and third party's shopping platform also can be subject to the loss in prestige, even jeopardize the existence of businessman and third party's shopping platform.Visible, some supplementary regions probably have harmfulness, and it had both affected consumer's normal consumption, have also affected the normal management of supvr to commodity.Conventionally, by comprising these images with harmfulness or the more outstanding supplementary region of interfere information, be called abnormal image.

Above-mentioned technical matters based on existing in prior art, at present in the urgent need to a kind of image-recognizing method that identifies supplementary region at the image for displaying merchandise is provided, and, the image filtering method of Exception Filter image in a kind of all images comprising the supplementary region of identifying.

Summary of the invention

In order to solve the problems of the technologies described above, the embodiment of the present application provides a kind of image-recognizing method, filter method and relevant apparatus, to identify supplementary region, and from the image that comprises supplementary region Exception Filter image, the harm of avoiding abnormal image to cause consumer or supvr.

The embodiment of the present application discloses following technical scheme:

, comprising:

Extract the character area piece on target image;

Adopt image border algorithm, extract the strong fringe region on target image;

Judge whether described strong fringe region is the background area of described character area piece;

If so, described character area piece expansion background area, the region after expansion is identified as to supplementary region, otherwise, described character area piece is identified as to supplementary region.

, comprising:

Extract the character area piece on target image;

Adopt image border algorithm, extract the strong fringe region on target image;

Judge whether described strong fringe region is the background area of character area piece;

If so, described character area piece expansion background area, the region after expansion is identified as to supplementary region, otherwise, described character area piece is identified as to supplementary region;

According to the area information in described supplementary region and any one information in content information or any number of information combination, determine the type of place, described supplementary region image.

, comprising:

Character area piece extraction module, for extracting the character area piece on target image;

Strong fringe region extraction module, for adopting image border algorithm, extracts the strong fringe region on target image;

Background area judge module, for judging whether described strong fringe region is the background area of character area piece;

Picture recognition module, for when judgment result is that of described background area judge module is, described character area piece expansion background area, is identified as supplementary region by the region after expansion, otherwise, described character area piece is identified as to supplementary region.

A device, comprising:

Character area piece extraction module, for extracting the character area piece on target image;

Strong fringe region extraction module, for adopting image border algorithm, extracts the strong fringe region on target image;

Background area judge module, for judging whether described strong fringe region is the background area of character area piece;

Picture recognition module, for when judgment result is that of described background area judge module is, described character area piece expansion background area, is identified as supplementary region by the region after expansion, otherwise, described character area piece is identified as to supplementary region;

Image filtering module, for according to any one information of the area information in described supplementary region and content information or any number of information combination, determines the type of place, described supplementary region image.

As can be seen from the above-described embodiment, extract the character area piece in target image, character area piece is carried out to background expansion, can identify the supplementary region in target image.Supplementary region based on identifying, according to the number in supplementary region, size, position or content, further filters out abnormal image, thus the harm that can avoid abnormal image to cause consumer or supvr.

Accompanying drawing explanation

In order to be illustrated more clearly in the embodiment of the present application or technical scheme of the prior art, to the accompanying drawing of required use in embodiment or description of the Prior Art be briefly described below, apparently, accompanying drawing in the following describes is only some embodiment of the application, for those of ordinary skills, do not paying under the prerequisite of creative work, can also obtain according to these accompanying drawings other accompanying drawing.

Fig. 1 is the method flow diagram of a kind of image-recognizing method of the embodiment of the present application one announcement;

Fig. 2 is the arrange statistics schematic diagram of information of the application's Chinese word;

Fig. 3 is with the commodity image schematic diagram of the background area of character area piece in the application;

Fig. 4 is the method flow diagram of a kind of image filtering method of the embodiment of the present application two announcements;

Fig. 5 is the contrast schematic diagram that filters commodity image in the application;

Fig. 6 is the method flow diagram of a kind of image filtering method of the embodiment of the present application three announcements;

Fig. 7 is the structure drawing of device of a kind of pattern recognition device of the embodiment of the present application four announcements;

Fig. 8 is the structure drawing of device of a kind of pattern recognition device of the embodiment of the present application four announcements.

Embodiment

For the application's above-mentioned purpose, feature and advantage can be become apparent more, below in conjunction with accompanying drawing, the embodiment of the present application is described in detail.

Embodiment mono-

Refer to Fig. 1, it is the method flow diagram of a kind of image-recognizing method of the embodiment of the present application one announcement, and the method comprises the following steps:

Step 101: extract the character area piece on target image;

Can, based on texture statistics method of the prior art or region analysis method, from target image, extract character area piece.Wherein, the character area piece extracting method based on texture statistics method is specially: target image is carried out to texture extraction, as, common are the texture extracting modes such as sobel (Sobel), lbp (local binary patterns) or small echo; According to the texture statistics word the extracting information of arranging; According to the word information of arranging, determine whether as character area piece, for example, the information conforms word the arrangement as shown in Figure 2 as long as word of statistics is arranged, is confirmed as character area piece.

In addition, the character area piece extracting method based on regional analysis is specially: image is carried out to extracted region, as, common are the extracted region modes such as canny or mser; According to features such as the foundation characteristic of character area, stroke feature or arrayed features, non-legible region is filtered.As, conventional foundation characteristic comprises: the ratio of ratio, region area and the fitted ellipse of region area or region area and the target image total area or the ratio of area circumference and region area etc.Conventional stroke feature comprises: the dispersion degree of symmetrical edge ratio, main stroke width or the stroke width of the average relative angle of stroke, stroke etc.Conventional arrayed feature comprises: the attributes similarities such as adjacent domain size, stroke width or color, how much alignment properties etc.

After extracting according to the method described above, the character area piece of acquisition is the boundary rectangle of character area, by the reference position of this boundary rectangle and the length and width of rectangle, identifies this boundary rectangle.

It should be noted that, the embodiment of the present application does not limit and adopts which kind of method to extract the character area piece on target image, except above-mentioned two kinds of methods enumerating, can adopt disclosed other method of prior art to extract character area piece yet.

Step 102: adopt image border algorithm, extract the strong fringe region on target image;

In a lot of supplementarys region, character area piece is not to be directly placed on commodity image, toward contact meeting, is added with a background area, outstanding to show.As shown in Figure 3, character area piece is all attended by a certain size background area, and the background area of these character area pieces belongs to the part in supplementary region too.Therefore,, when character area piece is extracted, also need to extract in the lump the background area of character area piece.

In the embodiment of the present application, provide a kind of background area identifying schemes based on strong fringe region.For example, adopt canny image border algorithm, extract the strong fringe region on target image.Certainly, the embodiment of the present application does not limit the strong fringe region that adopts which kind of method to extract target image, except the algorithm of canny image border, can adopt disclosed other method of prior art to extract strong fringe region yet.

Step 103: judge whether described strong fringe region is the background area of described character area piece, if so, enters step 104, otherwise, enter step 105;

Judge that whether described strong fringe region is that a kind of implementation of the background area of character area piece is: according to the main color rate of described strong fringe region, judge whether described strong fringe region is pure color region; If so, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

Except whether the strong fringe region that utilizes the judgement of main color rate to extract is the background area of character area piece, judge that whether described strong fringe region is that the another kind of implementation of the background area of character area piece is: judged whether that character area piece is positioned on described strong fringe region; If so, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

In addition, can also utilize main color to judge whether a strong fringe region is the background area of character area piece, judge that whether described strong fringe region is that the background area of character area piece comprises: whether the main color that judges described strong fringe region and character area piece is identical; If so, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

In addition, can also pass through Color Statistical histogram, obtain the main color of strong fringe region and character area piece.By color distance formula, calculate two distances between main color, if distance is less than a default distance threshold, the main color of strong fringe region and character area piece is close, and color distance formula is: $\mathrm{\ΔC}=\sqrt{(2+\stackrel{\‾}{r}/256)\×{\mathrm{\Δr}}^2+4\×{\mathrm{\Δg}}^2+(2+(255-\stackrel{\‾}{r})/256)\×{\mathrm{\Δb}}^2},$ Wherein,

Δ r=r ₁-r ₂, Δ g=g ₁-g ₂, Δ b=b ₁-b ₂, the main color of strong fringe region is: (r1, g1, b1), the main color of character area piece is: (r2, g2, b2).

It should be noted that, the present invention does not limit the concrete numerical value of above-mentioned default distance threshold, and the size of the different default distance thresholds of image is also different.In actual applications, can require constantly to adjust this distance threshold according to the degree of accuracy of identification.

Certainly, except utilizing respectively whether the strong fringe region of above-mentioned three kinds of methods judgement extraction is the background area of character area piece, in order to improve the accuracy of identification, can also utilize above-mentioned three kinds of methods to judge simultaneously, when three Rule of judgment all meet, think that the strong fringe region extracting is the background area of character area piece, otherwise, think that the strong fringe region extracting is not the background area of character area piece.

Step 104: described character area piece expansion background area, the region after expansion is identified as to supplementary region.

Step 105: described character area piece is identified as to supplementary region.

Finally, after determining that the strong fringe region extracting is the background area of character area piece, the character area piece expansion background area extracting in step 101; Otherwise, only retain the character area piece extracting in step 101.

As can be seen from the above-described embodiment, extract the character area piece in target image, character area piece is carried out to background expansion, can identify the supplementary region in target image.

Embodiment bis-

When identify supplementary region from target image after, some supplementary region has harmfulness, is also referred to as abnormal image, therefore, the embodiment of the present application provides a kind of image filtering method, to filter out abnormal image the image from comprising supplementary region.Refer to Fig. 4, it is the method flow diagram of a kind of image filtering method of the embodiment of the present application two announcements, comprises the following steps:

Step 401: extract the character area piece on target image;

Step 402: adopt image border algorithm, extract the strong fringe region on target image;

Step 403: judge whether described strong fringe region is the background area of character area piece, if so, enters step 404, otherwise, enter step 405;

Step 404: described character area piece expansion background area, the region after expansion is identified as to supplementary region, enters step 406;

Step 405: described character area piece is identified as to supplementary region;

The implementation of above-mentioned steps 401-405 can be referring to the step 101-105 in embodiment mono-, because this partial content describes in detail in embodiment mono-, so locate to repeat no more.

Step 406: according to the area information in described supplementary region and any one information in content information or any number of information combination, determine the type of place, described supplementary region image.

Preferably, the area information in supplementary region comprises: region number, size and the position in supplementary region.

Certainly, except above-mentioned preferred area information, can also determine according to other area information the type of place, supplementary region image, as, the color in supplementary region etc.

The content information in supplementary region is all the elements of recording in supplementary region, as word or word etc.

By above-mentioned image filtering method, its image type of distinguishing comprises abnormal image and normal picture.

Several situations take respectively below as example, illustrate according to the area information in supplementary region and content information, determine the method for the type of place, described supplementary region image.It should be noted that, when adopting a plurality of area informations to determine the type of place, supplementary region image, the present invention does not limit sequencing each other simultaneously; Can also adopt a plurality of area informations to determine the type of place, supplementary region image simultaneously, and set: when any one or any number of definite result (comprise and all determine results) are during for abnormal image, finally determine that image is abnormal image.

If determine the type of place, described supplementary region image according to the number in supplementary region, its implementation comprises: the number of adding up described supplementary region; Whether the number of judgement statistics is greater than the first preset number threshold value; If so, determine that the type of place, described supplementary region image is abnormal image, otherwise, determine that the type of place, described supplementary region image is normal picture.The numerical value of the first preset number threshold value can be set arbitrarily according to the demand of self by client, for example, by the first preset number threshold value setting, be 3, if the number in the supplementary region of statistics is greater than at 3 o'clock, the type of the image that comprises these 3 supplementary regions is abnormal image, if the number in the supplementary region of statistics is 1, the image that comprises this 1 supplementary region is normal picture.

If determine the type of place, described supplementary region image according to the size in supplementary region, its implementation comprises: the area that calculates described supplementary region; Whether the area that judgement is calculated is greater than preset area threshold value; If so, determine that the type of place, described supplementary region image is abnormal image, otherwise, determine that place, described supplementary region image is normal picture.The numerical value of above-mentioned preset area threshold value also can be set arbitrarily according to the demand of self by client.For example, by preset area threshold value setting, be 100 (pixels), calculate respectively the area in each supplementary region, if the area in supplementary region is greater than 100, this place, supplementary region image is abnormal image, otherwise, be normal picture.

Except directly utilizing the size of area, determine the type in supplementary region, can also be at the area that calculates each supplementary, obtain behind supplementary region that area is greater than preset area threshold value, further judge that this area is greater than the supplementary region of preset area threshold value and whether the area ratio of target image is greater than preset ratio, if, determine that supplementary region is abnormal image, otherwise, be normal picture.As, setting preset ratio is 10%, when the ratio that is greater than 100 the area in supplementary region and the area of target image when area is greater than 10%, this place, supplementary region image is abnormal image, otherwise, be normal picture.

Or, can also be according to the number in supplementary region and area, determine the type of place, supplementary region image, its implementation comprises: calculate the area in supplementary region, statistics area is greater than the number in the supplementary region of preset area threshold value, and whether the number of judgement statistics is greater than the first preset number threshold value, if, determine that place, supplementary region image is abnormal image, otherwise, be normal picture.For example, by preset area threshold value setting, be still 100, by the first preset number threshold value setting, be 3, through image recognition, supplementary region on certain image of determining is 5 (A, B, C, D and E), calculate the area in 5 supplementary regions, statistics area is greater than the number in 100 supplementary region, the area of supposing supplementary region A, B, C and D is greater than 100, the number of statistics is 4, the number of statistics is greater than 3, and supplementary region A, B, C, D and E place image are abnormal image.

If according to the type of place, supplementary region image described in the number in supplementary region and location positioning, its implementation comprises: add up the number in the described supplementary region of the appointed area that is positioned at target image, the fringe region of the nucleus that the appointed area of described target image is target image, the central area of target image or target image; Whether the number of judgement statistics is greater than the second preset number threshold value; If so, determine that the type of place, described supplementary region image is abnormal image, otherwise, determine that place, described supplementary region image is normal picture.

The size of the nucleus of above-mentioned target image, central area and fringe region can be set arbitrarily according to the demand of self by client, and the numerical value of the second preset number threshold value can be set arbitrarily according to the demand of self by client.For example, nucleus, central area and fringe region that the rectangle of take is divided target image are example, as the x by the rectangle upper left corner, the x in y coordinate and the lower right corner, when y coordinate is determined a rectangle, sets (0.3,0.3,0.7,0.7) be the nucleus of target image, set (0.15,0.15,0.85,0.85) be the central area of target image, the fringe region that remainder is target image.Statistics is positioned at the number in the supplementary region of nucleus, when the second preset number threshold value under setting nucleus is 1, if be positioned at the number in the supplementary region of nucleus, be greater than 1, place, the supplementary region image in this nucleus is abnormal image, otherwise, be normal picture.Or, statistics is positioned at the number in the supplementary region of central area, when the second preset number threshold value under setting central area is 2, if be positioned at the number in the supplementary region of central area, be greater than 2, place, the supplementary region image in this central area is abnormal image, otherwise, be normal picture.Or, statistics is positioned at the number in the supplementary region of fringe region, when the second preset number threshold value under setting fringe region is 3, if be positioned at the number in the supplementary region of fringe region, be greater than 3, place, the supplementary region image in this fringe region is abnormal image, otherwise, be normal picture.

If determine the type of place, described supplementary region image according to the content information in supplementary region, its implementation comprises: according to machine learning method, from described character area piece, identify word; The word that identifies of judgement whether with default dictionary in violated word match; If so, determine that the type of place, described supplementary region image is abnormal image, otherwise, determine that the type of place, described supplementary region image is normal picture.

Wherein, the embodiment of the present application adopts existing machine learning method, from described character area piece, identifies word, specifically comprises: by characteristics such as word height or spacing, character area piece is cut into the block that comprises one by one individual character; Block is carried out to feature description, as, the describing method based on marginal point sampling and the directional statistics that is connected thereof, or, the describing method based on HOG (Histograms of Oriented Gradients, gradient orientation histogram), or, the describing method based on stroke width direction etc.; By machine learning, according to the feature of block, block is identified as to word, as, the machine learning method SVM that google provides (Support Vector Machine, support vector machine) can be identified as word by block according to the feature of block.

When character area piece is identified as after word, in the default dictionary on backstage, store violated vocabulary, the word identifying is mated with the violated word in default dictionary, if there is violated word in character area piece, by this place, supplementary region spectral discrimination, it is abnormal image, otherwise, be normal picture.

Certainly, the embodiment of the present application can be according to the number in supplementary region, size, position and content, the comprehensive method of determining the type of place, described supplementary region image, that is to say, when all determining that by above-mentioned four kinds of modes place, supplementary region image is abnormal image respectively, this place, supplementary region image is finalized as abnormal image, otherwise this place, supplementary region image is finalized as normal picture.It should be noted that, when adopting a plurality of information (comprising any number of combinations in area information and content information) to determine the type of place, supplementary region image, the present invention does not limit sequencing each other simultaneously.Certainly, can also adopt a plurality of information to determine the type of place, supplementary region image simultaneously, and set: when any one or any number of definite result (comprise and all determine results) are during for abnormal image, this image is abnormal image.

As can be seen from the above-described embodiment, extract the character area piece in target image, character area piece is carried out to background expansion, can identify the supplementary region in target image.Supplementary region based on identifying, according to the number in supplementary region, size, position or content, further filters out abnormal image, thus the harm that can avoid abnormal image to cause consumer or supvr.

Embodiment tri-

The 5 groups of target images shown in Fig. 5 of take are below example, describe the filter method to target image in detail, to filter out abnormal image wherein.Refer to Fig. 6, it is the method flow diagram of a kind of image filtering method of the embodiment of the present application three announcements, comprises the following steps:

Step 601: extract the character area piece on target image;

Step 602: the character area piece of extraction is carried out to background expansion;

Step 603: the region after expansion is identified as to supplementary region;

Step 604: determine each region of supplementary region in target image;

Step 605: judge whether the area of supplementary region in specific region is greater than default area threshold, if so, determine that this place, supplementary region image is abnormal image, otherwise, enter step 606;

Step 606: judge whether the number of supplementary region in specific region is greater than default number threshold value, if so, determine that this place, supplementary region image is abnormal image, otherwise, enter step 607;

Step 607: the character area piece extracting is carried out to word identification;

Step 608: the word that identifies of judgement whether with violated dictionary in violated word match, if so, determine that this place, supplementary region image is abnormal image, otherwise, determine that this place, supplementary region image is normal picture.

As shown in Figure 5, first classifies the former figure of target image as, and second classifies the supplementary region (comprising character area piece and background area) of identifying in target image as, and the 3rd classifies the violated word identifying from character area piece as, and the 4th classifies filter result as.

As can be seen from the above-described embodiment, extract the character area piece in target image, character area piece is carried out to background expansion, can identify the supplementary region in target image.Supplementary region based on identifying, according to the number in supplementary region, size, position or content, further filters out abnormal image, thus the harm that can avoid abnormal image to cause consumer or supvr.

Embodiment tetra-

Corresponding with a kind of image-recognizing method in above-described embodiment one, the embodiment of the present application also provides a kind of pattern recognition device.Refer to Fig. 7, it is the structure drawing of device of a kind of pattern recognition device of the embodiment of the present application four announcements, and this device comprises: character area piece extraction module 701, strong fringe region extraction module 702, background area judge module 703 and picture recognition module 704.Principle of work below in conjunction with this device is further introduced its inner structure and annexation.

Character area piece extraction module 701, for extracting the character area piece on target image;

Strong fringe region extraction module 702, for adopting image border algorithm, extracts the strong fringe region on target image;

Background area judge module 703, for judging whether described strong fringe region is the background area of character area piece;

Picture recognition module 704, for when judgment result is that of described background area judge module is, described character area piece expansion background area, the region after expansion is identified as to supplementary region, otherwise, described character area piece is identified as to supplementary region.

Preferably, background area judge module 703 comprises: the first judgement submodule, for judging according to the main color rate of described strong fringe region whether described strong fringe region is pure color region; The first result is determined submodule, and for when judgment result is that of described the first judgement submodule is, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

Or, preferably you, background area judge module 703 comprises: second judgement submodule, for having judged whether that character area piece is positioned at described strong fringe region; The second result is determined submodule, and for when judgment result is that of described the second judgement submodule is, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

Again or, preferred, background area judge module 703 comprises: the 3rd judgement submodule, whether identical for judging the main color of described strong fringe region and character area piece; The 3rd result is determined submodule, and for when judgment result is that of described the 3rd judgement submodule is, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

Again or, preferred, background area judge module 703 comprises: apart from calculating sub module, for calculating the distance between described strong fringe region and the main color of character area piece by color distance formula, wherein, color distance formula is: $\mathrm{\ΔC}=\sqrt{(2+\stackrel{\‾}{r}/256)\×{\mathrm{\Δr}}^2+4\×{\mathrm{\Δg}}^2+(2+(255-\stackrel{\‾}{r})/256)\×{\mathrm{\Δb}}^2},$ Wherein,

Δ r=r ₁-r ₂, Δ g=g ₁-g ₂, Δ b=b ₁-b ₂, the main color of strong fringe region is: (r1, g1, b1), and the main color of character area piece is: (r2, g2, b2); The 4th judgement submodule, for judging whether the distance between described strong fringe region and the main color of character area piece is less than predeterminable range threshold value; The 4th result is determined submodule, and for when judgment result is that of described the 4th judgement submodule is, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

As can be seen from the above-described embodiment, extract the character area piece in target image, character area piece is carried out to background expansion, can identify the supplementary region in target image.

Embodiment five

Corresponding with a kind of image filtering method in above-described embodiment two, the embodiment of the present application also provides a kind of image filtering device.Refer to Fig. 8, it is the structure drawing of device of a kind of pattern recognition device of the embodiment of the present application four announcements, and this device comprises: character area piece extraction module 701, strong fringe region extraction module 702, background area judge module 703, picture recognition module 704 and image filtering module 705.Principle of work below in conjunction with this device is further introduced its inner structure and annexation.

Character area piece extraction module 701, for extracting the character area piece on target image;

Strong fringe region extraction module 702, for adopting image border algorithm, extracts the strong fringe region on target image;

Background area judge module 703, for judging whether described strong fringe region is the background area of character area piece;

Picture recognition module 704, for when judgment result is that of described background area judge module is, described character area piece expansion background area, the region after expansion is identified as to supplementary region, otherwise, described character area piece is identified as to supplementary region;

Image filtering module 705, for according to any one or any number of combination of the area information in described supplementary region and content information, determines the type of place, described supplementary region image.

Described type comprises abnormal image and normal picture.

Wherein, preferred, image filtering module 705 comprises: number statistics submodule, for adding up the number in described supplementary region; Number judgement submodule, for judging whether the number of statistics is greater than the first preset number threshold value; The first kind is determined submodule, for when judgment result is that of described number judgement submodule is, determines that the type of place, described supplementary region image is abnormal image, otherwise, determine that the type of place, described supplementary region image is normal picture.

Or preferred, image filtering module 705 comprises: area statistics submodule, for calculating the area in described supplementary region; Area judgement submodule, for judging whether the area of calculating is greater than preset area threshold value; Second Type is determined submodule, for when judgment result is that of described area judgement submodule is, determines that the type of place, described supplementary region image is abnormal image, otherwise, determine that the type of place, described supplementary region image is normal picture.

Again or, preferably, described image filtering module 705 comprises: integration statistics submodule, for adding up the number in the described supplementary region of the appointed area that is positioned at target image, the fringe region of the nucleus that the appointed area of described target image is icon image, the zone line of target image or target image; Integration judgement submodule, for judging whether the number of statistics is greater than the second preset number threshold value; The 3rd type is determined submodule, for when judgment result is that of described comprehensive judgement submodule is, determines that the type of place, described supplementary region image is abnormal image, otherwise, determine that place, described supplementary region image is normal picture.

Again or, preferred, described image filtering module 705 comprises: word recognin module, for according to machine learning method, from described character area piece, identify word; Characters matching judgement submodule, for judging whether the word identifying matches with the violated word of default dictionary; The 4th type is determined submodule, for when judgment result is that of described characters matching judgement submodule is, the type of determining place, described supplementary region image is abnormal image, otherwise, determine that the type of place, described supplementary region image is normal picture.

The structure of background area judge module 703 wherein can be referring to the description in embodiment tetra-, and the present embodiment repeats no more.

As can be seen from the above-described embodiment, extract the character area piece in target image, character area piece is carried out to background expansion, can identify the supplementary region in target image.Supplementary region based on identifying, according to the number in supplementary region, size, position or content, further filters out abnormal image, thus the harm that can avoid abnormal image to cause consumer or supvr.

It should be noted that, one of ordinary skill in the art will appreciate that all or part of flow process realizing in above-described embodiment method, to come the hardware that instruction is relevant to complete by computer program, described program can be stored in a computer read/write memory medium, this program, when carrying out, can comprise as the flow process of the embodiment of above-mentioned each side method.Wherein, described storage medium can be magnetic disc, CD, read-only store-memory body (Read-Only Memory, ROM) or random store-memory body (Random Access Memory, RAM) etc.

A kind of image-recognizing method, filter method and the relevant apparatus that above the application are provided are described in detail, applied specific embodiment herein the application's principle and embodiment are set forth, the explanation of above embodiment is just for helping to understand the application's method and core concept thereof; Meanwhile, for one of ordinary skill in the art, the thought according to the application, all will change in specific embodiments and applications, and in sum, this description should not be construed as the restriction to the application.

Claims (31)

1. an image-recognizing method, is characterized in that, comprising:

Extract the character area piece on target image;

Adopt image border algorithm, extract the strong fringe region on target image;

Judge whether described strong fringe region is the background area of described character area piece;

If so, described character area piece expansion background area, the region after expansion is identified as to supplementary region, otherwise, described character area piece is identified as to supplementary region.

2. method according to claim 1, is characterized in that, describedly judges that whether described strong fringe region is that the background area of character area piece comprises:

According to the main color rate of described strong fringe region, judge whether described strong fringe region is pure color region;

If so, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

3. method according to claim 1, is characterized in that, describedly judges that whether described strong fringe region is that the background area of character area piece comprises:

Judged whether that character area piece is positioned on described strong fringe region;

If so, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

4. method according to claim 1, is characterized in that, describedly judges that whether described strong fringe region is that the background area of character area piece comprises:

Whether the main color that judges described strong fringe region and character area piece is identical;

If so, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

5. method according to claim 1, is characterized in that, describedly judges that whether described strong fringe region is that the background area of character area piece comprises:

By color distance formula, calculate the distance between described strong fringe region and the main color of character area piece, wherein, color distance formula is: $\mathrm{\ΔC}=\sqrt{(2+\stackrel{\‾}{r}/256)\×{\mathrm{\Δr}}^2+4\×{\mathrm{\Δg}}^2+(2+(255-\stackrel{\‾}{r})/256)\×{\mathrm{\Δb}}^2},$ Wherein,

Δ r=r ₁-r ₂, Δ g=g ₁-g ₂, Δ b=b ₁-b ₂; The main color of strong fringe region is: (r1, g1, b1), and the main color of character area piece is: (r2, g2, b2);

Judge whether the distance between described strong fringe region and the main color of character area piece is less than predeterminable range threshold value;

If so, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

6. according to the method described in any one in claim 1-5, it is characterized in that, describedly from target image, extract character area piece and comprise:

Based on texture statistics method or based on region analysis method, from target image, extract character area piece.

7. an image filtering method, is characterized in that, comprising:

Extract the character area piece on target image;

Adopt image border algorithm, extract the strong fringe region on target image;

Judge whether described strong fringe region is the background area of character area piece;

If so, described character area piece expansion background area, the region after expansion is identified as to supplementary region, otherwise, described character area piece is identified as to supplementary region;

According to the area information in described supplementary region and any one information in content information or any number of information combination, determine the type of place, described supplementary region image.

8. method according to claim 7, is characterized in that, the area information in described supplementary region comprises: region number, size and the position in supplementary region.

9. method according to claim 7, is characterized in that, the described number according to described supplementary region determines that the type of place, described supplementary region image comprises:

Add up the number in described supplementary region;

Whether the number of judgement statistics is greater than the first preset number threshold value;

If so, determine that the type of place, described supplementary region image is abnormal image, otherwise, determine that the type of place, described supplementary region image is normal picture.

10. method according to claim 7, is characterized in that, the described size according to described supplementary region determines that the type of place, described supplementary region image comprises:

Calculate the area in described supplementary region;

Whether the area that judgement is calculated is greater than preset area threshold value;

If so, determine that the type of place, described supplementary region image is abnormal image, otherwise, determine that the type of place, described supplementary region image is normal picture.

11. methods according to claim 7, is characterized in that, the described type according to place, supplementary region image described in the number in described supplementary region and location positioning comprises:

Add up the number in the described supplementary region of the appointed area that is positioned at target image, the fringe region of the nucleus that the appointed area of described target image is target image, the zone line of target image or target image;

Whether the number of judgement statistics is greater than the second preset number threshold value;

If so, determine that the type of place, described supplementary region image is abnormal image, otherwise, determine that the type of place, described supplementary region image is normal picture.

12. methods according to claim 7, is characterized in that, the described content according to described supplementary region determines that the type of place, described supplementary region image comprises:

According to machine learning method, from described character area piece, identify word;

The word that identifies of judgement whether with default dictionary in violated word match;

If so, determine that the type of place, described supplementary region image is abnormal image, otherwise, determine that the type of place, described supplementary region image is normal picture.

13. methods according to claim 7, is characterized in that, describedly judge that whether described strong fringe region is that the background area of character area piece comprises:

According to the main color rate of described strong fringe region, judge whether described strong fringe region is pure color region;

If so, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

14. methods according to claim 7, is characterized in that, describedly judge that whether described strong fringe region is that the background area of character area piece comprises:

Judged whether that character area piece is positioned on described strong fringe region;

If so, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

15. methods according to claim 7, is characterized in that, describedly judge that whether described strong fringe region is that the background area of character area piece comprises:

Whether the main color that judges described strong fringe region and character area piece is identical;

If so, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

16. methods according to claim 7, is characterized in that, describedly judge that whether described strong fringe region is that the background area of character area piece comprises:

By color distance formula, calculate the distance between described strong fringe region and the main color of character area piece, wherein, color distance formula is: $\mathrm{\ΔC}=\sqrt{(2+\stackrel{\‾}{r}/256)\×{\mathrm{\Δr}}^2+4\×{\mathrm{\Δg}}^2+(2+(255-\stackrel{\‾}{r})/256)\×{\mathrm{\Δb}}^2},$ Wherein,

Δ r=r ₁-r ₂, Δ g=g ₁-g ₂, Δ b=b ₁-b ₂, the main color of strong fringe region is: (r1, g1, b1), and the main color of character area piece is: (r2, g2, b2);

Judge whether the distance between described strong fringe region and the main color of character area piece is less than predeterminable range threshold value;

If so, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

17. 1 kinds of pattern recognition devices, is characterized in that, comprising:

Character area piece extraction module, for extracting the character area piece on target image;

Strong fringe region extraction module, for adopting image border algorithm, extracts the strong fringe region on target image;

Background area judge module, for judging whether described strong fringe region is the background area of character area piece;

Picture recognition module, for when judgment result is that of described background area judge module is, described character area piece expansion background area, is identified as supplementary region by the region after expansion, otherwise, described character area piece is identified as to supplementary region.

18. devices according to claim 17, is characterized in that, described background area judge module comprises:

The first judgement submodule, for judging according to the main color rate of described strong fringe region whether described strong fringe region is pure color region;

The first result is determined submodule, and for when judgment result is that of described the first judgement submodule is, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

19. devices according to claim 17, is characterized in that, described background area judge module comprises:

The second judgement submodule, for having judged whether that character area piece is positioned at described strong fringe region;

The second result is determined submodule, and for when judgment result is that of described the second judgement submodule is, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

20. devices according to claim 17, is characterized in that, described background area judge module comprises:

Whether the 3rd judgement submodule is identical for judging the main color of described strong fringe region and character area piece;

The 3rd result is determined submodule, and for when judgment result is that of described the 3rd judgement submodule is, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

21. devices according to claim 17, is characterized in that, described background area judge module comprises:

Apart from calculating sub module, for calculating the distance between described strong fringe region and the main color of character area piece by color distance formula, wherein, color distance formula is: $\mathrm{\ΔC}=\sqrt{(2+\stackrel{\‾}{r}/256)\×{\mathrm{\Δr}}^2+4\×{\mathrm{\Δg}}^2+(2+(255-\stackrel{\‾}{r})/256)\×{\mathrm{\Δb}}^2},$ Wherein,

Δ r=r ₁-r ₂, Δ g=g ₁-g ₂, Δ b=b ₁-b ₂, the main color of strong fringe region is: (r1, g1, b1), and the main color of character area piece is: (r2, g2, b2);

The 4th judgement submodule, for judging whether the distance between described strong fringe region and the main color of character area piece is less than predeterminable range threshold value;

The 4th result is determined submodule, and for when judgment result is that of described the 4th judgement submodule is, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

22. 1 kinds of image filtering devices, is characterized in that, comprising:

Character area piece extraction module, for extracting the character area piece on target image;

Strong fringe region extraction module, for adopting image border algorithm, extracts the strong fringe region on target image;

Background area judge module, for judging whether described strong fringe region is the background area of character area piece;

Picture recognition module, for when judgment result is that of described background area judge module is, described character area piece expansion background area, is identified as supplementary region by the region after expansion, otherwise, described character area piece is identified as to supplementary region;

Image filtering module, for according to any one information of the area information in described supplementary region and content information or any number of information combination, determines the type of place, described supplementary region image.

23. devices according to claim 22, is characterized in that, the area information in described supplementary region comprises: region number, size and the position in supplementary region.

24. devices according to claim 22, is characterized in that, described image filtering module comprises:

Number statistics submodule, for adding up the number in described supplementary region;

Number judgement submodule, for judging whether the number of statistics is greater than the first preset number threshold value;

The first kind is determined submodule, for when judgment result is that of described number judgement submodule is, determines that the type of place, described supplementary region image is abnormal image, otherwise, determine that the type of place, described supplementary region image is normal picture.

25. devices according to claim 22, is characterized in that, described image filtering module comprises:

Area statistics submodule, for calculating the area in described supplementary region;

Area judgement submodule, for judging whether the area of calculating is greater than preset area threshold value;

Second Type is determined submodule, for when judgment result is that of described area judgement submodule is, determines that the type of place, described supplementary region image is abnormal image, otherwise, determine that the type of place, described supplementary region image is normal picture.

26. devices according to claim 22, is characterized in that, described image filtering module comprises:

Integration statistics submodule, for adding up the number in the described supplementary region of the appointed area that is positioned at target image, the fringe region of the nucleus that the appointed area of described target image is icon image, the zone line of target image or target image;

Integration judgement submodule, for judging whether the number of statistics is greater than the second preset number threshold value;

The 3rd type is determined submodule, for when judgment result is that of described comprehensive judgement submodule is, determines that the type of place, described supplementary region image is abnormal image, otherwise, determine that the type of place, described supplementary region image is normal picture.

27. devices according to claim 22, is characterized in that, described image filtering module comprises:

Word recognin module for according to machine learning method, identifies word from described character area piece;

Characters matching judgement submodule, for judging whether the word identifying matches with the violated word of default dictionary;

The 4th type is determined submodule, for when judgment result is that of described characters matching judgement submodule is, the type of determining place, described supplementary region image is abnormal image, otherwise, determine that the type of place, described supplementary region image is normal picture.

28. devices according to claim 22, is characterized in that, described background area judge module comprises:

The first judgement submodule, for judging according to the main color rate of described strong fringe region whether described strong fringe region is pure color region;

The first result is determined submodule, and for when judgment result is that of described the first judgement submodule is, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

29. devices according to claim 22, is characterized in that, described background area judge module comprises:

The second judgement submodule, for having judged whether that character area piece is positioned at described strong fringe region;

The second result is determined submodule, and for when judgment result is that of described the second judgement submodule is, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

30. devices according to claim 22, is characterized in that, described background area judge module comprises:

Whether the 3rd judgement submodule is identical for judging the main color of described strong fringe region and character area piece;

The 3rd result is determined submodule, and for when judgment result is that of described the 3rd judgement submodule is, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

31. devices according to claim 22, is characterized in that, described background area judge module comprises:

Apart from calculating sub module, for calculating the distance between described strong fringe region and the main color of character area piece by color distance formula, wherein, color distance formula is: $\mathrm{\ΔC}=\sqrt{(2+\stackrel{\‾}{r}/256)\×{\mathrm{\Δr}}^2+4\×{\mathrm{\Δg}}^2+(2+(255-\stackrel{\‾}{r})/256)\×{\mathrm{\Δb}}^2},$ Wherein,

Δ r=r ₁-r ₂, Δ g=g ₁-g ₂, Δ b=b ₁-b ₂, the main color of strong fringe region is: (r1, g1, b1), and the main color of character area piece is: (r2, g2, b2);

The 4th judgement submodule, for judging whether the distance between described strong fringe region and the main color of character area piece is less than predeterminable range threshold value;

The 4th result is determined submodule, and for when judgment result is that of described the 4th judgement submodule is, described strong fringe region is the background area of character area piece, otherwise described strong fringe region is not the background area of character area piece.

Images