CN110059635A - A kind of skin blemishes detection method and device - Google Patents

A kind of skin blemishes detection method and device Download PDF

Info

Publication number
CN110059635A
CN110059635A CN201910319728.3A CN201910319728A CN110059635A CN 110059635 A CN110059635 A CN 110059635A CN 201910319728 A CN201910319728 A CN 201910319728A CN 110059635 A CN110059635 A CN 110059635A
Authority
CN
China
Prior art keywords
module
face picture
trained
feature
characteristic pattern
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.)
Granted
Application number
CN201910319728.3A
Other languages
Chinese (zh)
Other versions
CN110059635B (en
Inventor
杨小栋
黄炜
王喆
张伟
许清泉
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.)
Xiamen Meitu Yifu Technology Co ltd
Original Assignee
Xiamen Meitu Technology Co Ltd
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 Xiamen Meitu Technology Co Ltd filed Critical Xiamen Meitu Technology Co Ltd
Priority to CN201910319728.3A priority Critical patent/CN110059635B/en
Publication of CN110059635A publication Critical patent/CN110059635A/en
Application granted granted Critical
Publication of CN110059635B publication Critical patent/CN110059635B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/0002Inspection of images, e.g. flaw detection
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/70Determining position or orientation of objects or cameras
    • G06T7/73Determining position or orientation of objects or cameras using feature-based methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168Feature extraction; Face representation
    • G06V40/171Local features and components; Facial parts ; Occluding parts, e.g. glasses; Geometrical relationships
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16Human faces, e.g. facial parts, sketches or expressions
    • G06V40/172Classification, e.g. identification
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20081Training; Learning
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20084Artificial neural networks [ANN]
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30004Biomedical image processing
    • G06T2207/30088Skin; Dermal
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30196Human being; Person
    • G06T2207/30201Face

Landscapes

  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

The embodiment of the present application provides a kind of skin blemishes detection method and device, which comprises pre-processes to the face picture to be tested comprising skin blemishes;Pretreated face picture to be tested inputs in the Defect Detection network model obtained according to lightweight network training, obtains the characteristic pattern of the face picture to be tested;According to the characteristic pattern of the face picture to be tested, the flaw classification and flaw location of the face picture to be tested are obtained.Skin blemishes are detected using the Defect Detection network model that lightweight network training provided by the present application obtains, classification and the flaw position of skin blemishes is can recognize that, reduces the memory consumption of computer equipment.

Description

A kind of skin blemishes detection method and device
Technical field
This application involves field of image recognition, in particular to a kind of skin blemishes detection method and device.
Background technique
For the skin blemishes detection in facial image, main there are two steps: the extraction of unwanted visual characteristic and using extracting Feature carry out flaw classification and positioning.Wherein, the extraction of unwanted visual characteristic is particularly important, and speed and the extraction operation of extraction disappear The memory of consumption is all the important reference of unwanted visual characteristic extraction efficiency.
Currently, one is manual feature extractions, such as not using scale there are mainly two types of the schemes extracted for unwanted visual characteristic Become eigentransformation (Scale-invariant feature transform, abbreviation SIFT) and histograms of oriented gradients (Histogram of Oriented Gradient, abbreviation HOG) is extracted;Another kind is that convolutional neural networks extract, such as Utilize VGG (Visual Geometry Group Network) network and ResNet (Residual Neural Network) net Network.When using manual feature extraction, robustness is generally not good enough, when being extracted using general convolutional neural networks, not only consumes Huge with memory, the speed of service is also very slow.
It is current problem to be solved in view of this, how to realize the rapidly extracting to unwanted visual characteristic.
Summary of the invention
The application provides a kind of skin blemishes detection method and device.
In a first aspect, the embodiment of the present application provides a kind of skin blemishes detection method, it is applied to computer equipment, the side Method includes:
Face picture to be tested comprising skin blemishes is pre-processed;
Pretreated face picture to be tested inputs the Defect Detection network model obtained according to lightweight network training In, obtain the characteristic pattern of the face picture to be tested;
According to the characteristic pattern of the face picture to be tested, the flaw classification and flaw of the face picture to be tested are obtained Position.
Optionally, the step of obtaining the Defect Detection network model the method also includes training, which includes:
It will be inputted after pretreatment using lightweight network MobileNetV2 as basic network framework after training face picture In the detection model built, wherein be labeled with the flaw classification and flaw location of skin blemishes in the face picture to be trained;
The detection model is trained using the face picture to be trained, and according to the loss of the detection model The functional value of function is adjusted the hyper parameter of the detection model, using trained detection model as the Defect Detection Network model.
Optionally, the lightweight network MobileNetV2 include output characteristic pattern be sequentially reduced the first module, the Two modules, third module, the 4th module and the 5th module, the method also includes obtaining the feature of the face picture to be trained The step of target flaw information in figure, which includes:
Institute will be used as after the fifth feature figure fusion of the fourth feature figure of 4th module output and the output of the 5th module State the characteristic pattern of face picture to be trained, wherein the fourth feature figure passes through described first by the face picture to be trained Module, the second module, third module and the 4th module obtain after successively handling, and the fifth feature figure is by the face to be trained Picture obtains after first module, the second module, third module, the 4th module and the 5th module are successively handled;
The target flaw information in the characteristic pattern of the face picture to be trained is obtained using anchor mechanism.
Optionally, the fifth feature figure by the fourth feature figure of the 4th module output and the output of the 5th module melts Characteristic pattern after conjunction as the face picture to be trained, comprising:
Bilinear interpolation is carried out to the fourth feature figure by the 5th module and obtains the fifth feature figure;
Reduce the port number of the 5th module by convolution, so that the channel of the fifth feature figure and fourth feature figure Number is identical;
The fourth feature figure is added pixel-by-pixel with the fifth feature figure after reduction port number, is obtained described wait instruct Practice the characteristic pattern of face picture.
Optionally, the target flaw in the characteristic pattern that the face picture to be trained is obtained using anchor mechanism is believed Breath, comprising:
According to the characteristic pattern of the face picture to be trained, the predicted position of anchor point is divided;
According to the skin blemishes size of the face picture to be trained, the size of the anchor point is set;
The characteristic pattern of the face picture to be trained is obtained in the predicted position by the anchor point after setting size In target flaw information.
Second aspect, the embodiment of the present application also provide a kind of skin blemishes detection device, are applied to computer equipment, described Device includes:
Processing module, for being pre-processed to the face picture to be tested comprising skin blemishes;
Input module inputs the flaw obtained according to lightweight network training for pretreated face picture to be tested It detects in network model, obtains the characteristic pattern of the face picture to be tested;
Output module obtains the face picture to be tested for the characteristic pattern according to the face picture to be tested Flaw classification and flaw location.
Optionally, the skin blemishes detection device further include:
Training module, for that will input after training face picture after pretreatment with lightweight network MobileNetV2 In the detection model built for basic network framework, wherein be labeled with the flaw of skin blemishes in the face picture to be trained Classification and flaw location;
The detection model is trained using the face picture to be trained, and according to the loss of the detection model The functional value of function is adjusted the hyper parameter of the detection model, using trained detection model as the Defect Detection Network model.
Optionally, the lightweight network MobileNetV2 include output characteristic pattern be sequentially reduced the first module, the Two modules, third module, the 4th module and the 5th module, described device include:
Fusion Module, the fifth feature figure of fourth feature figure and the output of the 5th module for exporting the 4th module Characteristic pattern after fusion as the face picture to be trained, wherein the fourth feature figure is by the face picture to be trained It is obtained after first module, the second module, third module and the 4th module are successively handled, the fifth feature figure is by institute Face picture to be trained is stated successively to handle by first module, the second module, third module, the 4th module and the 5th module After obtain;
Module is obtained, the target flaw letter in characteristic pattern for obtaining the face picture to be trained using anchor mechanism Breath.
Optionally, the Fusion Module is specifically used for:
Bilinear interpolation is carried out to the fourth feature figure by the 5th module and obtains the fifth feature figure;
Reduce the port number of the 5th module by convolution, so that the channel of the fifth feature figure and fourth feature figure Number is identical;
The fourth feature figure is added pixel-by-pixel with the fifth feature figure after reduction port number, is obtained described wait instruct Practice the characteristic pattern of face picture.
Optionally, the acquisition module is specifically used for:
According to the characteristic pattern of the face picture to be trained, the predicted position of anchor point is divided;
According to the skin blemishes size of the face picture to be trained, the size of the anchor point is set;
The characteristic pattern of the face picture to be trained is obtained in the predicted position by the anchor point after setting size In target flaw information.
The embodiment of the present application provides a kind of method and device of skin blemishes detection, is obtained using lightweight network training Defect Detection network model detects skin blemishes, can rapidly and accurately identify classification and the flaw institute of skin blemishes In position, reduce the memory consumption of computer equipment, improves the speed of service of Defect Detection network model.
Detailed description of the invention
Technical solution in ord to more clearly illustrate embodiments of the present application, below will be to needed in the embodiment attached Figure is briefly described.It should be appreciated that the following drawings illustrates only some embodiments of the application, therefore it is not construed as pair The restriction of range.It for those of ordinary skill in the art, without creative efforts, can also be according to this A little attached drawings obtain other relevant attached drawings.
Fig. 1 is the structural block diagram of computer equipment provided by the embodiments of the present application;
Fig. 2 is the step schematic process flow diagram of skin blemishes detection method provided by the embodiments of the present application;
Fig. 3 is other steps flow chart schematic block diagrams of skin blemishes detection method provided by the embodiments of the present application;
Fig. 4 is other steps flow chart schematic block diagrams of skin blemishes detection method provided by the embodiments of the present application;
Fig. 5 is the sub-step flow diagram of step S206 in Fig. 4;
Fig. 6 is the sub-step flow diagram of step S207 in Fig. 4;
Fig. 7 is the structural schematic diagram of skin blemishes detection device provided by the embodiments of the present application.
Icon: 100- computer equipment;110- skin blemishes detection device;111- memory;112- processor;113- is logical Believe unit;1101- processing module;1102- input module;1103- output module;1104- training module;1105- Fusion Module; 1106- obtains module.
Specific embodiment
To keep the purposes, technical schemes and advantages of the embodiment of the present application clearer, below in conjunction with the embodiment of the present application In attached drawing, the technical scheme in the embodiment of the application is clearly and completely described.Obviously, described embodiment is Some embodiments of the present application, instead of all the embodiments.The application being usually described and illustrated herein in the accompanying drawings is implemented The component of example can be arranged and be designed with a variety of different configurations.
Therefore, the detailed description of the embodiments herein provided in the accompanying drawings is not intended to limit below claimed Scope of the present application, but be merely representative of the selected embodiment of the application.Based on the embodiment in the application, this field is common Technical staff's every other embodiment obtained without creative efforts belongs to the model of the application protection It encloses.
It should also be noted that similar label and letter indicate similar terms in following attached drawing, therefore, once a certain Xiang Yi It is defined in a attached drawing, does not then need that it is further defined and explained in subsequent attached drawing.
In addition, term " first ", " second " etc. are only used for distinguishing description, it is not understood to indicate or imply relatively important Property.
In the description of the present application, it is also necessary to which explanation is unless specifically defined or limited otherwise, " setting ", " even Connect " etc. terms shall be understood in a broad sense, for example, " connection " may be a fixed connection, may be a detachable connection, or integrally connect It connects;It can be mechanical connection, be also possible to be electrically connected;It can be and be directly connected to, can also be indirectly connected with by intermediary, it can To be the connection inside two elements.For the ordinary skill in the art, can understand as the case may be above-mentioned The concrete meaning of term in this application.
With reference to the accompanying drawing, the specific embodiment of the application is described in detail.
Fig. 1 is please referred to, Fig. 1 is the structural block diagram of computer equipment provided by the embodiments of the present application.The computer equipment 100 include skin blemishes detection device 110, memory 111, processor 112 and communication unit 113.
The memory 111, processor 112 and each element of communication unit 113 are directly or indirectly electrical between each other Connection, to realize the transmission or interaction of data.For example, these elements can pass through one or more communication bus or letter between each other Number line, which is realized, to be electrically connected.The skin blemishes detection device 110 includes at least one can be with software or firmware (firmware) Form be stored in the memory 111 or be solidificated in the operating system (operating of the computer equipment 100 System, OS) in software function module.The processor 112 is for executing the executable mould stored in the memory 111 Block, such as software function module and computer program etc. included by the skin blemishes detection device 110.
Wherein, the memory 111 may be, but not limited to, random access memory (Random Access Memory, RAM), read-only memory (Read Only Memory, ROM), programmable read only memory (Programmable Read-Only Memory, PROM), erasable read-only memory (Erasable Programmable Read-Only Memory, EPROM), electricallyerasable ROM (EEROM) (Electric Erasable Programmable Read-Only Memory, EEPROM) etc..Wherein, memory 111 is for storing program or data.
The processor 112 may be a kind of IC chip, the processing capacity with data.Above-mentioned processor 112 can be general processor, including central processing unit (Central Processing Unit, CPU), network processing unit (Network Processor, NP) etc..It may be implemented or execute each method, step and the logic diagram in the disclosure.It is general Processor can be microprocessor or the processor is also possible to any conventional processor etc..
Communication unit 113 is used to establish the communication connection between computer equipment 100 and external communications terminals by network, Realize the transmitting-receiving operation of network signal and data.Above-mentioned network signal may include wireless signal or wire signal.
It is appreciated that structure shown in FIG. 1 is only to illustrate, computer equipment 100 may also include it is more than shown in Fig. 1 or The less component of person, or with the configuration different from shown in Fig. 1.Each component shown in Fig. 1 can using hardware, software or A combination thereof is realized.
Referring to figure 2., Fig. 2 is the step schematic process flow diagram of skin blemishes detection method provided by the embodiments of the present application. The method includes the steps S201 to step S203.
Step S201 pre-processes the face picture to be tested comprising skin blemishes.
Step S202, pretreated face picture to be tested input the Defect Detection obtained according to lightweight network training In network model, the characteristic pattern of the face picture to be tested is obtained.
Step S203 obtains the flaw of the face picture to be tested according to the characteristic pattern of the face picture to be tested Classification and flaw location.
It in the present embodiment, can be from the output result of the Defect Detection network model obtained by lightweight network training In, directly acquire classification and the position of flaw.Without taking the two-stage method for first positioning and classifying again, i.e., first train one two Class detector, in one subdivider of training.Do so can exist due to multistage calculating, loss of significance can be continuously increased and The slower problem of the excessive calculating speed of calculating process.
Referring to figure 3., Fig. 3 is other steps flow chart schematic blocks of skin blemishes detection method provided by the embodiments of the present application Figure.The step of obtaining the Defect Detection network model the method also includes training, which includes step S204 and step S205。
Step S204 will be inputted after pretreatment using lightweight network MobileNetV2 as base after training face picture In the detection model that present networks framework is built, wherein be labeled with the flaw classification of skin blemishes in the face picture to be trained And flaw location.
In the present embodiment, the preprocessing process of face picture to be trained, which can be, carries out face picture to be trained at random Reduction and flip horizontal, and be to be sized by the scaled of picture, such as 800*800, finally each pixel is normalized to Between 0-1.
In the present embodiment, the classification of skin blemishes can be divided into acne, acne print, mole, spot etc..In other embodiments, skin Skin flaw can also include other classifications.
Step S205 is trained the detection model using the face picture to be trained, and according to the detection The functional value of the loss function of model is adjusted the hyper parameter of the detection model, using trained detection model as institute State Defect Detection network model.
In the present embodiment, the loss function of the detection model may include three parts: the Classification Loss of Anchor (is handed over Pitch entropy), the position of object returns loss (mean square error loss) and the IOU loss of object (interaction is than loss).It can pass through Above three index measures detection model, so that detection model is more accurate.For example, for the Classification Loss of Anchor For, the classification of current flaw can be set as acne, be indicated with (1,0,0,0), but the predicted value of detection model be but (0,0, 0,0), this prediction differs greatly with true value, so loss is just very big, allows penalty values after being adjusted according to hyper parameter Reduce, so that prediction result is closer to true value, improves the accuracy of detection model.It should be understood that due to skin Acne, acne print, mole, the quantity of spot are uneven in flaw, and the quantity of especially spot accounts for the overwhelming majority, so when calculating loss function, Weight ratio between them can be respectively set to 1:1:1:0.1, weaken the most spot of quantity for the contribution of loss with this.
It should be understood that in the present embodiment, it, can be by defeated for trained Defect Detection network model Enter facial image to be tested to test it into Defect Detection network model.Defect Detection network model is to be measured in reception The classification confidence (flaw classification) of the corresponding characteristic pattern for exporting facial image to be tested, prediction block position after the facial image of examination (position of the flaw in facial image to be tested) and IOU confidence level (the fitting journey of prediction block position and physical location Degree).It is 0.3 that prediction threshold value (classification confidence * IOU confidence level), which can be set, if the value of classification confidence * IOU confidence level is big It it may be considered that the secondary prediction result is positive sample, otherwise is negative sample in prediction threshold value.In the present embodiment, can also lead to It crosses and non-very big threshold value is set for 0.1 to filter duplicate prediction block.
It in the present embodiment, can be with since eyes, mouth and the nostril in facial image to be tested are there is no flaw To by applying exposure mask (MASK) to eyes, mouth and nostril to filter apparent erroneous detection.
Referring to figure 4., Fig. 4 is other steps flow chart schematic blocks of skin blemishes detection method provided by the embodiments of the present application Figure.The lightweight network MobileNetV2 includes the first module, the second module, third that the characteristic pattern of output is sequentially reduced Module, the 4th module and the 5th module.The method also includes obtaining the target in the characteristic pattern of the face picture to be trained The step of flaw information, the step include step S206 and step S207.
Step S206 merges the fifth feature figure of the fourth feature figure of the 4th module output and the output of the 5th module Characteristic pattern as the face picture to be trained afterwards, wherein the fourth feature figure is passed through by the face picture to be trained First module, the second module, third module and the 4th module obtain after successively handling, the fifth feature figure by it is described to Training face picture obtains after first module, the second module, third module, the 4th module and the 5th module are successively handled It arrives.
In the present embodiment, the size of facial image to be trained can be 800*800, and the first module is defeated to the 5th module The step-length stride of characteristic pattern out successively can be 2,4,8,16,32.Therefore face picture to be trained by first module, The size for the fourth feature figure that second module, third module and the 4th module obtain after successively handling is 50*50, face to be trained Picture obtains the 5th spy after first module, the second module, third module, the 4th module and the 5th module are successively handled The size for levying figure is 25*25.
Step S207 obtains the target flaw information in the characteristic pattern of the face picture to be trained using anchor mechanism.
In the present embodiment, it can use anchor mechanism simulation sliding window to obtain the characteristic pattern of face picture to be trained In target flaw information, target flaw information may include flaw classification, the position of flaw predicted position and prediction and reality The compactness of position.Flaw classification can be indicated by the flaw confidence map of four-way, the Feature Mapping figure of four-way indicates the flaw Defect predicted position, logical Feature Mapping figure indicates the position of prediction and the compactness of physical location together.
Referring to figure 5., Fig. 5 is the sub-step flow diagram of step S206 in Fig. 4.In the present embodiment, step S206 It may include sub-step S2061, sub-step S2062 and sub-step S2063.
Step S2061 carries out bilinear interpolation to the fourth feature figure by the 5th module and obtains the described 5th Characteristic pattern.
In the present embodiment, the fifth feature figure size by the output of the 5th module can be 32*32, and it is linear to carry out double property Interpolation can be understood as up-sampling it, expands characteristic layer, is changed into original twice, the i.e. size of fifth feature figure Become 64*64.
Step S2062 reduces the port number of the 5th module by convolution, so that the fifth feature figure and the 4th spy The port number for levying figure is identical.
The fourth feature figure is added pixel-by-pixel with the fifth feature figure after reduction port number, obtains by step S2063 To the characteristic pattern of the face picture to be trained.
In the present embodiment, the size of the fourth feature figure exported by the 4th module can be 50*50, and port number can be with It is 48, the size of the fifth feature figure of the 5th module output can be 25*25, and port number can be 80, aforementioned to the 5th mould After the fifth feature figure of block output carries out bilinear interpolation, size becomes 50*50, port number 80.1*1 convolution can be used The channel of fifth feature figure is become 48, fifth feature figure is consistent with the port number of fourth feature figure at this time, can phase pixel-by-pixel Add and merge fourth feature figure and fifth feature figure, obtains the characteristic pattern of face picture to be trained.It should be understood that In the present embodiment, the characteristic pattern that the 4th module of lightweight network MobileNetV2 and the 5th module export is merged, It reduces calculating process as far as possible in the case where not reducing precision, improves the speed of Defect Detection.
Fig. 6 is please referred to, Fig. 6 is the sub-step flow diagram of step S207 in Fig. 4.In the present embodiment, step S207 It may include sub-step S2071, sub-step S2072 and sub-step S2073.
Step S2071 divides the predicted position of anchor point according to the characteristic pattern of the face picture to be trained.
The size of the anchor point is arranged according to the skin blemishes size of the face picture to be trained in step S2072.
Step S2073 obtains the face figure to be trained in the predicted position by the anchor point after setting size Target flaw information in the characteristic pattern of piece.
In the present embodiment, the predicted position of anchor point can be divided according to the characteristic pattern of face picture to be trained.For example, The size of the characteristic pattern of face picture to be trained can be 50*50, at this time it is considered that this feature figure is divided in order to 50*50's Grid can preset different size in each grid, the anchor point (i.e. prediction block) of different length-width ratio adapts to different flaws. It should be understood that the size and length-width ratio of each anchor point, are the size statistics according to the flaw in face figure to be trained. For example, five anchor points can be set to detect to the flaw in each square by statistics, the sizes of five anchor points can be with Respectively (0.1,0.1), (0.3,0.3), (0.5,0.5), (0.7,0.7) and (0.9,0.9), it is each in face picture to be trained The size of class skin blemishes can respectively correspond one in above-mentioned five anchor points.
Fig. 7 is please referred to, Fig. 7 is the structural schematic diagram of skin blemishes detection device 110 provided by the embodiments of the present application.It is described Device includes:
Processing module 1101, for being pre-processed to the face picture to be tested comprising skin blemishes.
Implementation about processing module 1101 can not go to live in the household of one's in-laws on getting married herein refering to the associated description of step S201 in Fig. 2 It states.
Input module 1102 is obtained for pretreated face picture input to be tested according to lightweight network training In Defect Detection network model, the characteristic pattern of the face picture to be tested is obtained.
Implementation about input module 1102 can not go to live in the household of one's in-laws on getting married herein refering to the associated description of step S202 in Fig. 2 It states.
Output module 1103 obtains the face figure to be tested for the characteristic pattern according to the face picture to be tested The flaw classification and flaw location of piece.
Implementation about output module 1103 can not go to live in the household of one's in-laws on getting married herein refering to the associated description of step S203 in Fig. 2 It states.
Further, the skin blemishes detection device 110 further include:
Training module 1104, for that will input after training face picture after pretreatment with lightweight network MobileNetV2 is in the detection model that basic network framework is built, wherein is labeled with skin in the face picture to be trained The flaw classification and flaw location of flaw;
The detection model is trained using the face picture to be trained, and according to the loss of the detection model The functional value of function is adjusted the hyper parameter of the detection model, using trained detection model as the Defect Detection Network model.
About training module 1104 implementation can refering to the associated description of step S204 and step S205 in Fig. 3, Therefore not to repeat here.
Further, the lightweight network MobileNetV2 include output characteristic pattern be sequentially reduced the first module, Second module, third module, the 4th module and the 5th module, described device include:
Fusion Module 1105, the 5th of fourth feature figure and the output of the 5th module for exporting the 4th module are special Characteristic pattern after sign figure fusion as the face picture to be trained, wherein the fourth feature figure is by the face to be trained Picture obtains after first module, the second module, third module and the 4th module are successively handled, the fifth feature figure First module, the second module, third module, the 4th module and the 5th module are passed through successively by the face picture to be trained It is obtained after processing.
Implementation about Fusion Module 1105 can not go to live in the household of one's in-laws on getting married herein refering to the associated description of step S206 in Fig. 4 It states.
Module 1106 is obtained, the target flaw in characteristic pattern for obtaining the face picture to be trained using anchor mechanism Defect information.
It can not go to live in the household of one's in-laws on getting married herein refering to the associated description of step S207 in Fig. 4 about the implementation for obtaining module 1106 It states.
Further, the Fusion Module 1105 is specifically used for:
Bilinear interpolation is carried out to the fourth feature figure by the 5th module and obtains the fifth feature figure.
Reduce the port number of the 5th module by convolution, so that the channel of the fifth feature figure and fourth feature figure Number is identical.
The fourth feature figure is added pixel-by-pixel with the fifth feature figure after reduction port number, is obtained described wait instruct Practice the characteristic pattern of face picture.
Further, the acquisition module 1106 is specifically used for:
According to the characteristic pattern of the face picture to be trained, the predicted position of anchor point is divided.
According to the skin blemishes size of the face picture to be trained, the size of the anchor point is set.
The characteristic pattern of the face picture to be trained is obtained in the predicted position by the anchor point after setting size In target flaw information.
Realization principle in the disclosure, in the realization principle of skin blemishes detection device and aforementioned skin blemishes detection method Similar, corresponding contents can be refering to the description in preceding method, thus therefore not to repeat here.
In conclusion the embodiment of the present application provides a kind of skin blemishes detection method and device using lightweight network training Obtained Defect Detection network model detects skin blemishes, can rapidly and accurately identify skin blemishes classification and Flaw position reduces the memory consumption of computer equipment, improves the speed of service of Defect Detection network model.
The foregoing is merely preferred embodiment of the present application, are not intended to limit this application, for the skill of this field For art personnel, various changes and changes are possible in this application.Within the spirit and principles of this application, made any to repair Change, equivalent replacement, improvement etc., should be included within the scope of protection of this application.

Claims (10)

1. a kind of skin blemishes detection method, which is characterized in that be applied to computer equipment, which comprises
Face picture to be tested comprising skin blemishes is pre-processed;
Pretreated face picture to be tested inputs in the Defect Detection network model obtained according to lightweight network training, obtains To the characteristic pattern of the face picture to be tested;
According to the characteristic pattern of the face picture to be tested, flaw classification and the flaw position of the face picture to be tested are obtained It sets.
2. the method according to claim 1, wherein the method also includes training to obtain the Defect Detection net The step of network model, the step include:
It will input after pretreatment after training face picture and be built by basic network framework of lightweight network MobileNetV2 Detection model in, wherein the flaw classification and flaw location of skin blemishes are labeled in the face picture to be trained;
The detection model is trained using the face picture to be trained, and according to the loss function of the detection model Functional value the hyper parameter of the detection model is adjusted, using trained detection model as the Defect Detection network Model.
3. according to the method described in claim 2, it is characterized in that, the lightweight network MobileNetV2 includes output The first module, the second module, third module, the 4th module and the 5th module that characteristic pattern is sequentially reduced, the method also includes The step of obtaining the target flaw information in the characteristic pattern of the face picture to be trained, which includes:
Will after the fourth feature figure of the 4th module output and the fifth feature figure fusion of the 5th module output as it is described to Training face picture characteristic pattern, wherein the fourth feature figure by face picture train pass through first module, Second module, third module and the 4th module obtain after successively handling, and the fifth feature figure is by the face picture to be trained It is obtained after first module, the second module, third module, the 4th module and the 5th module are successively handled;
The target flaw information in the characteristic pattern of the face picture to be trained is obtained using anchor mechanism.
4. according to the method described in claim 3, it is characterized in that, it is described by the 4th module output fourth feature figure and Characteristic pattern after the fifth feature figure fusion of 5th module output as the face picture to be trained, comprising:
Bilinear interpolation is carried out to the fourth feature figure by the 5th module and obtains the fifth feature figure;
Reduce the port number of the 5th module by convolution, so that the port number phase of the fifth feature figure and fourth feature figure Together;
The fourth feature figure is added pixel-by-pixel with the fifth feature figure after reduction port number, is obtained described to training of human The characteristic pattern of face picture.
5. according to the method described in claim 3, it is characterized in that, described obtain the face figure to be trained using anchor mechanism Target flaw information in the characteristic pattern of piece, comprising:
According to the characteristic pattern of the face picture to be trained, the predicted position of anchor point is divided;
According to the skin blemishes size of the face picture to be trained, the size of the anchor point is set;
Through the anchor point after setting size in the characteristic pattern that the predicted position obtains the face picture to be trained Target flaw information.
6. a kind of skin blemishes detection device, which is characterized in that be applied to computer equipment, described device includes:
Processing module, for being pre-processed to the face picture to be tested comprising skin blemishes;
Input module inputs the Defect Detection obtained according to lightweight network training for pretreated face picture to be tested In network model, the characteristic pattern of the face picture to be tested is obtained;
Output module obtains the flaw of the face picture to be tested for the characteristic pattern according to the face picture to be tested Classification and flaw location.
7. device according to claim 6, which is characterized in that the skin blemishes detection device further include:
Training module, for that will be inputted after pretreatment using lightweight network MobileNetV2 as base after training face picture In the detection model that present networks framework is built, wherein be labeled with the flaw classification of skin blemishes in the face picture to be trained And flaw location;
The detection model is trained using the face picture to be trained, and according to the loss function of the detection model Functional value the hyper parameter of the detection model is adjusted, using trained detection model as the Defect Detection network Model.
8. device according to claim 7, which is characterized in that the lightweight network MobileNetV2 includes output The first module, the second module, third module, the 4th module and the 5th module that characteristic pattern is sequentially reduced, described device further include:
Fusion Module, the fifth feature figure fusion of fourth feature figure and the output of the 5th module for exporting the 4th module Characteristic pattern as the face picture to be trained afterwards, wherein the fourth feature figure is passed through by the face picture to be trained First module, the second module, third module and the 4th module obtain after successively handling, the fifth feature figure by it is described to Training face picture obtains after first module, the second module, third module, the 4th module and the 5th module are successively handled It arrives;
Module is obtained, the target flaw information in characteristic pattern for obtaining the face picture to be trained using anchor mechanism.
9. device according to claim 8, which is characterized in that the Fusion Module is specifically used for:
Bilinear interpolation is carried out to the fourth feature figure by the 5th module and obtains the fifth feature figure;
Reduce the port number of the 5th module by convolution, so that the port number phase of the fifth feature figure and fourth feature figure Together;
The fourth feature figure is added pixel-by-pixel with the fifth feature figure after reduction port number, is obtained described to training of human The characteristic pattern of face picture.
10. device according to claim 8, which is characterized in that the acquisition module is specifically used for:
According to the characteristic pattern of the face picture to be trained, the predicted position of anchor point is divided;
According to the skin blemishes size of the face picture to be trained, the size of the anchor point is set;
Through the anchor point after setting size in the characteristic pattern that the predicted position obtains the face picture to be trained Target flaw information.
CN201910319728.3A 2019-04-19 2019-04-19 Skin defect detection method and device Active CN110059635B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910319728.3A CN110059635B (en) 2019-04-19 2019-04-19 Skin defect detection method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910319728.3A CN110059635B (en) 2019-04-19 2019-04-19 Skin defect detection method and device

Publications (2)

Publication Number Publication Date
CN110059635A true CN110059635A (en) 2019-07-26
CN110059635B CN110059635B (en) 2021-03-23

Family

ID=67319820

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910319728.3A Active CN110059635B (en) 2019-04-19 2019-04-19 Skin defect detection method and device

Country Status (1)

Country Link
CN (1) CN110059635B (en)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111275703A (en) * 2020-02-27 2020-06-12 腾讯科技(深圳)有限公司 Image detection method, image detection device, computer equipment and storage medium
CN111695463A (en) * 2020-05-29 2020-09-22 深圳数联天下智能科技有限公司 Training method of face impurity detection model and face impurity detection method
CN111767858A (en) * 2020-06-30 2020-10-13 北京百度网讯科技有限公司 Image recognition method, device, equipment and computer storage medium
CN111860169A (en) * 2020-06-18 2020-10-30 北京旷视科技有限公司 Skin analysis method, device, storage medium and electronic equipment
CN112263224A (en) * 2020-12-24 2021-01-26 南京信息工程大学 Medical information processing method based on FPGA edge calculation
CN113392861A (en) * 2020-03-12 2021-09-14 北京京东乾石科技有限公司 Model training method, map drawing method, device, computer device and medium
CN113627435A (en) * 2021-07-13 2021-11-09 南京大学 Method and system for detecting and identifying flaws of ceramic tiles
CN116993714A (en) * 2023-08-30 2023-11-03 深圳伯德睿捷健康科技有限公司 Skin detection method, system and computer readable storage medium

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106469302A (en) * 2016-09-07 2017-03-01 成都知识视觉科技有限公司 A kind of face skin quality detection method based on artificial neural network
CN107945173A (en) * 2017-12-11 2018-04-20 深圳市宜远智能科技有限公司 A kind of skin disease detection method and system based on deep learning
CN108780497A (en) * 2017-11-14 2018-11-09 深圳和而泰智能控制股份有限公司 A kind of skin blemishes point sorting technique and electronic equipment
CN108846311A (en) * 2018-04-28 2018-11-20 北京羽医甘蓝信息技术有限公司 The method and device of the facial pieces of skin shape defect of detection based on deep learning
CN109464132A (en) * 2019-01-11 2019-03-15 肖湘江 Robot, kindergarten morning check system

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106469302A (en) * 2016-09-07 2017-03-01 成都知识视觉科技有限公司 A kind of face skin quality detection method based on artificial neural network
CN108780497A (en) * 2017-11-14 2018-11-09 深圳和而泰智能控制股份有限公司 A kind of skin blemishes point sorting technique and electronic equipment
CN107945173A (en) * 2017-12-11 2018-04-20 深圳市宜远智能科技有限公司 A kind of skin disease detection method and system based on deep learning
CN108846311A (en) * 2018-04-28 2018-11-20 北京羽医甘蓝信息技术有限公司 The method and device of the facial pieces of skin shape defect of detection based on deep learning
CN109464132A (en) * 2019-01-11 2019-03-15 肖湘江 Robot, kindergarten morning check system

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
FERNANDO PEREIRA DOS SANTOS 等: "Robust feature spaces from pre-trained deep network layers for skin lesion classification", 《2018 31ST SIBGRAPI CONFERENCE ON GRAPHICS, PATTERNS AND IMAGES (SIBGRAPI)》 *
MARK SANDLER 等: "MobileNetV2: Inverted Residuals and Linear Bottlenecks", 《ARXIV:1801.04381V4 [CS.CV]》 *

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111275703A (en) * 2020-02-27 2020-06-12 腾讯科技(深圳)有限公司 Image detection method, image detection device, computer equipment and storage medium
CN111275703B (en) * 2020-02-27 2023-10-27 腾讯科技(深圳)有限公司 Image detection method, device, computer equipment and storage medium
CN113392861A (en) * 2020-03-12 2021-09-14 北京京东乾石科技有限公司 Model training method, map drawing method, device, computer device and medium
CN111695463B (en) * 2020-05-29 2022-07-26 深圳数联天下智能科技有限公司 Training method of face impurity detection model and face impurity detection method
CN111695463A (en) * 2020-05-29 2020-09-22 深圳数联天下智能科技有限公司 Training method of face impurity detection model and face impurity detection method
CN111860169A (en) * 2020-06-18 2020-10-30 北京旷视科技有限公司 Skin analysis method, device, storage medium and electronic equipment
CN111860169B (en) * 2020-06-18 2024-04-30 北京旷视科技有限公司 Skin analysis method, device, storage medium and electronic equipment
CN111767858A (en) * 2020-06-30 2020-10-13 北京百度网讯科技有限公司 Image recognition method, device, equipment and computer storage medium
CN111767858B (en) * 2020-06-30 2024-03-22 北京百度网讯科技有限公司 Image recognition method, device, equipment and computer storage medium
CN112263224A (en) * 2020-12-24 2021-01-26 南京信息工程大学 Medical information processing method based on FPGA edge calculation
CN112263224B (en) * 2020-12-24 2021-03-23 南京信息工程大学 Medical information processing method based on FPGA edge calculation
CN113627435A (en) * 2021-07-13 2021-11-09 南京大学 Method and system for detecting and identifying flaws of ceramic tiles
CN116993714A (en) * 2023-08-30 2023-11-03 深圳伯德睿捷健康科技有限公司 Skin detection method, system and computer readable storage medium

Also Published As

Publication number Publication date
CN110059635B (en) 2021-03-23

Similar Documents

Publication Publication Date Title
CN110059635A (en) A kind of skin blemishes detection method and device
Liu et al. A detection and recognition system of pointer meters in substations based on computer vision
CN106875381B (en) Mobile phone shell defect detection method based on deep learning
CN112052186B (en) Target detection method, device, equipment and storage medium
CN108229341A (en) Sorting technique and device, electronic equipment, computer storage media, program
CN113409284B (en) Circuit board fault detection method, device, equipment and storage medium
CN109598298B (en) Image object recognition method and system
Geng et al. An improved helmet detection method for YOLOv3 on an unbalanced dataset
CN115409990B (en) Medical image segmentation method, device, equipment and storage medium
CN112699824B (en) Method and device for detecting constant of electric energy meter and storage medium
CN108460346A (en) Fingerprint identification method and device
CN113780145A (en) Sperm morphology detection method, sperm morphology detection device, computer equipment and storage medium
CN110807416A (en) Digital instrument intelligent recognition device and method suitable for mobile detection device
CN110796039B (en) Face flaw detection method and device, electronic equipment and storage medium
CN114694130A (en) Method and device for detecting telegraph poles and pole numbers along railway based on deep learning
CN117351499B (en) Split-combined indication state identification method, system, computer equipment and medium
CN114219073A (en) Method and device for determining attribute information, storage medium and electronic device
CN111931721B (en) Method and device for detecting color and number of annual inspection label and electronic equipment
Shi et al. Detection and recognition of digital instrument in substation using improved YOLO-v3
Peng et al. Deep learning-based autonomous real-time digital meter reading recognition method for natural scenes
CN112560791A (en) Recognition model training method, recognition method and device and electronic equipment
CN111753618A (en) Image recognition method and device, computer equipment and computer readable storage medium
CN115546539A (en) Magnetic flip plate liquid level reading method and device based on machine vision and readable medium
CN112950652B (en) Robot and hand image segmentation method and device thereof
CN113780239B (en) Iris recognition method, iris recognition device, electronic device and computer readable medium

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
GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20211029

Address after: 361100 568, No. 942, tonglong Second Road, torch high tech Zone (Xiang'an) Industrial Zone, Xiang'an District, Xiamen City, Fujian Province

Patentee after: Xiamen Meitu Yifu Technology Co.,Ltd.

Address before: B1f-089, Zone C, Huaxun building, software park, torch high tech Zone, Xiamen City, Fujian Province

Patentee before: XIAMEN HOME MEITU TECHNOLOGY Co.,Ltd.