CN107844758A - Intelligence pre- film examination method, computer equipment and readable storage medium storing program for executing - Google Patents
Intelligence pre- film examination method, computer equipment and readable storage medium storing program for executing Download PDFInfo
- Publication number
- CN107844758A CN107844758A CN201711002884.4A CN201711002884A CN107844758A CN 107844758 A CN107844758 A CN 107844758A CN 201711002884 A CN201711002884 A CN 201711002884A CN 107844758 A CN107844758 A CN 107844758A
- Authority
- CN
- China
- Prior art keywords
- picture
- training
- feature
- caffenet
- models
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
Abstract
The invention discloses intelligence pre- film examination method, computer equipment and readable storage medium storing program for executing, it is related to technical field of video image processing, including：Obtain training pictures and checking pictures.Training picture in training pictures is pre-processed and marked, obtains training picture database.The average of training picture is calculated, and training picture is subtracted into average.The pre- review of a film by the censor training pattern caffenet models of intelligence are established under convolutional neural networks framework Caffe environment.It is trained using the training picture for subtracting average for caffenet models, the caffenet models after being trained.For the checking picture that checking picture is concentrated as the input picture of the caffenet models after training to being identified, recognition result is pre- review of a film by the censor result.The application employs image recognition sorting technique and technology based on deep learning, by there is the feature of supervision or unsupervised mode learning hierarchy to describe, so as to realize the pre- review of a film by the censor of intelligence.
Description
Technical field
The present invention relates to technical field of video image processing, and in particular to the pre- film examination method of intelligence, computer equipment and
Readable storage medium storing program for executing.
Background technology
Along with the progress of computer science and technology, the research and development and application of artificial intelligence are more and more deeply extensive, and a lot
Field and industry suffer from close combination, and progressively substitute the manually work in these fields or industry in future soon
Make, turn into following main economic stimuli.
Work is identified in video display industry or video image field, any one video or characteristics of image at present, is all
Completed by people.
Existing video image identification technology, it is necessary to largely train material and human cost, accuracy rate and discrimination it is low,
And the quality of model personnel Feature Engineering work is to determine the key of effect quality.
Due to not using artificial intelligence-depth learning technology, existing video image identification technology is limited only to existing
Set content is trained, lacks ability of self-teaching, so the image recognition work of cumbersome tera incognita complicated and changeable, nothing can not be adapted to
Method solves many practical problems of reality scene.
The content of the invention
It is an object of the invention to provide intelligence pre- film examination method, computer equipment and readable storage medium storing program for executing, based on AI
Technology, corresponding image recognition model is trained, solve some concrete application fields of video content in video display industry by this model
The feature recognition work of scape is simultaneously classified, judged according to identification content results, the work such as early warning, and be applied to video display expose the false,
Multiple reality scene fields such as film advertisement, movie and television contents Safety Examination.
To achieve the above object, technical scheme is as follows：
In a first aspect, the embodiment of the present application provides a kind of pre- film examination method of intelligence, including：
Obtain training pictures and checking pictures.
Training picture in training pictures is pre-processed and marked, obtains training picture database.
The average of training picture is calculated, and training picture is subtracted into average.
The pre- review of a film by the censor training pattern caffenet models of intelligence are established under convolutional neural networks framework Caffe environment.
It is trained using the training picture for subtracting average for the caffenet models, after being trained
Caffenet models；Caffenet models after the training are carried out after input picture is received to the input picture
Identification, exports the generic to input picture.
Input picture pair using the checking picture that the checking picture is concentrated as the caffenet models after the training
It is identified, the generic for exporting the input picture is pre- review of a film by the censor result.
In one or more embodiments of first aspect, the training picture in training pictures is pre-processed, wrapped
Include：
The filename for training picture, picture category, picture path are stored.
Adjusting training picture size is the uniform sizes of setting, and is converted to unified form.
In one or more embodiments of first aspect, the pre- review of a film by the censor training pattern caffenet models of intelligence are：
Low-level image feature extraction is carried out to input picture.
Feature coding is carried out to the low-level image feature extracted.
Space characteristics constraint is carried out to the low-level image feature after feature coding, obtains the description vectors of input picture.
The description vectors of image are classified using grader, obtain the recognition result of caffenet models.
In one or more embodiments of first aspect, low-level image feature extraction is carried out to the image of the pre- review of a film by the censor, including：
According to bottom of the local feature description in default compensation, yardstick extraction preliminary hearing picture as preliminary hearing picture
Feature.
In one or more embodiments of first aspect, it is special that local feature description includes scale invariant feature conversion SIFT
Sign, histograms of oriented gradients HOG features, local binary patterns LBP features.
In one or more embodiments of first aspect, feature coding is carried out to the low-level image feature extracted, including, adopt
Encoded with vector quantization, be in sparse coding, local linear constraint coding or Fisher vector coding a kind of to being extracted
Low-level image feature carries out feature coding；
In one or more embodiments of first aspect, space characteristics are carried out about to the low-level image feature after feature coding
Beam, the description vectors of image are obtained, including：Using the method for pyramid characteristic matching to entering to the low-level image feature after feature coding
Row space characteristics constrain, and obtain the description vectors of image.
In one or more embodiments of first aspect, grader is support vector machines or random forest.
Second aspect, the embodiment of the present application additionally provide a kind of computer equipment, including memory, processor and storage
On a memory and the computer program that can run on a processor, the application first aspect is realized during computing device program
The method that one or more embodiments provide.
The third aspect, the embodiment of the present application additionally provide a kind of computer-readable recording medium, are stored thereon with computer
Program, the method that one or more embodiments of the application first aspect provide is realized when computer program is executed by processor.
The inventive method has the following advantages that：
The conventional image recognition technology based on machine learning, it is difficult to which the image for adapting to cumbersome tera incognita complicated and changeable is known
Do not work, in this case, the application employs image recognition sorting technique and technology based on deep learning, by there is prison
Superintend and direct or the feature of unsupervised mode learning hierarchy describes, so as to realize the pre- review of a film by the censor of intelligence.The advantages of the application is exactly profit
Inputted with image pixel information, remain all information of input picture to the full extent, output is directly image recognition result.
Brief description of the drawings
Fig. 1 is the pre- review of a film by the censor method flow diagram of a kind of intelligence based on AI technologies that the embodiment of the present application provides；
Fig. 2 is the specific training step for the pre- review of a film by the censor training pattern caffenet models of intelligence that the embodiment of the present application provides；
Fig. 3 is the pre- film examination method of intelligence that embodiment is classified as with video display industry film.
Embodiment
Following examples are used to illustrate the present invention, but are not limited to the scope of the present invention.
Embodiment 1
Fig. 1 is refer to, Fig. 1 shows the pre- film examination method stream of a kind of intelligence based on AI technologies that the embodiment of the present application provides
Cheng Tu, including：
S101, obtain training pictures and checking pictures；It can be in film wherein to train picture and checking picture
Corresponding picture, training picture are used to be trained pre- review of a film by the censor training pattern, and checking picture term is instructed to the pre- review of a film by the censor trained
Practice model to be verified.
S102, the training picture in training pictures is pre-processed and manually marked, obtain training picture database；
Wherein the training picture in training pictures is pre-processed, including：
The filename for training picture, picture category, picture path are stored；
Adjusting training picture size is the uniform sizes of setting, and is converted to unified form.
Wherein artificial mark includes：
Artificial mark training picture generic.
S103, the average for calculating training picture, and training picture is subtracted into average；
S104, the pre- review of a film by the censor training pattern caffenet models of intelligence are established under convolutional neural networks framework Caffe environment；
S105, it is trained for caffenet models using the training picture for subtracting average, after being trained
Caffenet models；Caffenet models after training can export the institute to input picture after input picture is received
Belong to classification, as identify classification results.
S106, using verify picture concentrate checking picture as training after caffenet models input picture to progress
Identification, recognition result is the generic of input picture, as pre- review of a film by the censor result.
It refer to Fig. 2, Fig. 2 shows the specific of pre- review of a film by the censor training pattern caffenet models of intelligence in the embodiment of the present application
Training step, including
S201, the image to the pre- review of a film by the censor carry out low-level image feature extraction；Extracted largely according to fixed step size, yardstick from image
Local feature description.Conventional local feature includes SIFT, and (Scale-Invariant Feature Transform, yardstick is not
Become Feature Conversion), HOG (Histogram of Oriented Gradient, histograms of oriented gradients), LBP (Local
Bianray Pattern, local binary patterns) etc., typically also prevent from losing excessive useful letter using various features description
Breath.
S202, feature coding is carried out to the low-level image feature extracted；Bulk redundancy and noise are contained in low-level image feature, is
The robustness (Robustness) of feature representation is improved, it is necessary to compiled using a kind of eigentransformation algorithm to low-level image feature
Code, referred to as feature coding.Conventional feature coding include vector quantization coding, sparse coding, local linear constraint encode,
Fisher vector coding etc..
S203, space characteristics constraint is carried out to the low-level image feature after feature coding, obtain the description vectors of image；Feature is compiled
Space characteristics can typically be passed through after code to constrain, also referred to as feature converges.Feature convergence refers in a spatial dimension, to every
One-dimensional characteristic takes maximum or average value, can obtain the feature representation of certain feature invariant shape.Pyramid characteristic matching is
A kind of conventional feature assemblage method, this method propose, by image uniform piecemeal, feature convergence to be done in piecemeal.
S204, using grader the description vectors of image are classified, obtain the recognition result of the image of the pre- review of a film by the censor.
The latter image by preceding step can be described with the vector of a fixed dimension, next be exactly to pass through
Grader is crossed to classify to image.Usually used grader include SVM (Support Vector Machine, support to
Amount machine), random forest etc..
By above method flow, image recognition sorting technique and technology based on deep learning are employed, by there is prison
Superintend and direct or the feature of unsupervised mode learning hierarchy describes, so as to realize the pre- review of a film by the censor of intelligence.Utilize the convolution of deep learning
Neural network model carries out feature extraction and higher level of abstraction.Its advantages of is exactly to be inputted using image pixel information, at utmost
On remain all information of input picture, model output is directly image recognition result.
Embodiment 2
The embodiment of the present application additionally provides a kind of computer equipment, including memory, processor and is stored in memory
Computer program that is upper and can running on a processor, any possibility of the offer of embodiment 1 is realized during computing device program
Method flow.
Embodiment 3
The embodiment of the present application additionally provides a kind of computer-readable recording medium, is stored thereon with computer program, calculates
Machine program realizes any possible method flow that embodiment 1 provides when being executed by processor.
Embodiment 4
Fig. 3 is refer to, Fig. 3 is classified as embodiment with video display industry film, shows a kind of specific pre- review of a film by the censor side of intelligence
Method, specifically comprise the following steps：
S1. actual demand business scenario is combined, prepares training pictures and checking pictures.
S2. pretreatment and artificial mark, the picture database trained are carried out to the training picture in training pictures.
Image data pretreatment include：Preparing pictures inventory, adjustment picture size, structure picture database
S21. preparing pictures inventory, the contents such as picture file name, picture category, picture path are stored according to reference format
In train.txt and val.txt files.
S22. picture size is adjusted to 256*256, unified adjustment picture size, in order to preferably analyze.
S23. picture database is built, in caffe environment, carries out picture training, it is necessary to there is picture database, and it is non-straight
Input picture is connect, so train and val 2 groups of pictures are converted into lmbp lattice using create_imagenet.sh scripts
Formula.
S3. image average is calculated using make_imagenet_mean.sh, after calculating image average, training speed can be improved
Degree and precision.
S4. model is created, configuration file, training parameter is set.This is the most crucial place of this training pattern, Ye Shiben
The core of secondary invention.The caffenet models that model is just carried with program, change solver.prototxt and train_ therein
val.protxt。
S5. DSR, it is exactly finally training and test.If precision reaches reasonable demand, then it is assumed that training pattern
Success.If you have questions, then training data and training parameter, re-optimization training pattern are readjusted.
S6. reality scene is applied.The training picture and checking picture that this time we prepare, are for movie and television contents film point
Level prepares, can distinguish relate to it is yellow, relate to the content such as probably.Using this model, we just automatically identify the content in films and television programs, if
Related bottom line is touched, and early warning is carried out to its content, graded reporting is generated, so as to realize video display industry film hierarchical application.
Although above with general explanation and specific embodiment, the present invention is described in detail, at this
On the basis of invention, it can be made some modifications or improvements, this will be apparent to those skilled in the art.Therefore,
These modifications or improvements without departing from theon the basis of the spirit of the present invention, belong to the scope of protection of present invention.
Claims (10)
1. a kind of pre- film examination method of intelligence, it is characterised in that methods described includes：
Obtain training pictures and checking pictures；
Training picture in the training pictures is pre-processed and marks training picture generic, obtains training picture
Database；
The average of the training picture is calculated, and the training picture is subtracted into average；
The pre- review of a film by the censor training pattern caffenet models of intelligence are established under convolutional neural networks framework Caffe environment；
It is trained using the training picture for subtracting average for the caffenet models, after being trained
Caffenet models；Caffenet models after the training are carried out after input picture is received to the input picture
Identification, exports the generic to input picture；
The checking picture that the checking picture is concentrated is as the input picture of the caffenet models after the training to carrying out
Identification, the generic for exporting the input picture is pre- review of a film by the censor result.
2. the method as described in claim 1, it is characterised in that located in advance to the training picture in the training pictures
Reason, including：
The filename of the training picture, picture category, picture path are stored；
Uniform sizes of the training picture size for setting are adjusted, and are converted to unified form.
3. the method as described in claim 1, it is characterised in that the pre- review of a film by the censor training pattern caffenet models of intelligence are：
Low-level image feature extraction is carried out to input picture；
Feature coding is carried out to the low-level image feature extracted；
Space characteristics constraint is carried out to the low-level image feature after feature coding, obtains the description vectors of the input picture；
The description vectors of described image are classified using grader, obtain the recognition result of the caffenet models.
4. method as claimed in claim 3, it is characterised in that the image to the pre- review of a film by the censor carries out low-level image feature and carried
Take, including：
According to the local feature description in default compensation, the yardstick extraction preliminary hearing picture as the preliminary hearing picture
Low-level image feature.
5. method as claimed in claim 4, it is characterised in that the local feature description changes including scale invariant feature
SIFT feature, histograms of oriented gradients HOG features and local binary patterns LBP features.
6. method as claimed in claim 3, it is characterised in that the low-level image feature to being extracted carries out feature coding, bag
Include, encoded using vector quantization, be in sparse coding, local linear constraint coding or Fisher vector coding a kind of to being carried
The low-level image feature taken carries out feature coding.
7. method as claimed in claim 3, it is characterised in that the low-level image feature to after feature coding carries out space characteristics
Constraint, the description vectors of described image are obtained, including：Using the method for pyramid characteristic matching to the bottom after feature coding
Feature carries out space characteristics constraint, obtains the description vectors of described image.
8. method as claimed in claim 3, it is characterised in that the grader is support vector machines or random forest.
9. a kind of computer equipment, including memory, processor and storage can be run on a memory and on a processor
Computer program, it is characterised in that the claim 1-8 any described sides are realized during the computing device described program
Method.
10. a kind of computer-readable recording medium, is stored thereon with computer program, it is characterised in that the computer program
The claim 1-8 any described methods are realized when being executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711002884.4A CN107844758A (en) | 2017-10-24 | 2017-10-24 | Intelligence pre- film examination method, computer equipment and readable storage medium storing program for executing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711002884.4A CN107844758A (en) | 2017-10-24 | 2017-10-24 | Intelligence pre- film examination method, computer equipment and readable storage medium storing program for executing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107844758A true CN107844758A (en) | 2018-03-27 |
Family
ID=61663064
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711002884.4A Pending CN107844758A (en) | 2017-10-24 | 2017-10-24 | Intelligence pre- film examination method, computer equipment and readable storage medium storing program for executing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107844758A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111178167A (en) * | 2019-12-12 | 2020-05-19 | 咪咕文化科技有限公司 | Method and device for auditing through lens, electronic equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020647A (en) * | 2013-01-08 | 2013-04-03 | 西安电子科技大学 | Image classification method based on hierarchical SIFT (scale-invariant feature transform) features and sparse coding |
| CN103489004A (en) * | 2013-09-30 | 2014-01-01 | 华南理工大学 | Method for achieving large category image identification of deep study network |
| KR20170015639A (en) * | 2015-07-29 | 2017-02-09 | 대한민국(관리부서: 행정자치부 국립과학수사연구원장) | Personal Identification System And Method By Face Recognition In Digital Image |
| CN106980817A (en) * | 2017-02-27 | 2017-07-25 | 南京邮电大学 | A kind of terrified video frequency identifying method based on Caffe frameworks |
| CN107092926A (en) * | 2017-03-30 | 2017-08-25 | 哈尔滨工程大学 | Service robot object recognition algorithm based on deep learning |
-
2017
- 2017-10-24 CN CN201711002884.4A patent/CN107844758A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020647A (en) * | 2013-01-08 | 2013-04-03 | 西安电子科技大学 | Image classification method based on hierarchical SIFT (scale-invariant feature transform) features and sparse coding |
| CN103489004A (en) * | 2013-09-30 | 2014-01-01 | 华南理工大学 | Method for achieving large category image identification of deep study network |
| KR20170015639A (en) * | 2015-07-29 | 2017-02-09 | 대한민국(관리부서: 행정자치부 국립과학수사연구원장) | Personal Identification System And Method By Face Recognition In Digital Image |
| CN106980817A (en) * | 2017-02-27 | 2017-07-25 | 南京邮电大学 | A kind of terrified video frequency identifying method based on Caffe frameworks |
| CN107092926A (en) * | 2017-03-30 | 2017-08-25 | 哈尔滨工程大学 | Service robot object recognition algorithm based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 魏正: "基于Caffe平台深度学习的人脸识别研究与实现", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111178167A (en) * | 2019-12-12 | 2020-05-19 | 咪咕文化科技有限公司 | Method and device for auditing through lens, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111079639B (en) | Method, device, equipment and storage medium for constructing garbage image classification model | |
| CN111723654B (en) | High-altitude parabolic detection method and device based on background modeling, YOLOv3 and self-optimization | |
| CN105981008B (en) | Learn depth face representation | |
| CN112734775B (en) | Image labeling, image semantic segmentation and model training methods and devices | |
| CN105631426B (en) | The method and device of text detection is carried out to picture | |
| CN108108751B (en) | Scene recognition method based on convolution multi-feature and deep random forest | |
| JP4618098B2 (en) | Image processing system | |
| CN105404877A (en) | Human face attribute prediction method and apparatus based on deep study and multi-task study | |
| CN104298974A (en) | Human body behavior recognition method based on depth video sequence | |
| CN109165743A (en) | A kind of semi-supervised network representation learning algorithm based on depth-compression self-encoding encoder | |
| JP6798614B2 (en) | Image recognition device, image recognition method and image recognition program | |
| JPWO2012032788A1 (en) | General object image recognition apparatus and method using exclusive classifier | |
| CN109902672A (en) | Image labeling method and device, storage medium, computer equipment | |
| US20220180624A1 (en) | Method and device for automatic identification of labels of an image | |
| CN113610540A (en) | River crab anti-counterfeiting tracing method and system | |
| CN109657715A (en) | A kind of semantic segmentation method, apparatus, equipment and medium | |
| CN111291695B (en) | Training method and recognition method for recognition model of personnel illegal behaviors and computer equipment | |
| CN115050064A (en) | Face living body detection method, device, equipment and medium | |
| CN113255617A (en) | Image recognition method and device, electronic equipment and computer-readable storage medium | |
| CN116797248A (en) | Data traceability management method and system based on block chain | |
| CN113313215B (en) | Image data processing method, image data processing device, computer equipment and storage medium | |
| CN113762326A (en) | Data identification method, device and equipment and readable storage medium | |
| CN113762049B (en) | Content identification method, content identification device, storage medium and terminal equipment | |
| CN108229257A (en) | A kind of face recognition features' parallel training method based on deep learning and Spark | |
| CN107844758A (en) | Intelligence pre- film examination method, computer equipment and readable storage medium storing program for executing |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180327 |