WO2014000261A1 - Procédé de détection de marques déposées basé sur un pré-emplacement de composantes associées dans l'espace - Google Patents

Procédé de détection de marques déposées basé sur un pré-emplacement de composantes associées dans l'espace Download PDF

Info

Publication number
WO2014000261A1
WO2014000261A1 PCT/CN2012/077884 CN2012077884W WO2014000261A1 WO 2014000261 A1 WO2014000261 A1 WO 2014000261A1 CN 2012077884 W CN2012077884 W CN 2012077884W WO 2014000261 A1 WO2014000261 A1 WO 2014000261A1
Authority
WO
WIPO (PCT)
Prior art keywords
trademark
color
connected domain
domain
anchor
Prior art date
Application number
PCT/CN2012/077884
Other languages
English (en)
Chinese (zh)
Inventor
张树武
张渊
梁伟
Original Assignee
中国科学院自动化研究所
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 中国科学院自动化研究所 filed Critical 中国科学院自动化研究所
Priority to PCT/CN2012/077884 priority Critical patent/WO2014000261A1/fr
Publication of WO2014000261A1 publication Critical patent/WO2014000261A1/fr

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images

Definitions

  • the present invention relates to the field of target detection and recognition technology, and more particularly to a trademark detection method based on spatially connected domain pre-positioning, which can be used for rapid trademark detection, trademark retrieval, identification, and monitoring.
  • Trademark detection and identification is one of the challenging tasks in the field of target detection and identification. How to accurately detect and locate the trademark is a challenging task under the influence of scale transformation, perspective transformation, illumination, occlusion, background interference and so on.
  • various information media have developed rapidly, such as television, radio, and the Internet. These information media are filled with a large amount of advertising information every day, and with the large number of trademarks nowadays, how to effectively manage and monitor these advertising information to ensure that the safety of consumers and businesses is further emphasized.
  • the rapid trademark detection method based on the pre-positioning of the spatial connectivity domain is to meet the advertising monitoring requirements in the information security field pictures and videos.
  • the present invention is different from the above prior art, and based on the characteristics of the trademark itself, a trademark detection method based on the pre-positioning of the spatial connectivity domain is proposed.
  • the invention provides a trademark detecting method based on a predetermined position of a spatial connected domain, which is characterized in that the method comprises the following steps:
  • Step 1 Create a sample library of trademark images containing multiple trademark images
  • Step 2 Establish a spatial connected domain descriptor SCCD feature for the trademark image in the trademark image sample library
  • Step 3 input a test picture containing the target trademark, and establish a CCD feature for the test picture;
  • Step 4 obtaining the trademark pre-positioning area LPRs in the test picture by using the SCCD feature of the trademark image obtained in the step 2;
  • Step 5 using the feature established based on the trademark color and the shape information, the trademark pre-positioning area obtained in the step 4 is matched with the trademark image in the trademark photo database, and the matching trademark pre-positioning area is tested.
  • Step 2.4 Using the color of the trademark to have a certain regional characteristic, obtaining a color connected domain other than the background color in the trademark image as an effective connected domain;
  • Step 2.6 obtaining all valid connected domains according to the step 2.4, and establishing an SCCD feature for the trademark image, the SCCD feature including a connected domain prediction model and an effective connected domain pixel distribution histogram;
  • the step of establishing the CCD feature of the test picture in the step 3 further includes the following steps:
  • Step 3.2 Using the color of the trademark to have a certain regional characteristic, obtain a color connected domain other than the background color in the test image as the main connected domain;
  • Step 3.3 performing a connected domain partition of the gray space in the main connected domain obtained in step 3.2 to obtain an effective connected domain of the test picture;
  • Step 3.4 calculating CCD characteristics of each valid connected domain of the test picture according to the step 2.6;
  • the step 4 is divided into two steps:
  • Step 4.1 Searching for the trademark pre-positioning area LPRs in the test picture based on the prediction model; Step 4.2, filtering the searched LPRs based on the pixel distribution histogram.
  • the method of the invention fully utilizes the color, shape and regional characteristics of the trademark to detect and locate the trademark, further improves the speed and precision of the trademark detection and positioning recognition, and at the same time, the target is subject to scale transformation, the angle of view transformation under certain conditions, illumination
  • the detection of trademarks under the influence of occlusion, background interference, etc. has a good effect.
  • FIG. 1 is a flow chart of a method for detecting a trademark based on a predetermined position of a spatial connectivity domain according to the present invention.
  • Figure 2 is an example of some trademark images in the trademark image sample library.
  • Figure 3 is a flow chart for obtaining the spatial connected domain descriptor (SCCD) feature.
  • SCCD spatial connected domain descriptor
  • Figure 4 is a schematic diagram of the features of the SCCD.
  • Figure 5 is a schematic diagram of obtaining the LPR of Pepsi.
  • Figure 6 is a block diagram of a block color histogram.
  • Fig. 7 is a graph showing the change of the detection time of the picture to be tested with the area of the picture by the method of the present invention.
  • Figure 8 is a graph comparing the accuracy-recall ratio of the method of the present invention and the conventional method.
  • Figure 9 is a schematic diagram of test results in a database of images to be tested using the present invention.
  • the invention proposes a trademark detecting method based on a predetermined position of a spatial connected domain, which combines the trademark connected domain and its spatial relationship features to quickly detect and locate the trademark pre-positioning.
  • 1 is a flow chart of a method for detecting a rapid trademark based on a pre-positioning of a spatial connectivity domain according to the present invention.
  • the method for detecting a rapid trademark based on a predetermined location of a spatial connectivity domain includes the following steps: Step 1 a sample library of trademark images of multiple trademark images; Since the object of the present invention is trademark detection, that is, detecting whether a target trademark is included in the image to be tested, the number of target trademark images in the trademark image sample library is determined according to actual detection requirements.
  • Figure 2 shows some of the trademark images in the sample library of the trademark image of the present invention.
  • the background of the trademark is white, and the color of the main body of the trademark is not affected by external factors, such as illumination.
  • the boundaries of the trademark are clear and easy to separate from the background.
  • Step 2 Establish a spatial connected domain descriptor (SCCD) feature for all trademark images in the trademark image sample library;
  • SCCD spatial connected domain descriptor
  • FIG. 3 is a flow chart for acquiring a spatial connected domain descriptor (SCCD) feature. As shown in FIG. 3, the step 2 includes the following steps:
  • Step 2.1 converting the trademark image from red, green and blue RGB space to hue saturation brightness HSV space;
  • Step 2.2 Quantify 8 colors of the trademark image converted to the HSV space (for example, black and white, red, green, yellow, and orange), and obtain a trademark image after 8 colors;
  • the reference of the present invention is "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 )
  • the quantization process quantizes eight colors of the trademark image converted to the HSV space, but the present invention differs from the method in this document in that the present invention discards the gray used in the literature, that is, only quantizes the non-colored pixels to black and White, other processing procedures are consistent with the processing of the literature.
  • Step 2.3 judging the background color and noise color of the quantized trademark image, and performing color smoothing on the trademark image to reduce the influence of noise color;
  • Step 2.3.1 taking the color block of the quantized trademark image with the size of 5 X 5 at the four corners, and using the color with the largest number of pixels in each color block as the main color of the color block, and the four color blocks A color having a number of times greater than 2 in the main color is used as the background color of the trademark image, and white is defaulted to one of the background colors;
  • Step 2.3.2 determining a color of the number of pixels in the trademark image after the HSV color quantization is less than 20 as a noise color
  • Step 2.3.3 for a certain noise color pixel, the color of the noise color pixel is assigned by using the non-noise color which is the most frequently occurring among the eight pixels around it, thereby realizing the color smoothing of the trademark image.
  • Step 2.4 Using the color of the trademark to have a certain regional characteristic, obtaining a color connected domain other than the background color in the trademark image as an effective connected domain;
  • Step 2.5 Sort and number all valid connected domains according to the area from large to small, and use the effective connected domain of the top five as the anchor connected domain;
  • the top five valid connected domains are used as anchor connectivity domains for trademark positioning in subsequent test pictures. If the number of valid connected domains is less than 5, all valid connected domains are used as anchor connected domains.
  • Step 2.6 obtaining all valid connected domains according to the step 2.4, and establishing an SCCD feature for the trademark image;
  • the SCCD features include a connected domain prediction model and an effective connected domain pixel distribution histogram.
  • the former uses the color of each connected domain and its spatial positional relationship to represent the combined relationship between connected domains, while the latter represents the distribution of pixels in the effective connected domain of the trademark.
  • the two describe the layout of the trademark from different angles.
  • the connectivity domain prediction model feature is divided into two parts, namely, the anchor connection domain combination feature and the effective connected domain space set feature.
  • the feature of the anchor connected domain combination is described by the color-anchor connected domain combination set, that is, the arrangement and combination of the anchor connected domains under the same color, for example, the color of the trademark picture shown in Figure 4 (C) -
  • the anchor connected domain combination set can be described as:
  • the circumscribed rectangle of the connected domain of an anchor point is the smallest rectangular frame in which the connected domain of the anchor point is closely included, and the identifier of the circumscribed rectangular frame of the connected domain of the anchor point corresponds to the number of the connected domain of the anchor point.
  • the identifier of the circumscribed rectangle of the anchor connection domain 1 can be taken as R1, as shown in FIG.
  • the boundary point is the sum of the boundary points of all the connected points of the anchor points in the combination of each anchor point
  • the centroid is the center point of the circumscribed rectangle corresponding to the combination of the connected points of the anchor point, the distance The Euclidean distance between each boundary point and the center point of the corresponding circumscribed rectangle, the angle being the angle between each boundary point and the center point of the corresponding circumscribed rectangle.
  • the dimension of the boundary point-centroid distance angle histogram can be determined according to actual needs.
  • the present invention establishes a dimension of 10 X 12 (that is, the dimension of the distance is 10, the angle The boundary point of the dimension is 12) - the centroid distance angle histogram.
  • the longest Euclidean distance obtained can be divided into 10 equal parts, and all calculations are obtained.
  • the Euclidean distance is divided by the length of 10 equal parts to obtain the region where the Euclidean distance is in the distance direction of the histogram; dividing all the calculated angles by 30°, the angle is obtained in the angle direction of the histogram In the region where it is located, by finally counting the pixels contained in each region, a boundary point-centroid distance angle histogram with a dimension of 10 X 12 can be obtained.
  • the effective connected fields of the same color may be glued, and the anchor connection domain combination can handle this situation;
  • the second part, the feature of the effective connected domain space set uses the circumscribed rectangular frame - Anchor connection domain relationship To describe, that is, the relationship between the circumscribed rectangle of the anchor connected domain and the connected domain of the anchor included therein.
  • the effective connected domain pixel distribution histogram is used to filter trademark pre-positioning regions (LPRs), and the establishment of the effective connected domain pixel distribution histogram specifically includes the following steps:
  • Step 2.6.1 as shown in Figure 4 (d), divide the trademark image into 4 segments evenly;
  • Step 2.6.2 divide each segment into 8 segments (bm) horizontally and vertically.
  • the pixels in the valid connected domain of the trademark image fall into the corresponding bm according to their horizontal and vertical coordinate positions, and form two 8-dimensional pixel distribution histograms, and the above operations are sequentially performed for each block, and finally a total of 8 blocks is obtained.
  • a 8-dimensional pixel distribution histogram As shown in Figure 4 (d), divide the trademark image into 4 segments evenly;
  • Step 2.6.2 divide each segment into 8 segments (bm) horizontally and vertically.
  • the pixels in the valid connected domain of the trademark image fall into the corresponding bm according to their horizontal and vertical coordinate positions, and form two 8-dimensional pixel distribution histograms, and the above operations are sequentially performed for each block, and finally a total of 8 blocks is obtained.
  • a 8-dimensional pixel distribution histogram
  • Step 3 input a test picture containing the target trademark, and establish a connected domain descriptor (CCD) feature for the test picture;
  • CCD connected domain descriptor
  • the steps to establish the Connected Field Descriptor (CCD) feature of the test picture include the following steps:
  • Step 3.1 HSV spatial conversion, color quantization and color smoothing of the test image in the same way as the trademark image sample (as described in Step 2.1-Step 2.3);
  • Step 3.2 according to the step 2.4, using the color of the trademark to have a certain regional characteristic, obtaining a color connected domain other than the background color in the test picture as a main connected domain;
  • Step 3.3 step 3.2
  • the obtained connected domain performs the connected domain segmentation of the gray space to obtain an effective connected domain of the test picture;
  • the step 3.3 further includes the following steps:
  • Step 3.3.1 transforming the test picture into a gray space
  • the conversion of the gray space is a general technique, and will not be described here.
  • Step 3.3.2 detecting the connected domain area of the gray connected space in the main connected domain, if the area of a main connected domain is greater than a certain threshold (such as 100) and the variance of the gray value is greater than a certain value (such as 50) , using the Otsu method to perform gray space segmentation on the main connected domain to obtain two divided sub-connected domains;
  • a certain threshold such as 100
  • a certain value such as 50
  • the Otsu method is divided into general-purpose technologies in the field.
  • Step 3.3.3 if the average gray value difference between the two sub-connected domains is greater than a certain threshold (for example, 50), the two connected domains obtained after the segmentation are used to cover the main connected domain before the segmentation;
  • a certain threshold for example, 50
  • Step 3.3.4 divide all the main connected domains according to step 3.3.2 and step 3.3.3, until they can no longer be divided, and then remove the connected domain whose area is too small (such as less than 30). The next is the valid connected domain of the test picture;
  • Step 3.4 Calculate the CCD characteristics of each valid connected domain of the test picture according to the step 2.6, including color, area, circumscribed rectangular frame and boundary point-centroid distance angle histogram (the process is similar to step 2.5).
  • Step 4 obtaining the trademark pre-positioning area LPR in the test picture by using the SCCD feature of the trademark image obtained in the step 2;
  • the step 4 is divided into two steps:
  • Step 4.1 Searching the trademark pre-positioning area LPRs in the test picture based on the prediction model;
  • FIG. 5 is a schematic diagram of searching for LPRs based on the prediction model, and the step 4.1 specifically includes the following steps:
  • Step 4.1.1 calculate the effective connection between the anchor connection domain combination and the test image of the trademark image sample in the trademark image sample library under the same color.
  • the difference Si 0? te , / ⁇ ) and the effective connected domain in the test picture and the anchor connection domain combination of the trademark image sample correspond to the difference between the width ratio and the high ratio value 5 2 ? te , ),
  • Step 4.1.2 calculating the similarity S 3 (h tc , h lc ) of the boundary point-centroid distance angle histogram of the test picture and the trademark picture sample determined to be similar by the step 4.2 according to the formula (3), If 5 3 ( i te , i ie ) is greater than a threshold, then the two are considered to match, and a matching picture pair is obtained:
  • Step 4.1.3 using the relative position of the anchor connection domain of the trademark image sample in the sample of the trademark image and the area ratio between the matching image pairs obtained in the step 4.2, obtaining the test picture connected to the anchor point
  • the rectangular frame of the area where the valid connected domain of the domain matches, and the rectangular frame of the area is used as the trademark pre-positioning area LPR in the test picture.
  • Step 4.2 Filter the searched trademark pre-positioning area LPRs based on the pixel distribution histogram.
  • the step 4.2 further includes the following steps:
  • Step 4.2.1 Calculate the area coincidence between the two LPRs according to formula (4) for all obtained trademark pre-positioning regions LPRs:
  • 5 ⁇ 5 ⁇ is the area of each of the two LPRs
  • r t and ⁇ are the ratios of the two LPR coincident regions to the two LPRs respectively
  • ⁇ 4 is The threshold can be taken as 0.9.
  • Step 4.2.2 sequentially performing the following operations on each LPR obtained through the step 4.2.1:
  • the trademark image sample is effectively connected to all the external rectangular frames in the domain space set. Projected into the LPR, if more than 80% of the area of a connected domain in the LPR is in a circumscribed rectangle, the connected domain is considered to be a suspected trademark connected domain. In this way, all the suspected trademark connected domains in the LPR are obtained;
  • Step 4.2.3 according to the steps 2.6.1-2.6.3, the validity of the suspected trademark connected domain obtained in the step 4.2.2 is valid. Pixel distribution histogram;
  • Step 4.2.4 performing the similarity matching test on the effective pixel distribution histogram obtained in step 4.2.3 and the effective pixel distribution histogram of the trademark image sample by using the Euclidean distance as the similarity measure to satisfy a certain similarity degree.
  • the threshold required LPR is retained for subsequent trademark matching.
  • Step 5 using the feature established based on the trademark color and the shape information, the trademark pre-positioning area obtained in the step 4 is matched with the trademark image in the trademark photo database, and the matching trademark pre-positioning area is tested. The final detected mark in the picture.
  • the color and shape characteristics of the trademark are the most prominent features of the trademark.
  • the present invention mainly applies the two features of a block color histogram and a color boundary gradient symbiotic histogram (CEGCH).
  • CEGCH color boundary gradient symbiotic histogram
  • step 5 further includes the following steps:
  • Step 5.1 after obtaining all the LPRs, respectively extracting a block color histogram and a color boundary gradient symbiotic histogram of the trademark image in the LPR and the trademark image database; and extracting the reference of the block color histogram "Color indexing, MJ Swain And DH Ballard, International Journal of Computer Vision 7 (1), pp. 11-32, 1991" and the block mode shown in Figure 6; the extraction of the color boundary gradient symbiotic histogram reference "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" get on.
  • Step 5.2 referring to the descriptions of the block color histogram and the CEGCH similarity in the above two documents, the block color histogram similarity ( ⁇ , ⁇ ) of the trademark image in the LPR and the trademark image database is obtained.
  • Step 5.3 based on the calculated ⁇ : e , CJ and ⁇ : H e , CHJ calculates the comprehensive similarity of the LPR and the trademark image in the trademark image database ⁇ , :
  • LPR, 0 ⁇ / ⁇ 1.0, ⁇ C ⁇ C is the block color histogram similarity between the target trademark and LPR
  • C e represents the block color histogram of the target trademark
  • ⁇ represents the block color histogram of LPR
  • CEGCH similarity between the target trademark and LPR CEGCH of the target trademark of CH ⁇ , CEGCH 0 representing LPR
  • the LPR is the final detected trademark area, that is, the position with the highest similarity is considered to be the most likely trademark position.
  • the performance evaluation of the method of the present invention is carried out by verifying whether or not the test image contains a trademark.
  • the method of the invention is for different trademarks (Starbucks (S), Pepsi ( ⁇ ), Baidu ( ⁇ ), The average detection time of cattle (M), HSBC (H) and China Netcom (C) is shown in Table 1.
  • Figure 7 shows the curve of the test time as a function of the size (area) of the image to be measured. It can be seen from the curve that as the area increases, the pre-positioning and overall detection time increase steadily.
  • Figure 8 shows the accuracy-recall rate curve.
  • the three curves are the accuracy-recall rate curve of the SCCD pre-position and CEGCH feature detection mark in the original size database.
  • the SCCD pre-positioning and CEGCH features are used in the downsampling database (256x384). Detecting the accuracy of the trademark-recall rate curve, using the sliding window detection (OSW) and CEGCH features in the downsampling database (256x384) to detect the accuracy-recall rate curve of the trademark.
  • OSW sliding window detection
  • CEGCH features in the downsampling database
  • the method of the present invention uses SCCD pre-positioning and sliding window detection in the downsampling database (256x384), and the comparison table of the total number of detection errors and detection errors in the original image database using SCCD is shown in Table 2, where So indicates that the SCCD is pre-positioned.
  • Detection in the original image database Ss indicates detection using the SCCD pre-position in the downsampling database, and Os indicates detection using the sliding window in the downsampling database:
  • Figure 9 shows some test results in the image database to be tested using the present invention, from the results It can be seen that the method of the invention has certain robustness to factors such as scale, occlusion, illumination variation, rotation and the like.
  • the method proposed by the present invention improves the speed and accuracy of trademark detection and location recognition according to the characteristics of the trademark, and can be used for real-time monitoring of network advertisements.
  • the present invention proposes a new rapid mark detection and recognition method, which proposes an algorithm for quickly detecting and locating a trademark pre-position by combining a trademark connected domain and its spatial relationship features. You can quickly locate a suspected trademark location for brand matching.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)

Abstract

La présente invention concerne un procédé de détection de marques déposées basé sur un pré-emplacement de composantes associées dans l'espace. Le procédé comprend les étapes suivantes consistant à : établir une base de données d'échantillons d'images de marques déposées contenant de multiples images de marques déposées ; établir des caractéristiques des descripteurs des composantes associées dans l'espace (SCCD) pour les images de marques déposées dans la base de données d'échantillons d'images de marques déposées ; pour des images tests contenant des marques déposées cibles, établir des caractéristiques des descripteurs des composantes associées (CCD) des images tests ; à l'aide des caractéristiques SCCD des images de marques déposées dans la base de données d'échantillons d'images de marques déposées, obtenir des régions de pré-emplacements de marques déposées (LPR) dans les images tests ; et, en utilisant les caractéristiques qui sont établies sur la base des informations sur les formes et les couleurs des marques déposées, mettre en correspondance les LPR et les images de marques déposées dans la base de données d'échantillons d'images de marques déposées, les régions de pré-emplacements de marques déposées mises en correspondance avec succès étant les marques déposées finalement détectées. La présente invention porte entièrement sur les caractéristiques des marques déposées. La vitesse et la précision de la détection des marques déposées ainsi que l'identification des emplacements s'en trouvent encore améliorées. De plus, la détection des marques déposées, qui sont influencées par une modification de l'échelle, une modification de l'angle visuel, un éclairage, une occlusion, une interférence d'arrière-plan et autres, peut avoir un effet satisfaisant.
PCT/CN2012/077884 2012-06-29 2012-06-29 Procédé de détection de marques déposées basé sur un pré-emplacement de composantes associées dans l'espace WO2014000261A1 (fr)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2012/077884 WO2014000261A1 (fr) 2012-06-29 2012-06-29 Procédé de détection de marques déposées basé sur un pré-emplacement de composantes associées dans l'espace

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2012/077884 WO2014000261A1 (fr) 2012-06-29 2012-06-29 Procédé de détection de marques déposées basé sur un pré-emplacement de composantes associées dans l'espace

Publications (1)

Publication Number Publication Date
WO2014000261A1 true WO2014000261A1 (fr) 2014-01-03

Family

ID=49782106

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2012/077884 WO2014000261A1 (fr) 2012-06-29 2012-06-29 Procédé de détection de marques déposées basé sur un pré-emplacement de composantes associées dans l'espace

Country Status (1)

Country Link
WO (1) WO2014000261A1 (fr)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108846843A (zh) * 2018-04-03 2018-11-20 南昌奇眸科技有限公司 一种图像特征提取方法
US20180373960A1 (en) * 2015-12-15 2018-12-27 Qing Xu Trademark graph element identification method, apparatus and system, and computer storage medium
CN109643318A (zh) * 2016-04-14 2019-04-16 艾德计算机运作与贸易股份公司 商标图像的基于内容的搜索和检索
CN110766007A (zh) * 2019-10-28 2020-02-07 深圳前海微众银行股份有限公司 证件遮挡检测方法、装置、设备及可读存储介质
CN111160073A (zh) * 2018-11-08 2020-05-15 浙江宇视科技有限公司 车牌类型识别方法、装置及计算机可读存储介质
CN111985508A (zh) * 2020-09-01 2020-11-24 安徽萤瞳科技有限公司 一种适用于线阵ccd的目标连通域形状分析方法
CN112505278A (zh) * 2020-11-30 2021-03-16 深圳市联正通达科技有限公司 一种抽样式污水监控分析设备
CN116485797A (zh) * 2023-06-25 2023-07-25 潍坊亚贝涂料有限公司 基于人工智能的涂料色差快速检测方法
CN117218549A (zh) * 2023-11-07 2023-12-12 山东力加力钢结构有限公司 一种公路桥梁的状态评估方法及系统

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101246551A (zh) * 2008-03-07 2008-08-20 北京航空航天大学 一种快速的车牌定位方法
CN102799653A (zh) * 2012-06-29 2012-11-28 中国科学院自动化研究所 一种基于空间连通域预定位的商标检测方法

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101246551A (zh) * 2008-03-07 2008-08-20 北京航空航天大学 一种快速的车牌定位方法
CN102799653A (zh) * 2012-06-29 2012-11-28 中国科学院自动化研究所 一种基于空间连通域预定位的商标检测方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
ZHANG, YUAN ET AL.: "Spatial connected component pre-locating algorithm for rapid logo detection", IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING, ICASSP 2012, 25 March 2012 (2012-03-25), pages 1297 - 1300, XP032227369, DOI: doi:10.1109/ICASSP.2012.6288127 *

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180373960A1 (en) * 2015-12-15 2018-12-27 Qing Xu Trademark graph element identification method, apparatus and system, and computer storage medium
US10430687B2 (en) * 2015-12-15 2019-10-01 Qing Xu Trademark graph element identification method, apparatus and system, and computer storage medium
CN109643318A (zh) * 2016-04-14 2019-04-16 艾德计算机运作与贸易股份公司 商标图像的基于内容的搜索和检索
CN108846843A (zh) * 2018-04-03 2018-11-20 南昌奇眸科技有限公司 一种图像特征提取方法
CN111160073B (zh) * 2018-11-08 2023-09-19 浙江宇视科技有限公司 车牌类型识别方法、装置及计算机可读存储介质
CN111160073A (zh) * 2018-11-08 2020-05-15 浙江宇视科技有限公司 车牌类型识别方法、装置及计算机可读存储介质
CN110766007B (zh) * 2019-10-28 2023-09-22 深圳前海微众银行股份有限公司 证件遮挡检测方法、装置、设备及可读存储介质
CN110766007A (zh) * 2019-10-28 2020-02-07 深圳前海微众银行股份有限公司 证件遮挡检测方法、装置、设备及可读存储介质
CN111985508B (zh) * 2020-09-01 2023-09-19 安徽萤瞳科技有限公司 一种适用于线阵ccd的目标连通域形状分析方法
CN111985508A (zh) * 2020-09-01 2020-11-24 安徽萤瞳科技有限公司 一种适用于线阵ccd的目标连通域形状分析方法
CN112505278A (zh) * 2020-11-30 2021-03-16 深圳市联正通达科技有限公司 一种抽样式污水监控分析设备
CN112505278B (zh) * 2020-11-30 2024-05-17 福建钱江水务有限公司 一种抽样式污水监控分析设备
CN116485797A (zh) * 2023-06-25 2023-07-25 潍坊亚贝涂料有限公司 基于人工智能的涂料色差快速检测方法
CN116485797B (zh) * 2023-06-25 2023-08-29 潍坊亚贝涂料有限公司 基于人工智能的涂料色差快速检测方法
CN117218549A (zh) * 2023-11-07 2023-12-12 山东力加力钢结构有限公司 一种公路桥梁的状态评估方法及系统
CN117218549B (zh) * 2023-11-07 2024-03-05 山东力加力钢结构有限公司 一种公路桥梁的状态评估方法及系统

Similar Documents

Publication Publication Date Title
WO2014000261A1 (fr) Procédé de détection de marques déposées basé sur un pré-emplacement de composantes associées dans l'espace
Zhang et al. Towards reaching human performance in pedestrian detection
CN104166685B (zh) 一种检测视频片段的方法和装置
US7689044B2 (en) Method and apparatus for representing a group of images
CN108549874B (zh) 一种目标检测方法、设备及计算机可读存储介质
US8805117B2 (en) Methods for improving image search in large-scale databases
WO2020125216A1 (fr) Procédé et dispositif de ré-identification de piéton, dispositif électronique et support d'informations lisible par ordinateur
WO2020082577A1 (fr) Procédé de vérification anti-contrefaçon de sceau, dispositif et support d'informations lisible par ordinateur
CN107622489B (zh) 一种图像篡改检测方法及装置
US20140092244A1 (en) Object search method, search verification method and apparatuses thereof
WO2020252974A1 (fr) Procédé et dispositif de suivi de multiples objets cibles dans un état de mouvement
CN102799653A (zh) 一种基于空间连通域预定位的商标检测方法
Ren et al. Fusion of intensity and inter-component chromatic difference for effective and robust colour edge detection
US8942469B2 (en) Method for classification of videos
CN102176208A (zh) 基于三维空时特征的鲁棒视频指纹方法
CN103530638A (zh) 多摄像头下的行人匹配方法
WO2017193701A1 (fr) Procédé de détection de la chute d'une bicyclette partagée sur le sol
WO2014036813A1 (fr) Procédé et dispositif d'extraction de traits caractéristiques d'images
CN106599880A (zh) 一种面向无人监考的同人判别方法
Lin et al. Convolutional neural networks for face anti-spoofing and liveness detection
Liang et al. A new wavelet-Laplacian method for arbitrarily-oriented character segmentation in video text lines
CN105590086A (zh) 一种基于视觉标签识别的物品防盗检测方法
KR101711328B1 (ko) 영상 촬영 장치를 통하여 획득한 영상 이미지에서 신장 대비 머리 높이의 비를 이용하여 어린이와 성인을 구분하는 방법
CN109544614B (zh) 一种基于图像低频信息相似度的匹配图像对识别的方法
CN116665321A (zh) 一种基于边缘纳管技术的停车场车辆管理方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 12879742

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 12879742

Country of ref document: EP

Kind code of ref document: A1