CN102799653A - Logo detection method based on spatial connected domain prepositioning - Google Patents
Logo detection method based on spatial connected domain prepositioning Download PDFInfo
- Publication number
- CN102799653A CN102799653A CN2012102259223A CN201210225922A CN102799653A CN 102799653 A CN102799653 A CN 102799653A CN 2012102259223 A CN2012102259223 A CN 2012102259223A CN 201210225922 A CN201210225922 A CN 201210225922A CN 102799653 A CN102799653 A CN 102799653A
- Authority
- CN
- China
- Prior art keywords
- trade mark
- connected domain
- picture
- color
- anchor point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 41
- 238000012360 testing method Methods 0.000 claims abstract description 61
- 238000000034 method Methods 0.000 claims description 51
- 238000004891 communication Methods 0.000 claims description 15
- 230000008878 coupling Effects 0.000 claims description 7
- 238000010168 coupling process Methods 0.000 claims description 7
- 238000005859 coupling reaction Methods 0.000 claims description 7
- 239000003086 colorant Substances 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 238000011002 quantification Methods 0.000 claims description 5
- 230000031068 symbiosis, encompassing mutualism through parasitism Effects 0.000 claims description 5
- 230000013011 mating Effects 0.000 claims description 4
- 239000004744 fabric Substances 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 241000675108 Citrus tangerina Species 0.000 claims description 2
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 238000009499 grossing Methods 0.000 claims description 2
- 239000000843 powder Substances 0.000 claims description 2
- 101000608653 Homo sapiens UbiA prenyltransferase domain-containing protein 1 Proteins 0.000 claims 4
- 201000004224 Schnyder corneal dystrophy Diseases 0.000 claims 4
- 102100039547 UbiA prenyltransferase domain-containing protein 1 Human genes 0.000 claims 4
- 238000005286 illumination Methods 0.000 abstract description 5
- 230000009466 transformation Effects 0.000 abstract description 3
- 230000000694 effects Effects 0.000 abstract description 2
- 230000000007 visual effect Effects 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 9
- 230000008569 process Effects 0.000 description 6
- 230000008859 change Effects 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 235000021444 pepsi cola Nutrition 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 241000282341 Mustela putorius furo Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a logo detection method based on spatial connected domain prepositioning. The logo detection method disclosed by the invention comprises the following steps: creating a logo picture sample library comprising multiple logo pictures; creating spatial connected domain descriptor (SCCD) characteristics for the logo pictures in the logo picture sample library; for a test picture containing a target logo, creating connected domain descriptor (CCD) characteristics of the test picture; acquiring logo prepositioning regions (LPRs) in the test picture by utilizing the SCCD characteristics of the logo pictures in the logo picture sample library; and matching the LPRs with the logo pictures in the logo picture data library by utilizing the characteristics which are created based on logo colour and shape information, wherein the successfully matched logo prepositioning region is namely a finally detected logo. The logo detection method disclosed by the invention fully directs at the characteristics of a logo, speed and accuracy of logo detection and positioning identification are further improved, and detection on the logo, which is influenced by scale variation, visual angle transformation, illumination, overlap and background interference, can have good effect.
Description
Technical field
The present invention relates to target detection and distinguishment technical field, especially a kind of trade mark detection method based on spatial communication territory pre-determined bit can be used for quick trade mark detection, trade mark retrieval, identification and monitoring.
Background technology
Trade mark detects and identification is one of challenging task in target detection and the identification field.How influence down, detect accurately and locate that to identify trade mark be a challenging task receive change of scale, view transformation, illumination, block, background interference etc.At present, various information mediums have obtained swift and violent development, such as TV, and broadcasting, network etc.All be flooded with a large amount of advertising messages every day in these information mediums, adds present trade mark One's name is legion, how these advertising messages effectively managed and monitored, and further paid attention to the safety that guarantees consumer and businessman.Based on the quick trade mark detection method of spatial communication territory pre-determined bit in order to satisfy advertisement monitoring requirement in information security field picture and the video.
Most of target detection and the mode ferret out of discerning the search of employing multi-scale sliding window mouth; Traditional trade mark detects this method that also adopts; This method is simple traversal search; Therefore bring a large amount of useless calculating, and along with the increase of picture dimension is elongated detection time, these all quite are unfavorable for the application that real-time detects.Some prior art also adopts the method for target training, but possibly have type of pictures up to ten thousand for trade mark is this, if each picture is all trained, needs a large amount of pictures, and this is unfavorable for actual application very much.
Summary of the invention
In order to reach the purpose that quick and precisely detects, the present invention is different from above prior art, and the characteristics of the sample body of going into business are set out, and have proposed a kind of trade mark detection method based on spatial communication territory pre-determined bit.
A kind of trade mark detection method based on spatial communication territory pre-determined bit that the present invention proposes is characterized in that this method may further comprise the steps:
Step 3, input contains the test picture of target trade mark, for said test picture, sets up its CCD characteristic;
Step 4 utilizes the SCCD characteristic of the trade mark picture that said step 2 obtains in the test picture, to obtain trade mark pre-determined bit area L PRs;
Step 5; The trade mark coupling is carried out with the trade mark picture in the said trade mark picture database in the trade mark pre-determined bit zone that the characteristic that utilization is set up based on trade mark color and shape information obtains said step 4, mate successful trade mark pre-determined bit zone and is in the test picture trade mark that finally detection obtains;
Wherein, said step 2 further comprises following step:
Step 2.4 utilizes the color of trade mark to have certain zonal characteristic, obtains the effective connected domain of the conduct of the color connected domain except background colour in the trade mark picture;
Step 2.6 obtains all effective connected domains according to said step 2.4, and the trade mark picture is set up the SCCD characteristic, and said SCCD characteristic comprises connected domain forecast model and effective these two characteristics of connected domain pixel distribution histogram;
The step of setting up the CCD characteristic of test picture in the said step 3 further may further comprise the steps:
Step 3.2 utilizes the color of trade mark to have certain zonal characteristic, obtains the main connected domain of color connected domain conduct except background colour in the test picture;
The connected domain that step 3.3, the main connected domain that said step 3.2 is obtained are carried out gray space is cut apart, and obtains testing effective connected domain of picture;
Step 3.4 is calculated the CCD characteristic of testing each effective connected domain of picture according to said step 2.6;
Said step 4 further is divided into two steps:
Step 4.1 is based on the trade mark pre-determined bit area L PRs in the forecast model search test picture;
Step 4.2, the LPRs that screening searches based on the pixel distribution histogram.
Color, shape and provincial characteristics that the inventive method makes full use of trade mark detect and locate identification to trade mark; Further improved trade mark and detected speed and precision with location identification, simultaneously target receive under change of scale, the stable condition view transformation, illumination, block, detection that background interference etc. influences down trade mark has good effect.
Description of drawings
Fig. 1 is the process flow diagram that the present invention is based on the trade mark detection method of spatial communication territory pre-determined bit.
Fig. 2 is some the trade mark picture examples in the trade mark picture sample storehouse.
Fig. 3 is the process flow diagram that obtains spatial communication territory descriptor (SCCD) characteristic.
Fig. 4 is a synoptic diagram of setting up the SCCD characteristic.
Fig. 5 is the synoptic diagram that obtains the LPR of Pepsi Cola.
Fig. 6 is the piecemeal synoptic diagram of piecemeal color histogram.
Fig. 7 utilizes the curve map of picture to be measured detection time of the inventive method with the picture area change.
Fig. 8 is the precision-recall rate correlation curve figure of the inventive method and classic method.
Fig. 9 utilizes the test result synoptic diagram of the present invention in picture database to be measured.
Embodiment
For making the object of the invention, technical scheme and advantage clearer, below in conjunction with specific embodiment, and with reference to accompanying drawing, to further explain of the present invention.
The present invention proposes a kind of trade mark detection method, fast detecting and location identification that this method combines trade mark connected domain and spatial relationship characteristic thereof to carry out the trade mark pre-determined bit based on spatial communication territory pre-determined bit.Fig. 1 is the process flow diagram that the present invention is based on the quick trade mark detection method of spatial communication territory pre-determined bit, and is as shown in Figure 1, and the quick trade mark detection method that the present invention is based on spatial communication territory pre-determined bit may further comprise the steps:
Detect owing to the objective of the invention is trade mark, promptly detect and whether contain the target trade mark in the picture to be measured, so the quantity of target trade mark picture is decided according to the actual detected demand in the trade mark picture sample storehouse.
Fig. 2 has listed the part trade mark picture in the trade mark picture sample of the present invention storehouse, from these trade mark pictures, can find out, the background of trade mark all is a white, and the main body color of trade mark does not receive the influence of extraneous factor, such as illumination.The sharpness of border of trade mark splits from background easily simultaneously.
About about 100 * 100, most like this trade mark details can remain the size of the trade mark picture in the trade mark picture sample storehouse clearly greatly.In addition, the trade mark picture in the trade mark picture sample storehouse should be clear as much as possible as the basis of brand recognition, could guarantee the therefrom accuracy of the characteristic of extraction like this.
Fig. 3 is the process flow diagram that obtains spatial communication territory descriptor (SCCD) characteristic, and is as shown in Figure 3, and said step 2 further comprises following step:
Step 2.1 is transformed into tone saturation degree brightness HSV space with the trade mark picture from the RGB rgb space;
The current techique that converts this area in HSV space repeats no more at this.
Step 2.2, to the trade mark picture that is transformed into the HSV space carry out 8 kinds of colors quantification (such as can for: the red greenish-yellow blue tangerine powder of black and white), obtain through the trade mark picture behind 8 kinds of color quantizings;
Hsv color quantizing process in the list of references of the present invention " Content-based retrieval of logo and trademarks in unconstrained color image databases using Color Edge Gradient Co-occurrence Histograms (R.Phan and D.Androutsos; Computer Vision and Image Understanding (114); pp.66-84; 2010) " carries out the quantification of 8 kinds of colors to the trade mark picture that is transformed into the HSV space; But what the present invention was different with method in the document is; The present invention has given up the grey of using in the document, promptly only the achromaticity amount of pixels is turned to black and white, and other processing procedures all processing procedure with said document are consistent.
Step 2.3 is judged background colour and noise color through the trade mark picture after quantizing, and the trade mark picture is carried out the color smoothing processing, to reduce the influence of noise color;
Said step 2.3 further comprises following step:
Step 2.3.1; Get on four angles of trade mark picture after the quantification and be of a size of 5 * 5 color block; With number of pixels is maximum in each color block color main color as this color block; Occurrence number in the main color of four color block greater than 2 the color background colour as this trade mark picture, is defaulted as one of background colour with white simultaneously;
Step 2.3.2, number of pixels is less than 20 color and is judged to be the noise color in the trade mark picture after will passing through hsv color and quantizing;
Step 2.3.3 to certain noise colored pixels, utilizes around it in 8 pixels that maximum non-noise color of occurrence number that the color of this noise colored pixels is carried out assignment, thereby realizes level and smooth to the color of trade mark picture.
Step 2.4 utilizes the color of trade mark to have certain zonal characteristic, obtains the effective connected domain of the conduct of the color connected domain except background colour in the trade mark picture;
In the trade mark picture shown in Fig. 4 (a), except background white, also have red and blue two kinds of colors, therefore form 1 and 2 two color connected domain, these color connected domains are thought effective connected domain.
Step 2.5 sorts all effective connected domains from big to small according to area and numbers, with the first five effective connected domain of rank as the anchor point connected domain;
If effectively the area of connected domain is identical, then at random it is numbered.
Shown in Fig. 4 (a) and (b), the effective connected domain in the trade mark picture is numbered 1 and 2 from big to small according to area, and shown in Fig. 4 (c), (d), the effective connected domain in the trade mark picture is numbered 1,2,3 and 4 from big to small according to area.
Then with the first five effective connected domain of rank as the anchor point connected domain, be used for the trade mark location of follow-up test picture, if effective 5 of the lazy weights of connected domain, then whole effective connected domains is all as the anchor point connected domain.
Step 2.6 obtains all effective connected domains according to said step 2.4, and the trade mark picture is set up the SCCD characteristic;
Said SCCD characteristic comprises connected domain forecast model and effective these two characteristics of connected domain pixel distribution histogram.The former utilizes each connected domain color and spatial relation thereof to represent the syntagmatic between the connected domain, and the latter has then represented the pixel distribution situation of effective connected domain in the trade mark.Both are described the layout of trade mark respectively from different perspectives.
Said connected domain forecast model characteristic is divided into two parts, is respectively anchor point connected domain assemblage characteristic and effective connected domain space collection characteristic.
First, anchor point connected domain combined feature is described with color-anchor point connected domain combination of sets, the permutation and combination of color anchorage connected domain promptly of the same race, for example, the color of the trade mark picture shown in Fig. 4 (c)-anchor point connected domain combination of sets can be described as:
Red--[(1), (2), (3), (4), (12), (13), (14), (23), (24), (34), (123), (124), (134), (234)] }.
For each color in the said combination of sets-anchor point connected domain combination, write down its color, the knowledge of boundary rectangle collimation mark, the anchor point connected domain combination total area, frontier point-information such as centroid distance angular histogram.
Wherein, The boundary rectangle frame of a certain anchor point connected domain is the minimum rectangle frame that this anchor point connected domain is all closely included; The sign of the boundary rectangle frame of a certain anchor point connected domain is corresponding with the numbering of anchor point connected domain; Sign such as the boundary rectangle frame of anchor point connected domain 1 can be taken as R1, and is as shown in Figure 4.
In frontier point-centroid distance angular histogram; Frontier point is the frontier point summation of all anchor point connected domains in each anchor point connected domain combination; Barycenter makes up the central point of pairing boundary rectangle frame for this anchor point connected domain; Distance be each frontier point with the central point of corresponding boundary rectangle frame between Euclidean distance, angle is each frontier point and the angle of the relative horizontal direction of central point of boundary rectangle frame accordingly.The dimension of said frontier point-centroid distance angular histogram can be confirmed according to actual needs; Such as; In order to satisfy the demand of certain fineness, the present invention sets up frontier point-centroid distance angular histogram that a dimension is 10 * 12 (dimension that is distance is 10, and the dimension of angle is 12); Obtain said frontier point-centroid distance angular histogram for ease; Can the longest Euclidean distance of being tried to achieve be divided into 10 five equilibriums,, obtain the zone that this Euclidean distance belongs on this histogram middle distance direction the length of all Euclidean distances that calculate after divided by 10 branches such as grade; With all angles that calculate divided by 30 °; Obtain the zone that this angle belongs on the angle direction in this histogram; Final through adding up the pixel that is comprised in each zone, just can obtain dimension and be frontier point-centroid distance angular histogram of 10 * 12.When trade mark at the area of test in the picture hour, effective connected domain of color of the same race maybe be bonding, and the combination of anchor point connected domain just can be handled this situation;
Second portion, effectively the characteristic of connected domain space collection concerns with boundary rectangle frame-anchor point connected domain and describes, i.e. the relation of the boundary rectangle frame of anchor point connected domain and the anchor point connected domain that wherein comprised.When the boundary rectangle frame of certain anchor point connected domain is comprised by the boundary rectangle frame of another anchor point connected domain fully; This anchor point connected domain counts in the big pairing connected domain of boundary rectangle frame; For example the boundary rectangle frame R2 of the anchor point connected domain 2 among Fig. 4 (b) is comprised by the boundary rectangle frame R1 of anchor point connected domain 1 fully, and the boundary rectangle frame of the trade mark picture shown in Fig. 4 (b)-anchor point connected domain relation can be described as so: { R1-(12) }; For another example, the boundary rectangle frame of the trade mark picture shown in Fig. 4 (c)-anchor point connected domain relation can be described as: { R1-(14), R2-(2), R3-(3) }.
Effectively connected domain pixel distribution histogram is used to screen trade mark pre-determined bit zone (LPRs), and the histogrammic foundation of said effective connected domain pixel distribution specifically comprises following step:
Step 2.6.1 shown in Fig. 4 (d), evenly is divided into 4 piecemeals with the trade mark picture;
Step 2.6.2; Each piecemeal all is divided into 8 parts (bin) in the horizontal and vertical directions; Its horizontal ordinate position of pixel basis in the effective connected domain of trade mark picture is fallen among the corresponding bin; Form the pixel distribution histogram of two 8 dimensions, successively each piece is carried out aforesaid operations, finally obtain the pixel distribution histogram of 88 dimensions of total of 4 piecemeals;
Step 2.6.3; Utilize the area of each piecemeal that its corresponding pixel distribution histogram is carried out normalization; All obtain the normalization pixel distribution histogram of 88 dimensions like this for every width of cloth trade mark picture; Effective connected domain pixel distribution histogram of then every width of cloth trade mark picture is that (8 * 8=64), this has just described the space distribution information of trade mark to 64 dimensions to a certain extent.
Step 3, input contains the test picture of target trade mark, for said test picture, sets up its connected domain descriptor (CCD) characteristic;
The step of setting up connected domain descriptor (CCD) characteristic of test picture further may further comprise the steps:
Step 3.1 is carried out HSV space conversion, color quantizing and color level and smooth (of step 2.1-step 2.3) with the test picture according to the processing mode the same with trade mark picture sample;
Step 3.2 according to said step 2.4, utilizes the color of trade mark to have certain zonal characteristic, obtains the main connected domain of color connected domain conduct except background colour in the test picture;
The connected domain that step 3.3, the main connected domain that said step 3.2 is obtained are carried out gray space is cut apart, and obtains testing effective connected domain of picture;
Said step 3.3 further may further comprise the steps:
Step 3.3.1 will test picture and transform to gray space;
In this step, gray space convert current techique into, repeat no more at this.
Step 3.3.2; Said main connected domain is carried out the detection of the connected domain area of gray space; If the area of a main connected domain greater than certain threshold value (such as 100) and its gray-scale value variance greater than certain value (such as 50); Then utilize big Tianjin method that this main connected domain is carried out gray space and cut apart, two sub-connected domains after obtaining cutting apart;
Said big Tianjin method is divided into the current techique of this area; Concrete grammar is described and please refer to document " A threshold selection method from gray level histograms, " N.Otsu; IEEE Transactions onSystems Man and Cybernetics (9); Pp.62-66,1979, repeat no more here.
Step 3.3.3 is if the average gray value difference of said two sub-connected domains greater than certain threshold value (such as 50), then covers that the main connected domain before cutting apart with these two that obtain after cutting apart sub-connected domains;
Step 3.3.4 is cut apart all main connected domains according to said step 3.3.2 and step 3.3.3, ends up to not being further divided into, and removes the connected domain of area too little (such as less than 30) then, the remaining effective connected domain that is the test picture;
Step 3.4 is calculated the CCD characteristic of testing each effective connected domain of picture according to said step 2.6, comprises color, area, boundary rectangle frame and frontier point-centroid distance angular histogram (the similar step 2.5 of process).
Step 4 utilizes the SCCD characteristic of the trade mark picture that said step 2 obtains in the test picture, to obtain trade mark pre-determined bit area L PR;
Said step 4 further is divided into two steps:
Step 4.1 is based on the trade mark pre-determined bit area L PRs in the forecast model search test picture;
Fig. 5 has provided the synoptic diagram based on forecast model search LPRs, and said step 4.1 specifically comprises following step:
Step 4.1.1; Calculate under the color of the same race according to following formula (1) and (2); The ratio of width to height similarity of the boundary rectangle frame between the anchor point connected domain combination of the trade mark picture sample in the trade mark picture sample storehouse and the effective connected domain of test picture; Wherein, said the ratio of width to height similarity comprises the difference S of the ratio of width to height that the anchor point connected domain of the ratio of width to height and the trade mark picture sample of the effective connected domain of test in the picture makes up
1(R
Tc, R
Lc) and the anchor point connected domain of effective connected domain and the trade mark picture sample of test in the picture make up the difference S of corresponding wide ratio and ratios
2(R
Tc, R
Lc), if S
1(R
Tc, R
Lc) and S
2(R
Tc, R
Lc) simultaneously less than corresponding threshold value separately, explain that then both are similar:
S
1(R
tc，R
lc)＝W
tc/H
tc-W
lc/H
lc＜T1 (1)
S
2(R
tc，R
lc)＝W
tc/W
lc-H
tc/H
lc＜T2 (2)
Wherein, represent to test the anchor point connected domain combination of effective connected domain and trade mark picture sample in the picture respectively for subscript tc, lc; R representes the boundary rectangle frame, and W and H represent the wide and high of boundary rectangle frame respectively, and T1, T2 are threshold value, but the T1 value is 0.8, but the T2 value is 0.5.
Step 4.1.2 calculates the similarity S that is judged as the frontier point-centroid distance angular histogram of similar test picture and trade mark picture sample through said step 4.2 according to formula (3)
3(h
Tc, h
Lc), if S
3(h
Tc, h
Lc) greater than a threshold value, think that then both mate, it is right to obtain mating picture:
Wherein, h representes histogram; A and d represent the dimension of angle and distance respectively, and T3 is a threshold value, but value is 0.6.
Step 4.1.3; The coupling picture that relative position and the said step 4.2 of the anchor point connected domain of utilizing trade mark picture sample in trade mark picture sample obtains between area ratio; Obtain testing the regional rectangle frame at effective connected domain place of mating with said anchor point connected domain in the picture, this zone rectangle frame is as the trade mark pre-determined bit area L PR in the test picture.
Step 4.2, the trade mark pre-determined bit area L PRs that screening searches based on the pixel distribution histogram.
Said step 4.2 further comprises following step:
Step 4.2.1, the trade mark pre-determined bit area L PRs for all acquisitions, calculate the area registration between the LPR in twos according to formula (4):
r
t＝S
o/S
t＞T4&&r
a＝S
o/S
a＞T4 (4)
Wherein, S
oBe the area that two LPR overlap the zone, S
tAnd S
aBe respectively two LPR areas separately, r
tAnd r
aBe that two LPR overlap the ratio that regional area accounts for two LPR respectively, T4 is a threshold value, can be taken as 0.9.
To satisfy two LPR merging that the registration threshold value shown in the formula (4) requires, and those two LPR before replacing.LPRs to all merges according to the method described above, until can not be merged.
Through this step, just all LPRs have been carried out merging and gone heavily.
Step 4.2.2, operate each LPR that obtains through said step 4.2.1 successively as follows:
Area ratio according to trade mark picture sample and this LPR; All boundary rectangle frames of the effective connected domain spatial concentration of trade mark picture sample are projected among this LPR; If when existing the zone more than 80% of some connected domains to be in certain boundary rectangle frame among this LPR, think that then this connected domain is doubtful trade mark connected domain.So just obtained doubtful trade mark connected domains all among this LPR;
Step 4.2.3 according to said step 2.6.1-2.6.3, calculates its valid pixel distribution histogram for the doubtful trade mark connected domain that said step 4.2.2 obtains;
Step 4.2.4; The valid pixel distribution histogram that said step 4.2.3 is obtained and the valid pixel distribution histogram of trade mark picture sample are that measuring similarity carries out the similarity matching test with the Euclidean distance, the trade mark coupling after the LPR that satisfies certain similarity threshold requirement remains and carries out.
Step 5; The trade mark coupling is carried out with the trade mark picture in the said trade mark picture database in the trade mark pre-determined bit zone that the characteristic that utilization is set up based on trade mark color and shape information obtains said step 4, mate successful trade mark pre-determined bit zone and is in the test picture trade mark that finally detection obtains.
The color of trade mark and shape facility are trade mark notable attribute the most, and the present invention mainly uses piecemeal color histogram and these two characteristics of color boundaries gradient symbiosis histogram (CEGCH).This step is that filter in the trade mark pre-determined bit zone that standard comes said step 4 is obtained with the characteristic similarity, and the trade mark pre-determined bit regional determination that characteristic similarity satisfies certain threshold value is the final trade mark zone that obtains of detecting.
Concrete, said step 5 further may further comprise the steps:
Step 5.1 after getting access to all LPR, is extracted the piecemeal color histogram and the color boundaries gradient symbiosis histogram of the trade mark picture in LPR and the said trade mark picture database respectively;
Extraction list of references of said piecemeal color histogram " Color indexing; M.J.Swain and D.H.Ballard; International Journal of Computer Vision 7 (1), pp.11-32,1991 " and partitioned mode shown in Figure 6 are carried out; Mode described in the histogrammic extraction list of references of said color boundaries gradient symbiosis " Content-based retrieval of logo and trademarks in unconstrained color image databases using Color Edge Gradient Co-occurrence Histograms; R.Phan and D.Androutsos; Computer Vision and Image Understanding (114); pp.66-84,2010 " is carried out.
Step 5.2 about the description of piecemeal color histogram and CEGCH similarity, is tried to achieve the piecemeal color histogram similarity S (C of the trade mark picture in LPR and the said trade mark picture database in above-mentioned two documents of reference
Q, C
L) and and CEGCH similarity S (CH
Q, CH
L);
Step 5.3 is according to the S (C that calculates
Q, C
L) and S (CH
Q, CH
L) calculate the trade mark picture in LPR and the said trade mark picture database comprehensive similarity S (Q, L):
S(Q，L)＝k
1S(C
Q，C
L)+(1-k
1)S(CH
Q，CH
L) (5)
Wherein, Q represents the trade mark picture in the said trade mark picture database, i.e. target trade mark, and L represents LPR, 0＜k
1＜1.0, S (C
Q, C
L) be the piecemeal color histogram similarity of target trade mark and LPR, C
QRepresent the piecemeal color histogram of target trade mark, C
LRepresent the piecemeal color histogram of LPR, S (CH
Q, CH
L) be the CEGCH similarity of target trade mark and LPR, CH
QRepresent the CEGCH of target trade mark, CH
LRepresent the CEGCH of LPR.
(Q during L) greater than a certain threshold value (such as 0.6), thinks that then current LPR is the trade mark zone that final detection obtains, and that is to say that most possible trade mark position is thought in the position that similarity is the highest as comprehensive similarity S.
Algorithm performance is estimated
The performance evaluation of the inventive method is through to whether containing trade mark in the test picture verifying.The algorithm performance index has three, precision ξ, recall rate δ and detection time.
Judgement schematics is expressed as:
The inventive method is as shown in table 1 for the average detected time of different trade marks (Startbuck (S), Pepsi Cola (P), Baidu (B), Mongolia Ox (M), the The Hongkong and Shanghai Banking Corporation Limited (HSBC) (H) and China Netcom (C)):
Table 1
Fig. 7 has provided the test duration along with the curve that dimension of picture to be measured (area) changes, and from curve, can find out, along with the increase of area, pre-determined bit grew steadily with overall detection time.
Fig. 8 has provided precision-recall rate curve; Article 3, curve is respectively the precision-recall rate curve that in the original size database, adopts SCCD pre-determined bit and CEGCH feature detection trade mark; At the precision-recall rate curve of down-sampled data storehouse (256 * 384) employing SCCD pre-determined bit and CEGCH feature detection trade mark, adopt moving window to detect the precision-recall rate curve of (OSW) and CEGCH feature detection trade mark in down-sampled data storehouse (256 * 384).As can be seen from the figure, this method obviously is better than traditional OSW method, and need not carry out down-sampling and just can detect, and has reduced because down-sampling causes trade mark to detect the possibility of failure.
The inventive method adopts SCCD pre-determined bit and moving window to detect in down-sampled data storehouse (256 * 384); And adopt the detection of SCCD in the raw image data storehouse total contrast table correct and faults as shown in table 2; Wherein So representes to adopt the detection of SCCD pre-determined bit in the raw image data storehouse; Ss representes to adopt the detection of SCCD pre-determined bit in the down-sampled data storehouse, and Os representes to adopt the detection of moving window in the down-sampled data storehouse:
The correct detection and the number of false detections of the different trade marks of table 2
Fig. 9 has provided and has utilized the present invention's some test results in picture database to be measured, from the result, can find out, the inventive method to yardstick, block, factor such as illumination variation, rotation has certain robustness.
Experiment showed, with previous methods and compare that the method that the present invention proposes is to the characteristics of trade mark, further improved trade mark and detected speed and precision with location identification, can be used in to the real time monitoring of the web advertisement etc.
In sum; The present invention proposes a kind of new quick trade mark detects and recognition methods; This method has proposed a kind of with the combine algorithm of the fast detecting of carrying out the trade mark pre-determined bit and location identification of trade mark connected domain and spatial relationship characteristic thereof; Can navigate to doubtful trade mark position fast, thereby carry out the trade mark coupling.
Above-described specific embodiment; The object of the invention, technical scheme and beneficial effect have been carried out further explain, and institute it should be understood that the above is merely specific embodiment of the present invention; Be not limited to the present invention; All within spirit of the present invention and principle, any modification of being made, be equal to replacement, improvement etc., all should be included within protection scope of the present invention.
Claims (18)
1. trade mark detection method based on spatial communication territory pre-determined bit is characterized in that this method may further comprise the steps:
Step 1 is set up the trade mark picture sample storehouse that comprises a plurality of trade mark pictures;
Step 2 is for the trade mark picture in the trade mark picture sample storehouse is set up spatial communication territory descriptor SCCD characteristic;
Step 3, input contains the test picture of target trade mark, for said test picture, sets up its CCD characteristic;
Step 4 utilizes the SCCD characteristic of the trade mark picture that said step 2 obtains in the test picture, to obtain trade mark pre-determined bit area L PRs;
Step 5; The trade mark coupling is carried out with the trade mark picture in the said trade mark picture database in the trade mark pre-determined bit zone that the characteristic that utilization is set up based on trade mark color and shape information obtains said step 4, mate successful trade mark pre-determined bit zone and is in the test picture trade mark that finally detection obtains;
Wherein, said step 2 further comprises following step:
Step 2.4 utilizes the color of trade mark to have certain zonal characteristic, obtains the effective connected domain of the conduct of the color connected domain except background colour in the trade mark picture;
Step 2.6 obtains all effective connected domains according to said step 2.4, and the trade mark picture is set up the SCCD characteristic, and said SCCD characteristic comprises connected domain forecast model and effective these two characteristics of connected domain pixel distribution histogram;
The step of setting up the CCD characteristic of test picture in the said step 3 further may further comprise the steps:
Step 3.2 utilizes the color of trade mark to have certain zonal characteristic, obtains the main connected domain of color connected domain conduct except background colour in the test picture;
The connected domain that step 3.3, the main connected domain that said step 3.2 is obtained are carried out gray space is cut apart, and obtains testing effective connected domain of picture;
Step 3.4 is calculated the CCD characteristic of testing each effective connected domain of picture according to said step 2.6;
Said step 4 further is divided into two steps:
Step 4.1 is based on the trade mark pre-determined bit area L PRs in the forecast model search test picture;
Step 4.2, the LPRs that screening searches based on the pixel distribution histogram.
2. method according to claim 1 is characterized in that, said step 2 also comprised following step before said step 2.4:
Step 2.1 is transformed into tone saturation degree brightness HSV space with said trade mark picture from the RGB rgb space;
Step 2.2 is carried out the quantification of 8 kinds of colors to the trade mark picture that is transformed into the HSV space, obtains through the trade mark picture behind 8 kinds of color quantizings;
Step 2.3 is judged background colour and noise color through the trade mark picture after quantizing, and the trade mark picture is carried out the color smoothing processing, to reduce the influence of noise color.
3. method according to claim 2 is characterized in that, said 8 kinds of colors are the red greenish-yellow blue tangerine powder of black and white.
4. method according to claim 2 is characterized in that, said step 2.3 further comprises following step:
Step 2.3.1; Get on four angles of trade mark picture after the quantification and be of a size of 5 * 5 color block; With number of pixels is maximum in each color block color main color as this color block; Occurrence number in the main color of four color block greater than 2 the color background colour as this trade mark picture, is defaulted as one of background colour with white simultaneously;
Step 2.3.2, number of pixels is less than 20 color and is judged to be the noise color in the trade mark picture after will passing through hsv color and quantizing;
Step 2.3.3 to certain noise colored pixels, utilizes around it in 8 pixels that maximum non-noise color of occurrence number that the color of this noise colored pixels is carried out assignment, and is level and smooth the trade mark picture is carried out color.
5. method according to claim 1 is characterized in that, said step 2 further comprises after said step 2.4:
Step 2.5 sorts all effective connected domains from big to small according to area and numbers, with the first five effective connected domain of rank as the anchor point connected domain.
6. method according to claim 1; It is characterized in that; Said connected domain forecast model characteristic is divided into anchor point connected domain assemblage characteristic and effective connected domain space collection characteristic; Wherein, said anchor point connected domain assemblage characteristic is with color-anchor point connected domain combination of sets, and the permutation and combination of color anchorage connected domain promptly of the same race is described; Said effective connected domain space collection characteristic is with boundary rectangle frame-anchor point connected domain relation, and promptly the boundary rectangle frame of anchor point connected domain is described with the relation of the anchor point connected domain that is wherein comprised.
7. method according to claim 6; It is characterized in that; Said step 2.6 further comprises for each color in the said combination of sets-anchor point connected domain combination, writes down its color, the knowledge of boundary rectangle collimation mark, the anchor point connected domain combination total area and frontier point-centroid distance angular histogram information; Wherein,
The boundary rectangle frame of a certain anchor point connected domain is the minimum rectangle frame that this anchor point connected domain is all closely included;
In frontier point-centroid distance angular histogram; Frontier point is the frontier point summation of all anchor point connected domains in each anchor point connected domain combination; Barycenter makes up the central point of pairing boundary rectangle frame for this anchor point connected domain; Distance be each frontier point with the central point of corresponding boundary rectangle frame between Euclidean distance, angle is each frontier point and the angle of the relative horizontal direction of central point of boundary rectangle frame accordingly.
8. method according to claim 6; It is characterized in that; In said effective connected domain spatial concentration, when the boundary rectangle frame of certain anchor point connected domain was comprised by the boundary rectangle frame of another anchor point connected domain fully, this anchor point connected domain counted in the big pairing connected domain of boundary rectangle frame.
9. method according to claim 1 is characterized in that, the histogrammic foundation of said effective connected domain pixel distribution further comprises following step:
Step 2.6.1 evenly is divided into 4 piecemeals with the trade mark picture;
Step 2.6.2; Each piecemeal all is divided into 8 parts in the horizontal and vertical directions; Its horizontal ordinate position of pixel basis in the effective connected domain of trade mark picture is fallen in corresponding part; Form the pixel distribution histogram of two 8 dimensions, successively each piece is carried out identical operations, finally obtain the pixel distribution histogram of 88 dimensions of total of 4 piecemeals;
Step 2.6.3 utilizes the area of each piecemeal that its corresponding pixel distribution histogram is carried out normalization, then obtains effective connected domain pixel distribution histogram of 64 dimensions of every width of cloth trade mark picture.
10. method according to claim 2 is characterized in that, in the said step 3, before said step 3.2, further comprises:
Step 3.1, according to said step 2.1-step 2.3, it is level and smooth that the test picture is carried out HSV space conversion, color quantizing and color.
11. method according to claim 1 is characterized in that, said step 3.3 further may further comprise the steps:
Step 3.3.1 will test picture and transform to gray space;
Step 3.3.2; Said main connected domain is carried out the detection of the connected domain area of gray space; If the area of a main connected domain greater than certain threshold value and its gray-scale value variance greater than certain value; Then utilize big Tianjin method that this main connected domain is carried out gray space and cut apart, two sub-connected domains after obtaining cutting apart;
Step 3.3.3 is if the average gray value difference of said two sub-connected domains greater than certain threshold value, then covers that the main connected domain before cutting apart with these two that obtain after cutting apart sub-connected domains;
Step 3.3.4 is cut apart all main connected domains according to said step 3.3.2 and step 3.3.3, ends up to not being further divided into, and removes the too little connected domain of area then, the remaining effective connected domain that is the test picture.
12. method according to claim 1 is characterized in that, said step 4.1 further comprises following step:
Step 4.1.1; Calculate under the color of the same race; The ratio of width to height similarity of the boundary rectangle frame between the anchor point connected domain combination of the trade mark picture sample in the trade mark picture sample storehouse and the effective connected domain of test picture; If said similarity satisfies certain threshold condition, this trade mark picture sample and this test picture analogies are described then;
Step 4.1.2; Calculating is judged as the similarity of the frontier point-centroid distance angular histogram of similar test picture and trade mark picture sample through said step 4.2; If said similarity is greater than a threshold value; Think that then this trade mark picture sample and this test picture mate, thereby it is right to obtain mating picture;
Step 4.1.3; The coupling picture that relative position and the said step 4.2 of the anchor point connected domain of utilizing trade mark picture sample in trade mark picture sample obtains between area ratio; Obtain testing the regional rectangle frame at effective connected domain place of mating with said anchor point connected domain in the picture, this zone rectangle frame is as the trade mark pre-determined bit zone in the test picture.
13. method according to claim 12 is characterized in that, said the ratio of width to height similarity comprises the difference S of the ratio of width to height that the anchor point connected domain of the ratio of width to height and the trade mark picture sample of the effective connected domain of test in the picture makes up
1(R
Tc, R
Lc), the anchor point connected domain of effective connected domain and the trade mark picture sample of test in the picture makes up the difference S of corresponding wide ratio and ratios
2(R
Tc, R
Lc):
S
1(R
tc，R
lc)＝W
tc/H
tc-W
lc/H
lc，
S
2(R
tc，R
lc)＝W
tc/W
lc-H
tc/H
lc，
Wherein, subscript tc, lc represent to test the anchor point connected domain combination of effective connected domain and trade mark picture sample in the picture respectively; R representes the boundary rectangle frame, and W and H represent the wide and high of boundary rectangle frame respectively.
14. method according to claim 13 is characterized in that, said threshold condition is: S
1(R
Tc, R
Lc) and S
2(R
Tc, R
Lc) simultaneously less than corresponding threshold value separately.
15. method according to claim 12 is characterized in that, among the said step 4.1.2, and the similarity S of the frontier point-centroid distance angular histogram of test picture and trade mark picture sample
3(h
Tc, h
Lc) calculate according to following formula:
Wherein, h representes histogram, and a and d represent the dimension of angle and distance respectively.
16. method according to claim 9 is characterized in that, said step 4.2 further comprises following step:
Step 4.2.1, for the LPRs of all acquisitions, two LPR overlap the ratio r that regional area accounts for two LPR respectively according to computes
tAnd r
a:
r
t＝S
o/S
t，r
a＝S
o/S
a，
Wherein, S
oBe the area that two LPR overlap the zone, S
tAnd S
aBe respectively two LPR areas separately;
If satisfy: r
t＞T4&&r
a＞T4 then merges these two LPR, those two LPR before replacing;
All LPRs are merged according to identical method, until can not be merged;
Step 4.2.2, operate each LPR that obtains through said step 4.2.1 successively as follows:
Area ratio according to trade mark picture sample and this LPR; All boundary rectangle frames of the effective connected domain spatial concentration of trade mark picture sample are projected among this LPR; If when existing the zone more than 80% of some connected domains to be in certain boundary rectangle frame among this LPR, think that then this connected domain is doubtful trade mark connected domain;
Step 4.2.3 according to said step 2.6.1-2.6.3, calculates its valid pixel distribution histogram for the doubtful trade mark connected domain that said step 4.2.2 obtains;
Step 4.2.4, the valid pixel distribution histogram that said step 4.2.3 is obtained and the valid pixel distribution histogram of trade mark picture sample are that measuring similarity carries out the similarity matching test with the Euclidean distance, are met the LPR of certain similarity threshold requirement.
17. method according to claim 1 is characterized in that, in the said step 5, is characterized as piecemeal color histogram and color boundaries gradient symbiosis histogram CEGCH based on what trade mark color and shape information were set up.
18. method according to claim 1 is characterized in that, said step 5 further may further comprise the steps:
Step 5.1 after getting access to all LPR, is extracted the piecemeal color histogram and the color boundaries gradient symbiosis histogram of the trade mark picture in LPR and the said trade mark picture database respectively;
Step 5.2, the piecemeal color histogram similarity S (C of the trade mark picture in calculating LPR and the said trade mark picture database
Q, C
L) and CEGCH similarity S (CH
Q, CH
L);
Step 5.3 is according to the S (C that calculates
Q, C
L) and S (CH
Q, CH
L) calculate the trade mark picture in LPR and the said trade mark picture database comprehensive similarity S (Q, L):
S(Q，L)＝k
1S(C
Q，C
L)+(1-k
1)S(CH
Q，CH
L)，
Wherein, Q represents the trade mark picture in the said trade mark picture database, i.e. target trade mark, and L represents LPR, 0＜k
1＜1.0;
(Q during L) greater than a certain threshold value, thinks that then current LPR is the trade mark zone that final detection obtains as comprehensive similarity S.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012102259223A CN102799653A (en) | 2012-06-29 | 2012-06-29 | Logo detection method based on spatial connected domain prepositioning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012102259223A CN102799653A (en) | 2012-06-29 | 2012-06-29 | Logo detection method based on spatial connected domain prepositioning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102799653A true CN102799653A (en) | 2012-11-28 |
Family
ID=47198763
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2012102259223A Pending CN102799653A (en) | 2012-06-29 | 2012-06-29 | Logo detection method based on spatial connected domain prepositioning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102799653A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103258037A (en) * | 2013-05-16 | 2013-08-21 | 西安工业大学 | Trademark identification searching method for multiple combined contents |
| WO2014000261A1 (en) * | 2012-06-29 | 2014-01-03 | 中国科学院自动化研究所 | Trademark detection method based on spatial connected component pre-location |
| CN103854278A (en) * | 2012-12-06 | 2014-06-11 | 五邑大学 | Printed circuit board image registration method based on shape context of mass center of communicated region |
| CN104123528A (en) * | 2013-04-24 | 2014-10-29 | 阿里巴巴集团控股有限公司 | Image recognition method and apparatus |
| CN104299235A (en) * | 2014-10-10 | 2015-01-21 | 中国科学院长春光学精密机械与物理研究所 | Registration descriptor direction calculation method based on area integral formula |
| CN104331695A (en) * | 2014-09-01 | 2015-02-04 | 西北工业大学 | Robust round identifier shape quality detection method |
| CN104484853A (en) * | 2014-12-24 | 2015-04-01 | 北京奇虎科技有限公司 | Method and device for clipping pictures containing graphical signs |
| CN105303189A (en) * | 2014-07-29 | 2016-02-03 | 阿里巴巴集团控股有限公司 | Method and device for detecting specific identification image in predetermined area |
| CN106530194A (en) * | 2015-09-09 | 2017-03-22 | 阿里巴巴集团控股有限公司 | Method and apparatus for detecting pictures of suspected infringing products |
| CN107527055A (en) * | 2017-08-04 | 2017-12-29 | 佛山市国方商标服务有限公司 | Image divides card processing method, device and image search method, device and system |
| CN107844803A (en) * | 2017-10-30 | 2018-03-27 | ***股份有限公司 | The method and apparatus that a kind of picture compares |
| CN108763380A (en) * | 2018-05-18 | 2018-11-06 | 徐庆 | Brand recognition search method, device, computer equipment and storage medium |
| CN108875753A (en) * | 2018-04-03 | 2018-11-23 | 南昌奇眸科技有限公司 | A kind of trademark image retrieval feature extracting method |
| CN108898607A (en) * | 2018-04-03 | 2018-11-27 | 南昌奇眸科技有限公司 | A kind of Image Multiscale feature extracting method |
| CN108961252A (en) * | 2018-07-27 | 2018-12-07 | Oppo（重庆）智能科技有限公司 | It detects logo and pastes undesirable method, electronic device and computer readable storage medium |
| CN109376741A (en) * | 2018-09-10 | 2019-02-22 | 平安科技（深圳）有限公司 | Recognition methods, device, computer equipment and the storage medium of trademark infringement |
| CN110287993A (en) * | 2019-05-22 | 2019-09-27 | 广东精点数据科技股份有限公司 | A kind of data preprocessing method and system based on characteristics of image refinement |
| CN111126372A (en) * | 2019-11-29 | 2020-05-08 | 北京金山云网络技术有限公司 | Method and device for marking logo area in video and electronic equipment |
| CN117372719A (en) * | 2023-12-07 | 2024-01-09 | 四川迪晟新达类脑智能技术有限公司 | Target detection method based on screening |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101246551A (en) * | 2008-03-07 | 2008-08-20 | 北京航空航天大学 | Fast license plate locating method |
-
2012
- 2012-06-29 CN CN2012102259223A patent/CN102799653A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101246551A (en) * | 2008-03-07 | 2008-08-20 | 北京航空航天大学 | Fast license plate locating method |
Non-Patent Citations (1)
| Title |
|---|
| ZHANG YUAN等: "Spatial connected component pre-locating algorithm for rapid logo detection", 《2012 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING》, 30 March 2012 (2012-03-30) * |
Cited By (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014000261A1 (en) * | 2012-06-29 | 2014-01-03 | 中国科学院自动化研究所 | Trademark detection method based on spatial connected component pre-location |
| CN103854278A (en) * | 2012-12-06 | 2014-06-11 | 五邑大学 | Printed circuit board image registration method based on shape context of mass center of communicated region |
| CN104123528B (en) * | 2013-04-24 | 2018-01-02 | 阿里巴巴集团控股有限公司 | A kind of method and device for identifying picture |
| CN104123528A (en) * | 2013-04-24 | 2014-10-29 | 阿里巴巴集团控股有限公司 | Image recognition method and apparatus |
| CN103258037A (en) * | 2013-05-16 | 2013-08-21 | 西安工业大学 | Trademark identification searching method for multiple combined contents |
| US10360689B2 (en) | 2014-07-29 | 2019-07-23 | Alibaba Group Holding Limited | Detecting specified image identifiers on objects |
| US10885644B2 (en) | 2014-07-29 | 2021-01-05 | Banma Zhixing Network (Hongkong) Co., Limited | Detecting specified image identifiers on objects |
| CN105303189A (en) * | 2014-07-29 | 2016-02-03 | 阿里巴巴集团控股有限公司 | Method and device for detecting specific identification image in predetermined area |
| CN104331695B (en) * | 2014-09-01 | 2017-06-13 | 西北工业大学 | A kind of circle marker symbol shape quality detection method of robust |
| CN104331695A (en) * | 2014-09-01 | 2015-02-04 | 西北工业大学 | Robust round identifier shape quality detection method |
| CN104299235B (en) * | 2014-10-10 | 2017-06-13 | 中国科学院长春光学精密机械与物理研究所 | Registration based on area integral formula describes sub- direction calculating method |
| CN104299235A (en) * | 2014-10-10 | 2015-01-21 | 中国科学院长春光学精密机械与物理研究所 | Registration descriptor direction calculation method based on area integral formula |
| CN104484853A (en) * | 2014-12-24 | 2015-04-01 | 北京奇虎科技有限公司 | Method and device for clipping pictures containing graphical signs |
| CN106530194B (en) * | 2015-09-09 | 2020-02-07 | 阿里巴巴集团控股有限公司 | Method and device for detecting suspected infringement product picture |
| CN106530194A (en) * | 2015-09-09 | 2017-03-22 | 阿里巴巴集团控股有限公司 | Method and apparatus for detecting pictures of suspected infringing products |
| CN107527055A (en) * | 2017-08-04 | 2017-12-29 | 佛山市国方商标服务有限公司 | Image divides card processing method, device and image search method, device and system |
| CN107527055B (en) * | 2017-08-04 | 2018-12-11 | 佛山市国方商标服务有限公司 | Image divides card processing method, device and image search method, device and system |
| CN107844803A (en) * | 2017-10-30 | 2018-03-27 | ***股份有限公司 | The method and apparatus that a kind of picture compares |
| CN107844803B (en) * | 2017-10-30 | 2021-12-28 | ***股份有限公司 | Picture comparison method and device |
| CN108875753A (en) * | 2018-04-03 | 2018-11-23 | 南昌奇眸科技有限公司 | A kind of trademark image retrieval feature extracting method |
| CN108898607A (en) * | 2018-04-03 | 2018-11-27 | 南昌奇眸科技有限公司 | A kind of Image Multiscale feature extracting method |
| CN108763380A (en) * | 2018-05-18 | 2018-11-06 | 徐庆 | Brand recognition search method, device, computer equipment and storage medium |
| CN108763380B (en) * | 2018-05-18 | 2022-03-08 | 徐庆 | Trademark identification retrieval method and device, computer equipment and storage medium |
| CN108961252A (en) * | 2018-07-27 | 2018-12-07 | Oppo（重庆）智能科技有限公司 | It detects logo and pastes undesirable method, electronic device and computer readable storage medium |
| CN109376741A (en) * | 2018-09-10 | 2019-02-22 | 平安科技（深圳）有限公司 | Recognition methods, device, computer equipment and the storage medium of trademark infringement |
| CN110287993A (en) * | 2019-05-22 | 2019-09-27 | 广东精点数据科技股份有限公司 | A kind of data preprocessing method and system based on characteristics of image refinement |
| CN111126372A (en) * | 2019-11-29 | 2020-05-08 | 北京金山云网络技术有限公司 | Method and device for marking logo area in video and electronic equipment |
| CN117372719A (en) * | 2023-12-07 | 2024-01-09 | 四川迪晟新达类脑智能技术有限公司 | Target detection method based on screening |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102799653A (en) | Logo detection method based on spatial connected domain prepositioning | |
| CN102012939B (en) | Method for automatically tagging animation scenes for matching through comprehensively utilizing overall color feature and local invariant features | |
| CN102176208B (en) | Robust video fingerprint method based on three-dimensional space-time characteristics | |
| Ren et al. | Fusion of intensity and inter-component chromatic difference for effective and robust colour edge detection | |
| CN103473551A (en) | Station logo recognition method and system based on SIFT operators | |
| CN102915433B (en) | Character combination-based license plate positioning and identifying method | |
| CN105023008A (en) | Visual saliency and multiple characteristics-based pedestrian re-recognition method | |
| CN102881160B (en) | Outdoor traffic sign identification method under low-illumination scene | |
| CN105894655B (en) | Paper currency detection and recognition methods under complex environment based on RGB-D cameras | |
| CN106991370B (en) | Pedestrian retrieval method based on color and depth | |
| Awrangjeb et al. | Improved building detection using texture information | |
| CN103383733A (en) | Lane video detection method based on half-machine study | |
| CN104850850A (en) | Binocular stereoscopic vision image feature extraction method combining shape and color | |
| WO2014151681A2 (en) | Methods and apparatus to estimate demography based on aerial images | |
| Zhang et al. | A multi-feature fusion based traffic light recognition algorithm for intelligent vehicles | |
| CN102799859A (en) | Method for identifying traffic sign | |
| CN105678318B (en) | The matching process and device of traffic sign | |
| CN103810722A (en) | Moving target detection method combining improved LBP (Local Binary Pattern) texture and chrominance information | |
| CN104463232A (en) | Density crowd counting method based on HOG characteristic and color histogram characteristic | |
| CN103714181A (en) | Stratification specific figure search method | |
| CN107264570B (en) | Steel rail light band distribution detecting device and method | |
| CN103839267A (en) | Building extracting method based on morphological building indexes | |
| CN103440035A (en) | Gesture recognition system in three-dimensional space and recognition method thereof | |
| CN103530638A (en) | Method for matching pedestrians under multiple cameras | |
| WO2014000261A1 (en) | Trademark detection method based on spatial connected component pre-location |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20121128 |