Nothing Special   »   [go: up one dir, main page]

CN109284751A - A non-text filtering method based on spectral analysis and SVM for text localization - Google Patents

A non-text filtering method based on spectral analysis and SVM for text localization Download PDF

Info

Publication number
CN109284751A
CN109284751A CN201811281682.2A CN201811281682A CN109284751A CN 109284751 A CN109284751 A CN 109284751A CN 201811281682 A CN201811281682 A CN 201811281682A CN 109284751 A CN109284751 A CN 109284751A
Authority
CN
China
Prior art keywords
text
svm
spectrogram
frequency
preliminary positioning
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
Application number
CN201811281682.2A
Other languages
Chinese (zh)
Inventor
霍华
桂洋
吕靖
杜琰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Henan University of Science and Technology
Original Assignee
Henan University of Science and Technology
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 Henan University of Science and Technology filed Critical Henan University of Science and Technology
Priority to CN201811281682.2A priority Critical patent/CN109284751A/en
Publication of CN109284751A publication Critical patent/CN109284751A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The non-textual filtering method of text location based on spectrum analysis and SVM, comprising the following steps: file building data set is stored in the picture for needing to carry out text location and is used；Primary Location is carried out using picture of the method for candidate edge recombination and marginal classification to storage, obtains Primary Location region；By Primary Location region binaryzation, and Gray Projection in the vertical direction；The one-dimension array obtained after step 3 Gray Projection is transformed into frequency domain by Fast Fourier Transform (FFT) and obtains the spectrogram in Primary Location region, the characteristics of spectrogram is analyzed；Non-textual filter out is carried out to Primary Location region using SVM classifier.The invention has the advantages that: the accuracy rate of text location is higher.

Description

The non-textual filtering method of text location based on spectrum analysis and SVM
Technical field
The present invention relates to intelligent information processing technology fields, are related to the relevant technologies such as image procossing and artificial intelligence, specifically Ground say be the text location based on spectrum analysis and SVM non-textual filtering method.
Background technique
With the fast development of people's daily life, either people's mobile phone, camera shooting natural scene image, also It is the picture seen when browsing Web news web page, they all include a large amount of highly important text informations.When we think at once The text conversion in image at editable text information, then the text location in image becomes essential premise Link.So the text location in image efficiently and is accurately become extremely important to obtain these text informations just.
But due to the continuous development of information, the content and quality of picture also change constantly.Due to the size of picture, packet Various fonts containing text, line orientations, different lighting conditions, the factors such as weak character and complicated background in picture are quasi- True is still a challenge and a vital task by the text location in picture.However text is positioned as needing in picture The worker for big data analysis data are largely grabbed from network provides convenience, pushes the development of correlative study, is simultaneously Many forward position focus such as artificial intelligence provide relevant technical support.
Existing text location method has much, but the non-textual method filtered out is not very much, to sum up usually Two classes can be divided into: the method based on geometric attribute and text structure and the method based on machine learning classification, to be based on geometry The method of attribute and text structure is most widely used.In method wherein based on geometric attribute and design feature, by text The structural analysis of stroke is combined with color assignment, and non-textual region is eliminated in the methods of wiping out background interference etc..This kind of side Method is all simple and is easy to carry out, but the effect when disturbing factor is excessively complicated is not very good.
With the continuous development of research method, start the recurrent neural network for proposing that there is long-term short-term memory (LSTM) (RNN) to be identified by the feature that CNN is extracted from entire picture, this method be pass through explore contextual information and avoid by Mistake caused by dividing.There are also proposing that removing algorithm according to node is decomposed into simple assemblies for complex assemblies, then non-intersecting Skeleton combination on spliced corresponding text filed to obtain, finally use the categorizing system based on HMM to carry out false positive It filters out.Although these method accuracys rate are higher, carry out excessively complexity, and do not have wide applicability.
Summary of the invention
Non-textual technical problem to be solved by the invention is to provide the text location based on spectrum analysis and SVM filters out Method solves the problems, such as that current localization method effect is undesirable, improves the accuracy rate of text location.
Used technical solution is the present invention to solve above-mentioned technical problem: the text based on spectrum analysis and SVM is fixed The non-textual filtering method of position, comprising the following steps:
Step 1 is stored in file building data set use to the picture for needing to carry out text location；
Step 2 carries out Primary Location using the picture that the method for candidate edge recombination and marginal classification stores step 1, obtains Primary Location region；
Step 3, by Primary Location region binaryzation, and Gray Projection in the vertical direction；
Step 4, by Fast Fourier Transform (FFT) by the one-dimension array obtained after step 3 Gray Projection be transformed into frequency domain obtain just The spectrogram for walking localization region, analyzes the characteristics of spectrogram；
Step 5 carries out non-textual filter out to Primary Location region using SVM classifier；
Step 6 calculates locating accuracy.
The specific of Gray Projection is carried out to Primary Location region binaryzation in step 3 of the present invention and on numerical value direction Step are as follows:
The Primary Location area image that step 2 obtains is converted into the gray level image convenient for image procossing by step 3.1；
Step 3.2 carries out noise-filtering to the gray level image that step 3.1 obtains；
Step 3.3 is analyzed and processed the processed gray level image of step 3.2 to obtain its global threshold T, then by grayscale image Pixel of the gray value less than T value of the point after threshold process is set as 1 as in, otherwise the value of the point is set as after threshold process 0；
Step 3.4, the projection that the binary image obtained after step 3.3 processing is made to vertical direction, the black to add up on each column Pixel obtains its number.
Frequency domain is converted it to by Fast Fourier Transform (FFT) in step 4 of the present invention, the characteristics of to spectrogram into Row analysis method particularly includes:
The one-dimension array obtained after Gray Projection on vertical direction is carried out Fast Fourier Transform (FFT) by step 4.1, makes rows of week Phase signal is transformed into frequency domain and obtains the spectrogram in Primary Location region；
Step 4.2 obtains the spectrogram in whole Primary Location regions by above step, including text filed spectrogram With non-textual region spectrogram；
Spectrogram and corresponding Primary Location region are put on serial number by step 4.3, and a large amount of spectrogram is analyzed and summarized Non-textual and text difference characteristic parameters of spectra can be most represented out.
The non-textual specific method filtered out is carried out to Primary Location region using SVM classifier in step 5 of the present invention Are as follows:
Step 5.1, selected SVM classifier are as training tool；
Amplitude corresponding to frequency and frequency 12 arrive the magnitude peak between frequency 31 in step 5.2, selection frequency 2 to frequency n It is inputted as input feature vector value；
Step 5.3 sets the output of SVM as 2 one-dimension arrays；
Step 5.4, the functional relation for being determined by experiment output result.
The beneficial effects of the present invention are: present invention effectively removes leave after existing text location method Primary Location Non-legible region improves the accuracy rate of text location；This method applicability is than wide, not only to picture in news web page Non-legible filtration result is fine, also has preferable promotion to the non-legible filtration result of the picture in natural scene.
Detailed description of the invention
Fig. 1 is the flow chart of non-textual filtering method according to the present invention；
Fig. 2 is the original image in Primary Location region of the present invention；
Fig. 3 is the present invention image later to Primary Location area grayscaleization；
Fig. 4 is grayscale image of the present invention in vertical direction projection；
Fig. 5 is non-textual region of the invention and corresponding spectrogram；
Fig. 6 is the text filed and corresponding spectrogram of the present invention；
Fig. 7 is non-textual region of the invention and corresponding output result figure group；
Fig. 8 is the text filed and corresponding output result figure group of the present invention.
Specific embodiment
A specific embodiment of the invention (embodiment) is described with reference to the accompanying drawings of the specification, makes the skill of this field Art personnel better understood when the present invention.
The non-textual filtering method of text location based on spectrum analysis and SVM, comprising the following steps:
Step S1: it carries out encoding automatic refreshing Website page by Python and intercepts news picture automatically, obtain a certain amount of Picture is as first data set.The racing data collection (ICDAR2011) of downloading natural scene picture, which is stored in file, to be made For second data set.Two datasets are carried out respectively it is non-legible filter out, to obtain in news web page picture and natural field The non-legible filtration result of picture in scape.
Step S2: recombinating picture using candidate edge and the method for marginal classification carries out Primary Location, and it is tentatively fixed to obtain Position region.
Step S3: by the grayscale image binaryzation in Primary Location region, and Gray Projection in the vertical direction, specifically include with Lower step:
Step S31: Primary Location region is carried out first, in accordance with formula Gray=R × 0.299+G × 0.587+B × 0.114 The color image being collected into is converted into the gray level image convenient for image procossing by the color image gray processing of news or natural scene As shown in Figure 3.
Step S32: since the unnecessary information in gray level image will affect the corresponding spectrogram knot in Primary Location region Fruit, so to carry out noise-filtering to gray level image.
Step S33: setting global threshold T, then carry out thresholding to original image: all gray values are less than the pixel of T, The value of the point is 1 after threshold process；Conversely, all gray values are greater than the pixel of T, the value of the point is 0 after threshold process.
Step S34: the width of the bianry image in Primary Location region is set as W, a height of H, we make vertical direction to the region Projection, formula are as follows:
V (j) indicates that jth arranges the sum of upper black pixel point in formula, and NewPixel (i, j) indicates the value of each pixel, if The point is white then NewPixel (i, j)=0, and NewPixel (i, j)=1:j indicates pixel place if the point is black Column, and 0≤j < W:i indicates the row where the pixel.The grayscale image of vertical direction projection is as shown in Figure 4.
Step S4: converting it to frequency domain by Fast Fourier Transform (FFT), analyzes the characteristics of spectrogram, specifically Method are as follows:
Step S41: the one-dimension array obtained after Gray Projection on vertical direction is subjected to Fast Fourier Transform (FFT), makes rows of week Phase signal is transformed into frequency domain and obtains the spectrogram in Primary Location region.
Step S42: obtaining the spectrogram in whole Primary Location regions by above step, including non-textual region and Its spectrogram (as shown in Figure 5) and the text filed spectrogram (as shown in Figure 6) with it.
Step S43: analyzing the spectrogram in a large amount of non-textual region, and a large amount of text filed spectrogram, It was found that frequency text filed and non-textual difference from 0 to 1 is little；Have from frequency 2 to amplitude frequency 30 more apparent floating Dynamic, the frequency for summing up letter signal is approximately equal to the physical width of text.
Step S431: making marks to Primary Location region, and true positives zone marker is (1,0), by false positive zone marker For (0,1), Experimental comparison sums up the characteristic parameters of spectra that can most represent false positive and non-false positive difference.
Step S5: use SVM classifier as training tool, selecting frequency 2 to frequency n, (the two of average text width arrive Three times) amplitude and frequency 12 corresponding to interior frequency to the magnitude peak between frequency 31 be also used as input feature vector value to input, and it is defeated Out it is 2 one-dimension arrays:
(y11,y21,…,yi1), (y12,y22,…,yi2)；
The corresponding output result in text filed and non-textual region is respectively such as Fig. 7, shown in Fig. 8.It is determined by experiment output result Functional relation, such as export result meet wi=yi1-yi2 > Δ t be it is text filed, meet wi=yi1-y21≤Δ t is then non- It is text filed.
Step S6: accuracy rate, recall rate and F measure formulas after calculating non-textual filter out are respectively as follows:
P in formula, R, F are accuracy rate, recall rate and F value respectively；n1It is the quantity for the text block being correctly detecting, m1It is non-text The quantity of this block, m2It is the quantity for losing text block.

Claims (4)

1.基于频谱分析和SVM的文字定位的非文本滤除方法，其特征在于：包括以下步骤：1. the non-text filtering method based on the character location of spectrum analysis and SVM, is characterized in that: comprise the following steps: 步骤1、对需要进行文字定位的图片存入文件构建数据集用；Step 1. Save the pictures that need text positioning into the file to construct a data set; 步骤2、使用候选边缘重组和边缘分类的方法对步骤1存储的图片进行初步定位，得到初步定位区域；Step 2. Use the method of candidate edge reorganization and edge classification to perform preliminary positioning on the picture stored in step 1 to obtain a preliminary positioning area; 步骤3、将初步定位区域二值化，并在竖直方向上灰度投影；Step 3. Binarize the preliminary positioning area and project the grayscale in the vertical direction; 步骤4、通过快速傅里叶变换将步骤3灰度投影后得到的一维数组转换到频率域得到初步定位区域的频谱图，对频谱图的特点进行分析；Step 4. Convert the one-dimensional array obtained after the grayscale projection in step 3 into the frequency domain through fast Fourier transform to obtain a spectrogram of the preliminary positioning area, and analyze the characteristics of the spectrogram; 步骤5、使用SVM分类器对初步定位区域进行非文本滤除；Step 5. Use the SVM classifier to filter out the non-text of the preliminary positioning area; 步骤6、计算定位准确率。Step 6: Calculate the positioning accuracy. 2.根据权利要求1所述的基于频谱分析和SVM的文字定位的非文本滤除方法，其特征在于：所述步骤3中对初步定位区域二值化并在数值方向上进行灰度投影的具体步骤为：2. the non-text filtering method based on the character location of spectrum analysis and SVM according to claim 1, it is characterized in that: in described step 3, to preliminary location area binarization and carry out grayscale projection in numerical value direction. The specific steps are: 步骤3.1、将步骤2得到的初步定位区域图像转换成便于图像处理的灰度图像；Step 3.1, convert the preliminary positioning area image obtained in step 2 into a grayscale image that is convenient for image processing; 步骤3.2、对步骤3.1得到的灰度图像进行噪音滤除；Step 3.2, perform noise filtering on the grayscale image obtained in step 3.1; 步骤3.3、对步骤3.2处理过的灰度图像进行分析处理得到其全局阈值T，然后将灰度图像中灰度值小于T的像素点经阈值处理后该点的值设为1，反之经阈值处理后该点的值设为0；Step 3.3, analyze and process the grayscale image processed in step 3.2 to obtain its global threshold T, and then set the value of the pixel whose grayscale value is less than T in the grayscale image to 1 after thresholding, and vice versa. After processing, the value of this point is set to 0; 步骤3.4、将步骤3.3处理后得到的二值化图像作垂直方向的投影，累加各列上的黑色像素点得到其个数。Step 3.4: Project the binarized image obtained after the processing in step 3.3 in the vertical direction, and accumulate the black pixels on each column to obtain the number thereof. 3.根据权利要求1所述的基于频谱分析和SVM的文字定位的非文本滤除方法，其特征在于：所述步骤4中通过快速傅里叶变换将其转换到频率域，对频谱图的特点进行分析的具体方法为：3. the non-text filtering method based on the character location of spectrum analysis and SVM according to claim 1, is characterized in that: in described step 4, it is converted into frequency domain by fast Fourier transform, to the frequency domain of spectrogram. The specific methods for analyzing the characteristics are as follows: 步骤4.1、将竖直方向上灰度投影后得到的一维数组进行快速傅里叶变换，使成行的周期信号转换到频率域得到初步定位区域的频谱图；Step 4.1. Perform fast Fourier transform on the one-dimensional array obtained after grayscale projection in the vertical direction, so that the periodic signal in a row is converted into the frequency domain to obtain the spectrogram of the preliminary positioning area; 步骤4.2、通过以上步骤得到全部初步定位区域的频谱图，其中包括文本区域的频谱图和非文本区域频谱图；Step 4.2, obtain the spectrograms of all the preliminary positioning areas through the above steps, including the spectrograms of the text area and the non-text area spectrograms; 步骤4.3、将频谱图和对应的初步定位区域标上序号，对大量的频谱图进行分析并总结出最能代表非文本和文本差异的频谱特征值。Step 4.3: Mark the spectrogram and the corresponding preliminary positioning area with serial numbers, analyze a large number of spectrograms, and summarize the spectral feature values that can best represent the difference between non-text and text. 4.根据权利要求1所述的基于频谱分析和SVM的文字定位的非文本滤除方法，其特征在于：所述步骤5中使用SVM分类器对初步定位区域进行非文本滤除的具体方法为：4. the non-text filtering method based on the character location of spectrum analysis and SVM according to claim 1, it is characterized in that: in described step 5, use SVM classifier to carry out the concrete method of non-text filtering to preliminary positioning area as : 步骤5.1、选定SVM分类器作为训练手段；Step 5.1, select the SVM classifier as the training method; 步骤5.2、选择频率2到频率n内频率所对应的幅值和频率12到频率31之间的幅值峰值作为输入特征值输入；Step 5.2, select the amplitude corresponding to the frequency in the frequency 2 to the frequency n and the amplitude peak value between the frequency 12 and the frequency 31 as the input characteristic value input; 步骤5.3、设定SVM的输出为2个一维数组；Step 5.3, set the output of SVM as 2 one-dimensional arrays; 步骤5.4、通过实验确定输出结果的函数关系。Step 5.4: Determine the functional relationship of the output results through experiments.
CN201811281682.2A 2018-10-31 2018-10-31 A non-text filtering method based on spectral analysis and SVM for text localization Pending CN109284751A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811281682.2A CN109284751A (en) 2018-10-31 2018-10-31 A non-text filtering method based on spectral analysis and SVM for text localization

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811281682.2A CN109284751A (en) 2018-10-31 2018-10-31 A non-text filtering method based on spectral analysis and SVM for text localization

Publications (1)

Publication Number Publication Date
CN109284751A true CN109284751A (en) 2019-01-29

Family

ID=65174664

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811281682.2A Pending CN109284751A (en) 2018-10-31 2018-10-31 A non-text filtering method based on spectral analysis and SVM for text localization

Country Status (1)

Country Link
CN (1) CN109284751A (en)

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20000060673A (en) * 1999-03-18 2000-10-16 이준환 Method of extracting caption regions and recognizing character from compressed news video image
CN102144236A (en) * 2008-09-03 2011-08-03 索尼公司 Text localization for image and video OCR
CN102208023A (en) * 2011-01-23 2011-10-05 浙江大学 Method for recognizing and designing video captions based on edge information and distribution entropy
CN102547147A (en) * 2011-12-28 2012-07-04 上海聚力传媒技术有限公司 Method for realizing enhancement processing for subtitle texts in video images and device
CN103136523A (en) * 2012-11-29 2013-06-05 浙江大学 Arbitrary direction text line detection method in natural image
CN105825216A (en) * 2016-03-17 2016-08-03 中国科学院信息工程研究所 Method of locating text in complex background image
CN107145888A (en) * 2017-05-17 2017-09-08 重庆邮电大学 Method for real-time translation of video subtitles
CN107945125A (en) * 2017-11-17 2018-04-20 福州大学 It is a kind of to merge spectrum estimation method and the fuzzy image processing method of convolutional neural networks
CN108241874A (en) * 2018-02-13 2018-07-03 河南科技大学 Video Text Region Location Method Based on BP Neural Network and Spectrum Analysis

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20000060673A (en) * 1999-03-18 2000-10-16 이준환 Method of extracting caption regions and recognizing character from compressed news video image
CN102144236A (en) * 2008-09-03 2011-08-03 索尼公司 Text localization for image and video OCR
CN102208023A (en) * 2011-01-23 2011-10-05 浙江大学 Method for recognizing and designing video captions based on edge information and distribution entropy
CN102547147A (en) * 2011-12-28 2012-07-04 上海聚力传媒技术有限公司 Method for realizing enhancement processing for subtitle texts in video images and device
CN103136523A (en) * 2012-11-29 2013-06-05 浙江大学 Arbitrary direction text line detection method in natural image
CN105825216A (en) * 2016-03-17 2016-08-03 中国科学院信息工程研究所 Method of locating text in complex background image
CN107145888A (en) * 2017-05-17 2017-09-08 重庆邮电大学 Method for real-time translation of video subtitles
CN107945125A (en) * 2017-11-17 2018-04-20 福州大学 It is a kind of to merge spectrum estimation method and the fuzzy image processing method of convolutional neural networks
CN108241874A (en) * 2018-02-13 2018-07-03 河南科技大学 Video Text Region Location Method Based on BP Neural Network and Spectrum Analysis

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
CHONG YU ET AL: "Text detection and recognition in natural scene with edge analysis", 《IET COMPUTER VISION》 *
MOHAMMAD KHODADADI AZADBONI ET AL: "Text detection and character extraction in color images using FFT domain filtering and SVM classification", 《6TH INTERNATIONAL SYMPOSIUM ON TELECOMMUNICATIONS (IST)》 *
PALAIAHNAKOTE SHIVAKUMARA ET AL: "New Fourier-Statistical Features in RGB Space for Video Text Detection", 《IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY》 *
傅泽田等: "《面向移动终端的农业信息智能获取》", 30 September 2015 *
孙红星等: "基于小波变换和SVM的文本区域定位", 《东北大学学报(自然科学版)》 *

Similar Documents

Publication Publication Date Title
Peyrard et al. ICDAR2015 competition on text image super-resolution
WO2018145470A1 (en) Image detection method and device
CN106228159A (en) A kind of gauge table meter copying device based on image recognition and method thereof
CN106548169B (en) Fuzzy literal Enhancement Method and device based on deep neural network
CN112232371B (en) American license plate recognition method based on YOLOv3 and text recognition
CN105718866B (en) A kind of detection of sensation target and recognition methods
CN112862849B (en) Image segmentation and full convolution neural network-based field rice ear counting method
Hati et al. Plant recognition from leaf image through artificial neural network
CN109740639A (en) A method, system and electronic device for detecting cloud in remote sensing image of Fengyun satellite
CN105335716A (en) Improved UDN joint-feature extraction-based pedestrian detection method
CN111738048B (en) Pedestrian re-identification method
CN110334703B (en) A method for ship detection and recognition in day and night images
CN104794479A (en) Method for detecting text in natural scene picture based on local width change of strokes
CN108805102A (en) A kind of video caption detection and recognition methods and system based on deep learning
CN108509950B (en) Railway contact net support number plate detection and identification method based on probability feature weighted fusion
Ma et al. Towards improved accuracy of UAV-based wheat ears counting: A transfer learning method of the ground-based fully convolutional network
CN111507967A (en) A high-precision detection method for mangoes in a natural orchard scene
CN111260645A (en) Tampered image detection method and system based on deep learning of block classification
CN115082776A (en) Electric energy meter automatic detection system and method based on image recognition
Feng et al. A novel saliency detection method for wild animal monitoring images with WMSN
CN115578364A (en) Weak target detection method and system based on mixed attention and harmonic factor
CN111414855B (en) Telegraph pole sign target detection and identification method based on end-to-end regression model
CN109460767A (en) Rule-based convex print bank card number segmentation and recognition methods
CN116188943A (en) Solar radio spectrum burst information detection method and device
CN102915449A (en) Photo classification method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20190129

RJ01 Rejection of invention patent application after publication