WO2018113512A1 - 图像处理方法以及相关装置 - Google Patents

图像处理方法以及相关装置 Download PDF

Info

Publication number
WO2018113512A1
WO2018113512A1 PCT/CN2017/114568 CN2017114568W WO2018113512A1 WO 2018113512 A1 WO2018113512 A1 WO 2018113512A1 CN 2017114568 W CN2017114568 W CN 2017114568W WO 2018113512 A1 WO2018113512 A1 WO 2018113512A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
sample
preset
feature
trained
Prior art date
Application number
PCT/CN2017/114568
Other languages
English (en)
French (fr)
Inventor
郑永森
黄凯宁
Original Assignee
腾讯科技（深圳）有限公司
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 腾讯科技（深圳）有限公司 filed Critical 腾讯科技（深圳）有限公司
Publication of WO2018113512A1 publication Critical patent/WO2018113512A1/zh
Priority to US16/356,346 priority Critical patent/US10956783B2/en

Links

Images

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
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/04—Context-preserving transformations, e.g. by using an importance map
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/001—Texturing; Colouring; Generation of texture or colour
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/60—Editing figures and text; Combining figures or text
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4046—Scaling of whole images or parts thereof, e.g. expanding or contracting using neural networks
    • 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
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • 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
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
    • G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
    • G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/161—Detection; Localisation; Normalisation

Definitions

  • the present application relates to the field of computer technology, and in particular to image processing.
  • a common picture can be converted into a artistic rendering with a specific style through style conversion.
  • the style conversion is mainly based on the neural network artificial intelligence algorithm to the background of the ordinary picture.
  • White purification treatment is carried out, and then combined with the artistic style of the painting genre, and finally the image is intelligently processed to obtain an artistic rendering.
  • the more commonly used artificial intelligence algorithm is the style network method.
  • the artistic effect picture after stylized processing can be quickly obtained.
  • the relevant parameters in the offline style conversion model are fixed, some parameters do not affect the overall result output, which will cause computational redundancy and space waste.
  • the solution supporting the offline style conversion model requires about 13 megabytes (English full name: MByte, English abbreviation: MB) computing resources, so the operation on the mobile terminal application will be too large, and can not support more A forward neural network coexists.
  • the embodiment of the invention provides an image processing method and related device, which can remove some sample images that do not satisfy the preset image feature value extraction condition in the training model, thereby effectively reducing computational redundancy and space waste.
  • This type of model has a high compression ratio and can be applied to mobile terminal applications, thereby improving the practicability of the solution.
  • An embodiment of the present invention provides an image processing method, which is applied to an image processing apparatus, and includes:
  • the image to be processed is processed by using a preset training model, wherein the preset training model is a functional relationship model between an image of the feature sample and an activation function of the feature sample image, and the image of the feature sample includes a preset An image of an image feature value extraction condition;
  • Another aspect of the embodiments of the present invention provides an image processing apparatus, the apparatus comprising:
  • a first acquiring module configured to acquire an image to be processed
  • a processing module configured to process, by using a preset training model, the image to be processed acquired by the first acquiring module, where the preset training model is a function of an activation function of the feature sample image and the feature sample image a relationship model, wherein the feature sample image includes an image that satisfies a preset image feature value extraction condition;
  • a second acquiring module configured to acquire, according to the processing result obtained by the processing module by the preset training model, the target image corresponding to the image to be processed.
  • Another aspect of the embodiments of the present invention provides an image processing apparatus, the apparatus comprising:
  • the memory is configured to store program code and transmit the program code to the processor
  • the processor is configured to execute the image processing method according to an instruction in the program code.
  • Another aspect of an embodiment of the present invention provides a storage medium for storing program code for executing the image processing method described above.
  • Another aspect of an embodiment of the present invention provides a computer program product comprising instructions which, when run on a computer, cause the computer to perform the image processing method described above.
  • an image processing method is provided.
  • the image processing device first acquires an image to be processed; secondly, the image to be processed is processed by using a preset training model, wherein the preset training model is a feature sample image and feature. a function relationship model of the activation function of the sample image, and the feature sample image includes an image that satisfies the preset image feature value extraction condition; finally, the image processing device acquires the target image corresponding to the image to be processed according to the processing result of the preset training model .
  • FIG. 1 is a schematic diagram showing an effect of performing style conversion on a picture according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram showing the effect of style conversion using the style network method in the prior art
  • FIG. 3 is a schematic structural diagram of a preset training model according to an embodiment of the present invention.
  • FIG. 4 is a block diagram showing the hardware structure of an image processing apparatus according to an embodiment of the present invention.
  • FIG. 5 is a schematic diagram of an image processing method according to an embodiment of the present invention.
  • FIG. 6 is a schematic diagram showing images of a plurality of convolution samples in accordance with an embodiment of the present invention.
  • FIG. 7 is a schematic diagram showing a plurality of sample images in a to-be-trained image set according to an embodiment of the present invention.
  • FIG. 8 is a schematic diagram showing semi-automatic selection of sample images in accordance with an embodiment of the present invention.
  • FIG. 9 is a schematic diagram of a template of a target region of interest in an embodiment of the present invention.
  • FIG. 10 is a schematic flow chart of automatically selecting a sample image according to an embodiment of the present invention.
  • FIG. 11 is a flow chart showing a process of a feature sample image assistant module in an application scenario according to the present invention.
  • FIG. 12 is a schematic diagram of analysis of a visualization technique based on a nonlinear activation function in an application scenario
  • FIG. 13 is a schematic diagram of an image processing apparatus according to an embodiment of the present invention.
  • FIG. 14 is a schematic diagram showing another image processing apparatus according to an embodiment of the present invention.
  • FIG. 15 is a schematic diagram showing another image processing apparatus according to an embodiment of the present invention.
  • FIG. 16 is a schematic diagram showing another image processing apparatus according to an embodiment of the present invention.
  • FIG. 17 is a diagram showing another image processing apparatus according to an embodiment of the present invention.
  • FIG. 18 is a schematic diagram showing another image processing apparatus according to an embodiment of the present invention.
  • FIG. 19 is a schematic diagram showing another image processing apparatus according to an embodiment of the present invention.
  • FIG. 20 is a block diagram showing the structure of an image processing apparatus according to an embodiment of the present invention.
  • the embodiment of the invention provides an image processing method and related device, which can remove some sample images that do not satisfy the preset image feature value extraction condition in the training model, thereby effectively reducing computational redundancy and space waste, and at the same time
  • the class model has a high compression ratio and can be applied to mobile terminal applications, thereby improving the practicability of the solution.
  • FIG. 3 is a schematic structural diagram of a preset training model according to an embodiment of the present invention, and a style network.
  • the traditional offline style conversion model adopted by the method is different.
  • the structure of the preset training model in the present invention is "323 structure", that is, three convolution layers, two residual layers, and three deconvolution layers.
  • the traditional offline style conversion model has 3 convolutional layers, 5 residual layers and 3 deconvolution layers. The relevant parameters in the model are fixed, but some parameters have no effect on the overall output. Will result in computational redundancy and wasted space.
  • Preset training of the present invention The model is effectively compressed without changing the overall output result, eliminating the image that does not satisfy the preset image feature value extraction condition, reducing computational redundancy and space occupation, and the preset training model of the present invention is effectively compressed. It can usually be below 1MB, so it can be easily applied to mobile terminals. It should be noted that the present invention uses only two residual layers as one indication. In practical applications, the number of residual layers can also be changed, for example, three residual layers or four residual layers are used.
  • the parameters of the convolutional layer can actually be seen as a series of filters that can be trained or learned.
  • the neurons in each layer of the convolutional neural network are only connected to some local areas of the upper layer.
  • the forward calculation process we first input the image data in the target input image to the trainable filter for filtering. After convolving the image data, the convolved sample image is generated, and then the pixels in the convolved sample image are summed, weighted and biased, and then an activation function is used to obtain the sample image.
  • the convolutional layer needs to be convolution (English full name: Convolution, English abbreviation: Conv) to calculate, modify the nonlinear activation function (English full name: rectified linear units, English abbreviation: ReLU) calculation and batch standardization (English Full name: batch normalization, English abbreviation: BN) calculation.
  • the first feature map can be obtained by Conv calculation, then the first feature map is calculated by ReLU, and the second feature map is obtained.
  • the ReLU calculation can filter out the valuable activation signals by using the activation function.
  • BN can be tried when the neural network training encounters a slow convergence rate, or a situation such as a gradient explosion that cannot be trained.
  • the BN algorithm can be added to speed up the training and improve the accuracy of the model under normal use.
  • the residual layer contains important information about the basic assumptions of the model. If the regression model is correct, we can regard the residual as the observation of the error, it should conform to the assumptions of the model, and have some properties of the error, using the information provided by the residual to examine the rationality and data of the model hypothesis. The reliability is called residual analysis.
  • the deconvolution network can be seen as the inverse of the convolutional network and is used for unsupervised learning. However, the deconvolution process does not have the ability to learn. It is only used to visualize a trained convolutional network model without the process of learning and training.
  • the deconvolution visualization takes the feature map obtained by each layer as input, performs deconvolution, and obtains a deconvolution result to verify the feature map extracted by each layer.
  • FIG. 4 is a schematic diagram showing the hardware structure of an image processing apparatus according to an embodiment of the present invention. As shown in FIG. 4, the image processing apparatus may include:
  • the processor 1, the communication interface 2, the memory 3 and the display screen 5 complete communication with each other via the communication bus 4.
  • FIG. 5 is a schematic diagram of an image processing method according to an embodiment of the present invention, which includes the following steps:
  • the image processing apparatus first acquires at least one image to be processed, and the image to be processed can also be the original image of the input.
  • the image processing apparatus needs to perform the background white purification processing on the original image of the input, that is, the input original is eliminated as much as possible. The noise in the figure.
  • the preset processing model is used to process the processed image, wherein the preset training model is a functional relationship model of the activation function of the feature sample image and the feature sample image, and the feature sample image includes an image that satisfies the preset image feature value extraction condition. ;
  • the image processing device uses the preset training model to process the image to be processed, and specifically, the image to be processed is input to the preset training model that has been trained.
  • the preset training model is actually a functional relationship model between the feature sample image and the activation function of the feature sample image, but the feature sample images are sample images obtained after screening, and satisfy the preset image feature value extraction condition, not All sample images are used for training to reduce the computational resources used in training.
  • the image processing apparatus obtains the output processing result according to the preset training model, and the processing result is the target image corresponding to the image to be processed, and the target image is an image that has been subjected to white purification processing, and can be used for Combined with other pictures, the image is intelligently processed to obtain an artistic rendering.
  • an image processing method is provided.
  • the image processing device first acquires an image to be processed, and then processes the image to be processed by using a preset training model, wherein the preset training model is a feature sample image and a feature sample image.
  • the function relationship model of the function is activated, and the image of the feature sample image includes an image that satisfies the extraction condition of the preset image feature value.
  • the image processing device acquires the target image corresponding to the image to be processed according to the processing result of the preset training model.
  • some sample images that do not satisfy the preset image feature value extraction condition are eliminated in the training model, which can effectively reduce computational redundancy and space waste.
  • such models have high compression ratio and can be applied to mobile. The application of the terminal, thereby improving the usability of the solution.
  • the first image processing method provided by the embodiment of the present invention is based on the embodiment corresponding to FIG. 5 above.
  • the method before the processed image is processed by using the preset training model, the method further includes:
  • the sample image satisfies the preset image feature value extraction condition, the sample image is determined as a feature sample image, and the feature sample image is used to perform training of the preset training model.
  • the preset training model needs to be trained.
  • the image processing apparatus first acquires at least one image set to be trained, each image set to be trained respectively corresponds to a target input image to be trained, and each image set to be trained includes a plurality of sample images, and the sample images are The target input image is obtained after a series of calculations.
  • the image processing apparatus sequentially determines whether each sample image satisfies the preset image feature value extraction condition, and determines the sample image that satisfies the condition as the feature sample image, and after traversing all the sample images in one to-be-trained image set, Then, all the sample images in the next to-be-trained image set are filtered until the feature sample images in each of the to-be-trained image sets are acquired, and the feature sample images are trained to obtain the preset training model.
  • a method for the image processing apparatus to select a feature sample image in advance which mainly acquires a to-be-trained image set including a plurality of sample images, and then extracts according to a preset image feature value extraction condition, which can be used as a training.
  • a preset image feature value extraction condition which can be used as a training.
  • For the feature sample image of the sample only the part of the feature sample image is needed to train the preset training model.
  • the resources of the sample training thereby improving the practicality of the program.
  • the sample image does not satisfy the preset image feature value extraction condition, the sample image is deleted from the image set to be trained.
  • the image processing apparatus determines whether each sample image satisfies the preset image feature value extraction condition one by one, and if it is determined that a certain sample image does not satisfy the preset image feature value extraction condition, then the sample image is taken. Removed from its corresponding set of images to be trained.
  • the training image set After deleting the sample image of the image set to be trained that does not satisfy the preset image feature value extraction condition, The training image set is compressed, thereby reducing the number of parameters used for training, and thus completing the compression task of the preset training model.
  • the sample image that has not passed the screening may also be deleted from the image set to be trained.
  • the sample image that does not meet the requirements in the image set to be trained can be eliminated, thereby compressing the image set to be trained, thereby reducing computational complexity in the sample training process and saving computing resources in the network.
  • obtaining the image set to be trained may include:
  • the target input image is convoluted by a linear filter, and multiple convolution sample images are acquired;
  • a plurality of convolution sample images are calculated using a nonlinear activation function, and a plurality of sample images in the image set to be trained are acquired.
  • the image processing apparatus acquires at least one target input image.
  • a target input image will be described as an example.
  • FIG. 6 shows multiple convolution samples according to an embodiment of the present invention.
  • a schematic diagram of the image, as shown in the figure, is a feature map Ci obtained after calculation by the Conv layer, that is, a convolved sample image.
  • x i,j in the formula (1) represents an element of the i-th row and the j-th column of the target input image, and each weight of the filter is numbered, and w i,j represents a weight value of the i-th row and the j-th column, w b denotes an offset term of the filter, numbering each element of the convolved sample image, a i,j denotes an element of the i-th row and the j-th column of the convolved sample image, and denotes a nonlinear activation function by f, the present invention
  • the nonlinear activation function chosen is the ReLU function.
  • FIG. 7 is a flowchart of an embodiment of the present invention. Schematic diagram of training multiple sample images in an image set.
  • the calculation formula of the ReLU layer is as follows:
  • R j in the formula (2) represents an output result of the j-th RELU layer, that is, a sample image to be employed in the present invention.
  • the numerical values are from blue to red in the corresponding feature image from large to small.
  • the Relu layer can filter out the valuable activation signals after passing the activation function. More convenient.
  • all the sample images in the training image set need to be pre-processed, that is, at least one target input image is acquired first, and the target input image is sequentially After the convolution calculation and the nonlinear activation function, multiple sample images can be obtained.
  • a plurality of sample images are obtained by calculation of a nonlinear activation function, and a nonlinear activation function can filter and process valuable activation signals, making analysis more convenient, and the nonlinear activation function itself is also It has the advantage of fast calculation speed, can alleviate the problem of gradient disappearance, and the sparsity of activation rate.
  • determining whether the sample image in the image set to be trained satisfies the preset image feature value, on the basis of the third embodiment corresponding to FIG. Extraction conditions can include:
  • the image processing apparatus may acquire a region of interest of each sample image in at least one image set to be trained (English name: region of interest, English abbreviation: ROI), wherein the ROI of the sample image is preset by the user.
  • the selection may be selected from the background of the sample image, or may be selected from the layer, and the shape of the selected area is not limited.
  • the ROI layer is used to find the portion of the sample image in which the ROI is activated, wherein, as shown in FIG. 7, the brighter background is activated by the ReLU function, and then the background is brighter.
  • the weight value and the offset value of the sample image are both set to 0. It is not difficult to find in the substitution formula (1) that the sample image does not satisfy the preset image feature value extraction condition.
  • the determination may be made by a first preset threshold, if the ROI brightness value is less than or Equivalent to the first preset threshold, if yes, determining that the sample image satisfies the preset image feature value extraction condition, and if the ROI brightness value is greater than the first preset threshold, the sample image does not satisfy the preset image feature value extraction condition. Thereby culling it.
  • the first preset threshold is a preset brightness preset value, which is usually a value determined according to a series of empirical data, and may be manually set or automatically generated by the device, which is not limited herein. .
  • the image processing apparatus may further acquire the ROI brightness value of the sample image in the image set to be trained in advance, and determine whether the preset image is satisfied by determining whether the ROI brightness value is less than or equal to the first preset threshold.
  • Eigenvalue extraction conditions In the above manner, the sample image is activated based on the nonlinear activation function, and the sample image satisfying the extraction condition of the preset image feature value is selected by acquiring the brightness value of the ROI, and the ROI with higher brightness is activated by the nonlinear activation function. In part, this part will cause the image output to be not clean enough, so these sample images will not be used in the process of training the model, thereby improving the white purification effect of the output image.
  • determining whether the ROI brightness value is less than or equal to the first preset threshold may include :
  • the sample images in the sample image that meet the brightness value less than or equal to the first preset threshold are manually selected, and the sample images are feature sample images.
  • the user can search for the sample image with the background illuminated, wherein the illuminated background is the part where the background is activated, and the image output background is not caused.
  • the reason for the cleanness is therefore the right medicine.
  • the weight value w i,j and the offset value w b of the brighter background image are set to 0, that is, the sample image is erased, and thus the effect of parameter reduction and model compression is also achieved.
  • the sample image with brighter background in FIG. 7 (for example, sample image No. 0, sample image No. 1 and sample image No.
  • FIG. 8 is an embodiment of the present invention.
  • the user manually selects a sample picture in which the ROI luminance value in FIG. 7 is less than or equal to the first preset threshold, thereby triggering the sample extraction instruction, so that the image processing apparatus can know according to the sample extraction instruction.
  • the sample images are less than or equal to the first preset threshold, and it can be determined that the sample images satisfy the preset image feature value extraction condition.
  • a method for semi-automatically selecting a sample image where the image is The device receives the sample images manually selected by the user and models the sample images as feature samples.
  • the user can select a feature sample image that meets the requirements according to the requirements of the preset training model, thereby improving the flexibility and feasibility of the solution.
  • determining whether the ROI brightness value is less than or equal to the first preset threshold may include :
  • FIG. 9 is a schematic diagram of the template of the target region of interest in the embodiment of the present invention.
  • the target ROI area that is, the area on the face of the person is selected, and then the corresponding target ROI template is formed, and the brightness value of the target ROI template is Fm, as shown in the right figure of FIG.
  • FIG. 10 is a schematic flowchart of automatically selecting a sample image according to an embodiment of the present invention, specifically :
  • step 201 the image processing apparatus starts to automatically select the sample image from the image set to be trained
  • step 202 Fn represents the nth sample image of each layer, Fm represents the target ROI template corresponding to the ROI, and N represents the total number of sample images of the layer.
  • Fn-Fm is smaller than T, and T is preset. The second preset threshold, if yes, proceeds to step 203, otherwise, proceeds to step 206;
  • step 203 when Fn-Fm ⁇ T is established, the weight value and the offset value of the sample image Fn are both set to 0;
  • step 204 after the weight value and the offset value of the sample image Fn are both set to 0, the process of judging the next sample image is entered, that is, when n becomes n+1, the next sample can be entered.
  • Image screening After the weight value and the offset value of the sample image Fn are both set to 0, the process of judging the next sample image is entered, that is, when n becomes n+1, the next sample can be entered.
  • step 205 it is determined whether n is less than N. If yes, proceed to step 202 to start a new round of loop determination and calculation. If not, then jump to step 209 to complete automatic selection of the sample image;
  • step 206 the image processing apparatus does not perform any operation on the sample image
  • step 207 the next sample image is continuously judged, that is, n becomes n+1, so that the next sample image is filtered;
  • step 208 it is determined whether n is less than N. If yes, proceed to step 202 to start a new round of loop determination and calculation. If not, then jump to step 209 to complete automatic selection of the sample image;
  • step 209 the image processing apparatus ends the automatic selection of the sample image.
  • a method for automatically selecting a sample image that is, the image processing device obtains a luminance difference value between a ROI luminance value and a target ROI luminance value, and then if the luminance difference value is greater than or equal to The second preset threshold determines that the sample image satisfies the preset image feature value extraction condition.
  • the image sample processing device can automatically extract the feature sample image that satisfies the requirement, thereby improving the convenience and practicability of the solution.
  • FIG. 11 shows the flow of the feature sample image assistant module in the application scenario according to the present invention.
  • the model file loading module 301 is configured to load at least one target input image in the image database, extract one of the target input images in the image input module 302, and then visualize the target input image in the feature map visualization module 303.
  • the processing method is to first calculate the convolved sample image through the Conv layer, and calculate the convolved sample image by using the Relu algorithm to obtain a visualized sample image. Then, the user can select the ROI in the sample image through the region of interest definition module 304. Finally, the observer decision module 305 filters out the image of the feature sample that meets the requirements, and uses the feature sample image to train the preset training model.
  • FIG. 12 is a schematic diagram of a visualization technique based on a nonlinear activation function in an application scenario, as shown in the figure, a visualization technique
  • the analysis mainly includes four steps, which are definition of interest area, search based on Relu algorithm, iterative optimization strategy result generation and observer decision.
  • the black border indicates the region of interest of the image, that is, the ROI
  • the middle map is based on the Relu algorithm to determine the activated ROI portion of the sample image, and then the weight value and the offset value of the sample image are both Set to 0, then perform the same operation on the next sample image to generate optimization results.
  • the image processing apparatus in the present invention is an image processing apparatus applied to an image processing method, and the image processing apparatus 40 includes:
  • a first obtaining module 401 configured to acquire an image to be processed
  • the processing module 402 is configured to process the to-be-processed image acquired by the first acquiring module 401 by using a preset training model, where the preset training model is an activation function of the feature sample image and the feature sample image. a function relationship model, wherein the feature sample image includes an image that satisfies a preset image feature value extraction condition;
  • the second obtaining module 403 is configured to acquire, according to the processing result obtained by the processing module 402 by using the preset training model, the target image corresponding to the image to be processed.
  • the first acquiring module 401 acquires an image to be processed
  • the processing module 402 processes the image to be processed acquired by the first acquiring module 401 by using a preset training model, wherein the preset training model is a function relationship model of the feature sample image and the activation function of the feature sample image, wherein the feature sample image includes an image that satisfies a preset image feature value extraction condition
  • the second acquisition module 403 passes the pre-processing according to the processing module 402.
  • the processing result obtained by the training model is processed, and the target image corresponding to the image to be processed is acquired.
  • the image processing apparatus can implement the following functions: first acquiring an image to be processed, and then processing the image to be processed by using a preset training model, wherein the preset training model is an activation function of the feature sample image and the feature sample image.
  • the function relation model, and the feature sample image includes an image that satisfies the preset image feature value extraction condition
  • the image processing device acquires the target image corresponding to the image to be processed according to the processing result of the preset training model.
  • FIG. 14 is a schematic diagram of another image processing apparatus according to an embodiment of the present invention, in which the image processing apparatus is 40 also includes:
  • the third obtaining module 404 is configured to: before the processing module 402 processes the image to be processed by using a preset training model, acquiring a to-be-trained image set, where the to-be-trained image set includes multiple sample images;
  • the determining module 405 is configured to determine whether the sample image in the to-be-trained image set acquired by the third acquiring module 404 meets the preset image feature value extraction condition;
  • a determining module 406 configured to: if the determining module 405 determines that the sample image meets the preset image The feature value extraction condition determines the sample image as the feature sample image, and the feature sample image is used to perform training of the preset training model.
  • a method for the image processing apparatus to select a feature sample image in advance which mainly acquires a to-be-trained image set including a plurality of sample images, and then extracts according to a preset image feature value extraction condition, which can be used as a training.
  • a preset image feature value extraction condition which can be used as a training.
  • For the feature sample image of the sample only the part of the feature sample image is needed to train the preset training model.
  • the resources of the sample training thereby improving the practicality of the program.
  • FIG. 15 is a schematic diagram of another image processing apparatus according to an embodiment of the present invention. As shown in FIG. The image processing apparatus 40 further includes:
  • a deleting module 407 configured to determine, after the determining module 405 determines whether the sample image in the to-be-trained image set meets the preset image feature value extraction condition, if the sample image does not satisfy the preset image feature value extraction Condition, the sample image is deleted from the set of images to be trained.
  • the sample image that has not passed the screening may also be deleted from the image set to be trained.
  • the sample image that does not meet the requirements in the image set to be trained can be eliminated, thereby compressing the image set to be trained, thereby reducing computational complexity in the sample training process and saving computing resources in the network.
  • FIG. 16 is a schematic diagram of another image processing apparatus according to an embodiment of the present invention, in which the third acquisition is performed.
  • Module 404 includes:
  • a first obtaining unit 4041 configured to acquire a target input image
  • the convolution unit 4042 is configured to perform convolution processing on the target input image acquired by the first acquiring unit 4041 by using a linear filter, and acquire a plurality of convolution sample images;
  • the calculating unit 4043 is configured to calculate, by using a nonlinear activation function, the plurality of convolution sample images obtained by convolution processing by the convolution unit 4042, and acquire a plurality of sample images in the image set to be trained .
  • all the sample images in the training image set need to be pre-processed, that is, at least one target input image is acquired first, and the target input image is sequentially After the convolution calculation and the nonlinear activation function, multiple sample images can be obtained.
  • a plurality of sample images are obtained by calculation of a nonlinear activation function, and a nonlinear activation function can filter and process valuable activation signals, making analysis more convenient, and the nonlinear activation function itself is also It has the advantage of fast calculation speed, can alleviate the problem of gradient disappearance, and the sparsity of activation rate.
  • FIG. 17 is a schematic diagram of another image processing apparatus according to an embodiment of the present invention.
  • the determining module 405 include:
  • a second acquiring unit 4051 configured to acquire a region of interest ROI brightness value of the sample image in the to-be-trained image set
  • the determining unit 4052 is configured to determine whether the ROI brightness value acquired by the second acquiring unit 4051 is less than or equal to a first preset threshold, and if yes, determine that the sample image satisfies the preset image feature value extraction condition.
  • the image processing apparatus may further acquire the ROI brightness value of the sample image in the image set to be trained in advance, and determine whether the preset image is satisfied by determining whether the ROI brightness value is less than or equal to the first preset threshold.
  • Eigenvalue extraction conditions In the above manner, the sample image is activated based on the nonlinear activation function, and the sample image satisfying the extraction condition of the preset image feature value is selected by acquiring the brightness value of the ROI, and the ROI with higher brightness is activated by the nonlinear activation function. In part, this part will cause the image output to be not clean enough, so these sample images will not be used in the process of training the model, thereby improving the white purification effect of the output image.
  • FIG. 18 is a schematic diagram of another image processing apparatus according to an embodiment of the present invention, in which the image processing apparatus is
  • the determining unit 4052 includes:
  • the receiving sub-unit 40521 is configured to receive a user-triggered sample extraction instruction, where the sample extraction instruction is used to indicate that the ROI brightness value of the sample image is less than or equal to the first preset threshold;
  • the determining subunit 40522 is configured to determine, according to the sample extraction instruction received by the receiving subunit 40521, that the ROI luminance value of the sample picture is less than or equal to the first preset threshold.
  • the image processing apparatus receives the sample images manually selected by the user, and performs model training on the sample images as the feature sample images.
  • the user can select a feature sample image that meets the requirements according to the requirements of the preset training model, thereby improving the flexibility and feasibility of the solution.
  • FIG. 19 is a schematic diagram of another image processing apparatus according to an embodiment of the present invention, in which the image processing apparatus is
  • the determining unit 4052 includes:
  • the obtaining sub-unit 40523 is configured to obtain a luminance difference value between the ROI luminance value and the target RIO luminance value, where the target RIO luminance value is preset;
  • the determining sub-unit 40524 determines whether the brightness difference value acquired by the acquiring sub-unit 40523 is greater than or equal to a second preset threshold, and if yes, determining that the ROI brightness value is less than or equal to the first preset threshold.
  • a method for automatically selecting a sample image that is, the image processing device obtains a luminance difference value between a ROI luminance value and a target ROI luminance value, and then if the luminance difference value is greater than or equal to The second preset threshold determines that the sample image satisfies the preset image feature value extraction condition.
  • the image sample processing device can automatically extract the feature sample image that satisfies the requirement, thereby improving the convenience and practicability of the solution.
  • the present invention also provides an image processing apparatus, the apparatus comprising:
  • the memory is configured to store program code and transmit the program code to the processor
  • the processor is configured to execute the image processing method in the method embodiment shown in FIG. 5 described above according to an instruction in the program code.
  • an embodiment of the present invention further provides a storage medium for storing program code for executing the image processing method in the method embodiment shown in FIG. 5 described above.
  • embodiments of the present invention also provide a computer program product comprising instructions that, when run on a computer, cause the computer to perform the image processing method of the method embodiment illustrated in Figure 5 described above.
  • the image processing device may be any terminal including a mobile phone, a tablet computer, a personal digital assistant (Personal Digital Assistant, English abbreviation: PDA), a sales terminal (English name: Point of Sales, English abbreviation: POS), a car computer, and the like. device.
  • PDA Personal Digital Assistant
  • POS Point of Sales
  • car computer and the like. device.
  • the hardware structure of the image processing apparatus provided by the embodiment of the present invention is explained by taking a mobile phone as an example.
  • FIG. 20 is a block diagram showing a partial configuration of an image processing apparatus according to an embodiment of the present invention.
  • the mobile phone includes: radio frequency (English full name: Radio Frequency, English abbreviation: RF) circuit 510, memory 520, input unit 530, display unit 540, sensor 550, audio circuit 560, wireless fidelity (English full name: wireless fidelity , English abbreviation: WiFi) module 570, processor 580, and power supply 590 and other components.
  • radio frequency English full name: Radio Frequency, English abbreviation: RF
  • memory 520 includes: input unit 530, input unit 530, display unit 540, sensor 550, audio circuit 560, wireless fidelity (English full name: wireless fidelity , English abbreviation: WiFi) module 570, processor 580, and power supply 590 and other components.
  • WiFi wireless fidelity
  • processor 580 processor 580
  • power supply 590 power supply 590 and other components.
  • the RF circuit 510 can be used for receiving and transmitting signals during the transmission or reception of information or during a call. Specifically, after receiving the downlink information of the base station, it is processed by the processor 580. In addition, the uplink data is designed to be sent to the base station.
  • the RF circuit 510 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low noise amplifier (English name: Low Noise Amplifier, LNA), a duplexer, and the like.
  • RF circuitry 510 can also communicate with the network and other devices via wireless communication.
  • the above wireless communication may use any communication standard or protocol, including but not limited to the global mobile communication system (English full name: Global System of Mobile communication, English abbreviation: GSM), general packet radio service (English full name: General Packet Radio Service, GPRS) ), code division multiple access (English full name: Code Division Multiple Access, English abbreviation: CDMA), wideband code division multiple access (English full name: Wideband Code Division Multiple Access, English abbreviation: WCDMA), long-term evolution (English full name: Long Term Evolution, English abbreviation: LTE), e-mail, short message service (English full name: Short Messaging Service, SMS).
  • GSM Global System of Mobile communication
  • GPRS General Packet Radio Service
  • CDMA Code Division Multiple Access
  • WCDMA Wideband Code Division Multiple Access
  • LTE Long Term Evolution
  • SMS Short Messaging Service
  • the memory 520 can be used to store software programs and modules, and the processor 580 executes various functional applications and data processing of the mobile phone by running software programs and modules stored in the memory 520.
  • the memory 520 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application required for at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may be stored according to Data created by the use of the mobile phone (such as audio data, phone book, etc.).
  • memory 520 can include high speed random access memory, and can also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device.
  • the input unit 530 can be configured to receive input numeric or character information and to generate key signal inputs related to user settings and function controls of the handset.
  • the input unit 530 may include a touch panel 531 and other input devices 532.
  • the touch panel 531 also referred to as a touch screen, can collect touch operations on or near the user (such as the user using a finger, a stylus, or the like on the touch panel 531 or near the touch panel 531. Operation), and drive the corresponding connecting device according to a preset program.
  • the touch panel 531 can include two parts: a touch detection device and a touch controller.
  • the touch detection device detects the touch orientation of the user, and detects a signal brought by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device, converts the touch information into contact coordinates, and sends the touch information.
  • the touch panel 531 can be implemented in various types such as resistive, capacitive, infrared, and surface acoustic waves.
  • the input unit 530 may also include other input devices 532. Specifically, other input devices 532 may include, but are not limited to, one or more of a physical keyboard, function keys (such as volume control buttons, switch buttons, etc.), trackballs, mice, joysticks, and the like.
  • the display unit 540 can be used to display information input by the user or information provided to the user as well as various menus of the mobile phone.
  • the display unit 540 can include a display panel 541.
  • a liquid crystal display (English name: Liquid Crystal Display, English abbreviation: LCD), an organic light emitting diode (English name: Organic Light-Emitting Diode, English abbreviation: OLED), etc.
  • the display panel 541 is configured in a form.
  • the touch panel 531 can cover the display panel 541. When the touch panel 531 detects a touch operation on or near it, the touch panel 531 transmits to the processor 580 to determine the type of the touch event, and then the processor 580 according to the touch event.
  • the type provides a corresponding visual output on display panel 541.
  • the touch panel 531 and the display panel 541 are used as two independent components to implement the input and input functions of the mobile phone in FIG. 20, in some embodiments, the touch panel 531 and the display panel 541 may be integrated. Realize the input and output functions of the phone.
  • the handset may also include at least one type of sensor 550, such as a light sensor, motion sensor, and other sensors.
  • the light sensor may include an ambient light sensor and a proximity sensor, wherein the ambient light sensor may adjust the brightness of the display panel 541 according to the brightness of the ambient light, and the proximity sensor may close the display panel 541 and/or when the mobile phone moves to the ear. Or backlight.
  • the accelerometer sensor can detect the magnitude of acceleration in all directions (usually three axes). When it is stationary, it can detect the magnitude and direction of gravity.
  • the mobile phone can be used to identify the gesture of the mobile phone (such as horizontal and vertical screen switching, related Game, magnetometer attitude calibration), vibration recognition related functions (such as pedometer, tapping), etc.; as for the mobile phone can also be configured with gyroscopes, barometers, hygrometers, thermometers, infrared sensors and other sensors, no longer Narration.
  • the gesture of the mobile phone such as horizontal and vertical screen switching, related Game, magnetometer attitude calibration
  • vibration recognition related functions such as pedometer, tapping
  • the mobile phone can also be configured with gyroscopes, barometers, hygrometers, thermometers, infrared sensors and other sensors, no longer Narration.
  • Audio circuit 560, speaker 561, and microphone 562 provide an audio interface between the user and the handset.
  • the audio circuit 560 can transmit the converted electrical data of the received audio data to the speaker 561, and convert it into a sound signal output by the speaker 561.
  • the microphone 562 converts the collected sound signal into an electrical signal, and the audio circuit 560 is used by the audio circuit 560. After receiving, it is converted into audio data, and then processed by the audio data output processor 580, sent to the other mobile phone via the RF circuit 510, or outputted to the memory 520 for further processing.
  • WiFi is a short-range wireless transmission technology
  • the mobile phone can help users to send and receive emails, browse web pages, and access streaming media through the WiFi module 570, which provides users with wireless broadband Internet access.
  • FIG. 20 shows the WiFi module 570, it can be understood that it does not belong to the necessary configuration of the mobile device, and is completely It is omitted as needed within the scope of not changing the essence of the invention.
  • the processor 580 is the control center of the handset, and connects various portions of the entire handset using various interfaces and lines, by executing or executing software programs and/or modules stored in the memory 520, and invoking data stored in the memory 520, executing The phone's various functions and processing data, so that the overall monitoring of the phone.
  • the processor 580 may include one or more processing units; preferably, the processor 580 may integrate an application processor and a modem processor, where the application processor mainly processes an operating system, a user interface, an application, and the like.
  • the modem processor primarily handles wireless communications. It will be appreciated that the above described modem processor may also not be integrated into the processor 580.
  • the handset also includes a power source 590 (such as a battery) that supplies power to the various components.
  • a power source 590 such as a battery
  • the power source can be logically coupled to the processor 580 via a power management system to manage functions such as charging, discharging, and power management through the power management system.
  • the mobile phone may further include a camera, a Bluetooth module, and the like, and details are not described herein again.
  • the processor 580 included in the terminal further has the following functions:
  • the image to be processed is processed by using a preset training model, wherein the preset training model is a functional relationship model between an image of the feature sample and an activation function of the feature sample image, and the image of the feature sample includes a preset An image of an image feature value extraction condition;
  • the disclosed system, apparatus, and method may be implemented in other manners.
  • the device embodiments described above are merely illustrative.
  • the division of the unit is only a logical function division.
  • there may be another division manner for example, multiple units or components may be combined or Can be integrated into another system, or some features can be ignored or not executed.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, device or unit, and may be in an electrical, mechanical or other form.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units.
  • the present embodiment can be implemented by selecting some or all of the units according to actual needs. The purpose of the program.
  • each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
  • the above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.
  • the integrated unit if implemented in the form of a software functional unit and sold or used as a standalone product, may be stored in a computer readable storage medium.
  • the technical solution of the present invention which is essential or contributes to the prior art, or all or part of the technical solution, may be embodied in the form of a software product stored in a storage medium.
  • a number of instructions are included to cause a computer device (which may be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention.
  • the foregoing storage medium includes: a U disk, a mobile hard disk, a read only memory (English full name: Read-Only Memory, English abbreviation: ROM), a random access memory (English full name: Random Access Memory, English abbreviation: RAM), magnetic A variety of media that can store program code, such as a disc or a disc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Multimedia (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Human Computer Interaction (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

本发明实施例公开了一种图像处理方法及装置，通过预置训练模型将一些不满足预设图像特征值提取条件的样本图像在训练模型时进行剔除，从而有效地减少计算冗余和和空间浪费，同时这类模型的压缩率较高，能够应用于移动终端的应用程序，以此提升方案的实用性。

Description

图像处理方法以及相关装置
本申请要求于2016年12月21日提交中国专利局、申请号为201611191518.3、申请名称为“一种图像处理的方法以及相关装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及计算机技术领域，具体涉及图像处理。
背景技术
随着图像处理技术的日益成熟，可以通过风格转换将一张普通图片转换为具有特定风格的艺术效果图，如图1所示，风格转换主要是采用神经网络的人工智能算法将普通图片的背景进行白净化处理，然后与绘画流派的艺术风格相结合，最后经过图像智能化处理得到艺术效果图。
目前，较为常用的人工智能算法为风格网络法，如图2所示，通过将原图输入至已经训练好的离线风格转换模型，便可以快速得到风格化处理后的艺术效果图。由于离线风格转换模型中的相关参数固定不变，然而部分参数对整体的结果输出并不影响，这将会造成计算冗余和空间浪费。通常情况下，支持离线风格转换模型的方案需要约13兆(英文全称：MByte，英文缩写：MB)的计算资源，因此在移动终端的应用程序上进行操作就会显得过大，且无法支持多个前向神经网络并存。
发明内容
本发明实施例提供了一种图像处理方法以及相关装置，可以将一些不满足预设图像特征值提取条件的样本图像在训练模型时进行剔除，可以有效地减少计算冗余和和空间浪费，同时这类模型的压缩率较高，能够应用于移动终端的应用程序，从而提升方案的实用性。
本发明实施例一方面提供了一种图像处理方法，该方法应用于图像处理装置中，该方法包括：
获取待处理图像；
采用预置训练模型对所述待处理图像进行处理，其中，所述预置训练模型为特征样本图像与所述特征样本图像的激活函数的函数关系模型，所述特征样本图像中包含满足预设图像特征值提取条件的图像；
根据所述预置训练模型的处理结果，获取所述待处理图像所对应的目标图像。
本发明实施例另一方面提供了一种图像处理装置，该装置包括：
第一获取模块，用于获取待处理图像；
处理模块，用于采用预置训练模型对所述第一获取模块获取的所述待处理图像进行处理，其中，所述预置训练模型为特征样本图像与所述特征样本图像的激活函数的函数关系模型，所述特征样本图像中包含满足预设图像特征值提取条件的图像；
第二获取模块，用于根据所述处理模块通过所述预置训练模型处理得到的处理结果，获取所述待处理图像所对应的目标图像。
本发明实施例另一方面提供了一种图像处理装置，该装置包括：
处理器以及存储器；
所述存储器，用于存储程序代码，并将所述程序代码传输给所述处理器；
所述处理器，用于根据所述程序代码中的指令执行上述图像处理方法。
本发明实施例另一方面提供了一种存储介质，该存储介质用于存储程序代码，程序代码用于执行上述图像处理方法。
本发明实施例另一方面提供了一种包括指令的计算机程序产品，当其在计算机上运行时，使得计算机执行上述图像处理方法。
从以上技术方案可以看出，本发明实施例具有以下优点：
本发明实施例中，提供了一种图像处理方法，首先，图像处理装置先获取待处理图像；其次，采用预置训练模型对待处理图像进行处理，其中，预置训练模型为特征样本图像与特征样本图像的激活函数的函数关系模型，且特征样本图像中包含满足预设图像特征值提取条件的图像；最后，图像处理装置根据预置训练模型的处理结果，获取待处理图像所对应的目标图像。通过上述方法，将一些不满足预设图像特征值提取条件的样本图像在训练模型时进行剔除，可以有效地减少计算冗余和和空间浪费，同时这类模型的压缩率较高，能够应用于移动终端的应用程序，从而提升方案的实用性。
附图说明
图1所示为根据本发明实施例中将图片进行风格转换的效果示意图；
图2所示为现有技术中采用风格网络法进行风格转换的效果示意图；
图3所示为根据本发明实施例中预置训练模型的结构示意图；
图4所示为根据本发明实施例中图像处理装置硬件结构示意图
图5所示为根据本发明实施例中一种图像处理方法的示意图；
图6所示为根据本发明实施例中多个卷积样本图像的示意图；
图7所示为根据本发明实施例中待训练图像集合中多个样本图像的示意图；
图8所示为根据本发明实施例中半自动化选取样本图像的示意图；
图9所示为根据本发明实施例中目标感兴趣区域的模板的示意图；
图10所示为根据本发明实施例中自动选取样本图像的流程示意图；
图11所示为根据本发明应用场景中特征样本图像调参模块的流程设计图；
图12所示为应用场景中基于非线性激活函数的可视化技术分析示意图；
图13所示为根据本发明实施例中一种图像处理装置的示意图；
图14所示为根据本发明实施例中另一种图像处理装置的示意图；
图15所示为根据本发明实施例中另一种图像处理装置的示意图；
图16所示为根据本发明实施例中另一种图像处理装置的示意图；
图17所示为根据本发明实施例中另一种图像处理装置的示意图；
图18所示为根据本发明实施例中另一种图像处理装置的示意图；
图19所示为根据本发明实施例中另一种图像处理装置的示意图；
图20所示为根据本发明实施例中一种图像处理装置的结构示意图。
具体实施方式
本发明实施例提供了一种图像处理方法以及相关装置，可以将一些不满足预设图像特征值提取条件的样本图像在训练模型时进行剔除，可以有效地减少计算冗余和空间浪费，同时这类模型的压缩率较高，能够应用于移动终端的应用程序，从而提升方案的实用性。
应理解，本发明主要采用一种经过改良的预置训练模型进行图像的白净化处理，请参阅图3，图3所示为根据本发明实施例中预置训练模型的结构示意图，与风格网络法所采用的传统离线风格转换模型不同，本发明中的预置训练模型的结构为“323结构”，即3个卷积层，2个残差层以及3个反卷积层。而传统的离线风格转换模型具有3个卷积层，5个残差层以及3个反卷积层，该模型中相关的参数固定不变，但部分参数对整体的输出结果并没有影响，这将造成计算冗余和空间浪费。本发明的预置训练 模型在不改变整体输出结果的前提下，进行了有效压缩，剔除不满足预设图像特征值提取条件的图像，减少了计算冗余和空间占用，同时本发明的预置训练模型通过有效压缩后，通常可以在1MB以下，因此，其可以便捷地应用到移动终端中。需要说明的是，本发明采用2个残差层仅为一个示意，在实际应用中还可以对残差层数量进行改变，例如采用三个残差层或者采用四个残差层。
直观看来，卷积层的参数其实可以看作一系列的可训练或者学习的过滤器。卷积神经网络中每一层的神经元只会和上一层的一些局部区域相连，在前向计算过程中，我们首先将目标输入图像中的图像数据输入至可训练的滤波器，在滤波器对图像数据进行卷积后会产生卷积样本图像，然后对卷积样本图像中的像素再进行求和，加权值以及加偏置处理，接着，再通过一个激活函数就可以得到样本图像。
具体地，在卷积层需要依次进行卷积(英文全称：Convolution，英文缩写：Conv)计算、修正的非线性激活函数(英文全称：rectified linear units，英文缩写：ReLU)计算以及批量规范化(英文全称：batch normalization，英文缩写：BN)计算。利用Conv计算可以得到第一特征图，然后采用ReLU计算第一特征图，并得到第二特征图，相比于Conv计算，ReLU计算由于采用激活函数，因此可以把有价值的激活信号都过滤出来，使得分析起来更加方便。在神经网络训练时遇到收敛速度很慢，或梯度***等无法训练的状况时可以尝试BN来解决。此外，在一般使用情况下也可以加入BN算法来加快训练速度，提高模型精度。
残差层蕴含了有关模型基本假设的重要信息。如果回归模型正确的话，我们可以将残差看作误差的观测值，它应符合模型的假设条件，且具有误差的一些性质，利用残差所提供的信息，来考察模型假设的合理性及数据的可靠性称为残差分析。
反卷积网络可以看成是卷积网络的逆过程，是用于无监督学习的。然而反卷积过程并不具备学习的能力，仅仅是用于可视化一个已经训练好的卷积网络模型，没有学习训练的过程。反卷积可视化以各层得到的特征图作为输入，进行反卷积，得到反卷积结果，用以验证显示各层提取到的特征图。
本发明实施例提供的图像处理方法是基于图像处理装置实现的，在介绍本申请的图像处理方法之前，首先介绍一下图像处理装置，该图像处理装置可以是电脑、笔记本、平板电脑、智能手机等设备，参见图4，图4所示为根据本发明实施例中图像处理装置的硬件结构示意图，如图4所示，该图像处理装置可以包括：
处理器1，通信接口2，存储器3，通信总线4，和显示屏5。
其中，处理器1、通信接口2、存储器3和显示屏5通过通信总线4完成相互间的通信。
下面将从图像处理装置的角度，对本发明中图像处理方法进行介绍，请参阅图5，图5所示为根据本发明实施例中一种图像处理方法的示意图，该方法包括以下步骤：
101、获取待处理图像；
本实施例中，图像处理装置首先获取至少一张待处理图像，待处理图像也可以成为输入的原图，图像处理装置需要将这个输入的原图进行背景白净化处理，也就是尽量消除输入原图中的噪声。
102、采用预置训练模型对待处理图像进行处理，其中，预置训练模型为特征样本图像与特征样本图像的激活函数的函数关系模型，特征样本图像中包含满足预设图像特征值提取条件的图像；
本实施例中，图像处理装置采用预置训练模型对待处理图像进行处理，具体可以是将待处理图像输入至已经训练得到的预置训练模型。
其中，该预置训练模型实际上是特征样本图像与特征样本图像的激活函数的函数关系模型，但是这些特征样本图像是经过筛选后得到的样本图像，且满足预设图像特征值提取条件，并非将所有的样本图像都用于训练，以此减少训练所采用的计算资源。
103、根据预置训练模型的处理结果，获取待处理图像所对应的目标图像。
本实施例中，图像处理装置根据采用的预置训练模型得到输出的处理结果，该处理结果即为待处理图像所对应的目标图像，目标图像为已经经过白净化处理后的图像，可以用于与其他图画相结合，最后经过图像智能化处理得到艺术效果图。
本发明实施例中，提供了一种图像处理方法，图像处理装置先获取待处理图像，然后采用预置训练模型对待处理图像进行处理，其中，预置训练模型为特征样本图像与特征样本图像的激活函数的函数关系模型，且特征样本图像中包含满足预设图像特征值提取条件的图像，最后图像处理装置根据预置训练模型的处理结果，获取待处理图像所对应的目标图像。通过上述方式，将一些不满足预设图像特征值提取条件的样本图像在训练模型时进行剔除，可以有效地减少计算冗余和空间浪费，同时这类模型的压缩率较高，能够应用于移动终端的应用程序，从而提升方案的实用性。
在上述图5对应的实施例的基础上，本发明实施例提供的图像处理方法的第一个 可选实施例中，采用预置训练模型对待处理图像进行处理之前，还可以包括：
获取待训练图像集合，其中，待训练图像集合中包含多个样本图像；
判断待训练图像集合中的样本图像是否满足预设图像特征值提取条件；
若样本图像满足预设图像特征值提取条件，则将样本图像确定为特征样本图像，特征样本图像用于进行预置训练模型的训练。
本实施例中，图像处理装置在采用预置训练模型对待处理图像进行处理之前，还需要对该预置训练模型进行训练。
具体地，首先图像处理装置获取至少一个待训练图像集合，每个待训练图像集合都分别对应一个待训练的目标输入图像，且每个待训练图像集合中包含多个样本图像，这些样本图像是目标输入图像经过一系列计算后得到的。接着由图像处理装置按照顺序一一判断每个样本图像是否满足预设图像特征值提取条件，将满足条件的样本图像确定为特征样本图像，遍历完一个待训练图像集合中的所有样本图像之后，再对下一个待训练图像集合中的所有样本图像进行筛选，直到获取每个待训练图像集合中的特征样本图像，即可采用这些特征样本图像训练得到预置训练模型。
其次，本发明实施例中，说明了图像处理装置预先选取特征样本图像的方法，主要是先获取包含多个样本图像的待训练图像集合，然后根据预设图像特征值提取条件来提取可以作为训练样本的特征样本图像，只需要采用这部分特征样本图像即可训练得到预置训练模型。通过上述方式，限定了在训练预置训练模型的过程中，并非将所有的样本图像都进行训练，而且选择一部分价值较高的样本图像，以此提升了样本训练的准确性，同时也减少了样本训练的资源，进而提升方案的实用性。
在上述图5对应的第一个实施例的基础上，本发明实施例提供的图像处理方法的第二个可选实施例中，判断待训练图像集合中的样本图像是否满足预设图像特征值提取条件之后，还可以包括：
若样本图像不满足预设图像特征值提取条件，则从待训练图像集合中删除样本图像。
本实施例中，图像处理装置按照顺序一一判断每个样本图像是否满足预设图像特征值提取条件，若判断得到某个样本图像不满足预设图像特征值提取条件，那就把这个样本图像从它对应的待训练图像集合中删除。
删除了待训练图像集合中不满足预设图像特征值提取条件的样本图像后，相当于 对待训练图像集合进行了压缩，从而减少了用于训练的参数数量，进而完成了预置训练模型的压缩任务。
再次，本发明实施例中，在图像处理装置根据预设图像特征值提取条件提取特征样本图像的过程中，还可以把没有通过筛选的样本图像从待训练图像集合中删除。通过上述方式，能够剔除待训练图像集合中不满足要求的样本图像，以此来压缩待训练图像集合，从而降低样本训练过程中的计算复杂度以及节省网络中的计算资源。
在上述图5对应的第一个实施例的基础上，本发明实施例提供的图像处理方法的第二个可选实施例中，获取待训练图像集合，可以包括：
获取目标输入图像；
采用线性滤波器对目标输入图像进行卷积处理，并获取到多个卷积样本图像；
采用非线性激活函数对多个卷积样本图像进行计算，并获取到待训练图像集合中的多个样本图像。
本实施例中，将介绍如何对待训练图像集合中的样本图像进行预处理。首先图像处理装置获取至少一张目标输入图像，为了便于理解，下面将以一张目标输入图像为例进行说明，请参阅图6，图6所示为根据本发明实施例中多个卷积样本图像的示意图，如图所示，这是经过Conv层计算后所得到的特征图Ci，即卷积样本图像。
Conv层的计算公式如下：
Figure PCTCN2017114568-appb-000001
其中，式(1)中xi,j表示目标输入图像第i行第j列的元素，对滤波器的每一个权重进行编号，wi,j表示第i行第j列的权重值，wb表示滤波器的偏置项，对卷积样本图像的每个元素进行编号，ai,j表示卷积样本图像第i行第j列的元素，用f表示非线性激活函数，本发明所选用的非线性激活函数为ReLU函数。最后将第i个Conv层的输出结果记为Ci。
经过Conv层的计算之后，接着进行ReLU层的计算，并得到如图7所示的多个样本图像的示意图，其中，样本图像具体为特征图，特征图可以看作是一个函数将数据向量映射到特征空间，先采用多个线性滤波器对目标输入图像进行卷积操作，然后加上一个偏置，最后应用于一个非线性激活函数。请参阅图7，图7为本发明实施例中待 训练图像集合中多个样本图像的示意图，ReLU层的计算公式如下：
Rj＝max(0,Ci)；        (2)
其中，式(2)中Rj表示第j个ReLU层的输出结果，即本发明所要采用的样本图像。
图6和图7中，数值标号从大到小所对应特征图像中从蓝到红，相比Conv层，Relu层由于经过了激活函数之后，可以把有价值的激活信号都过滤出来，分析起来更加方便。
再次，本发明实施例中，在图像处理装置训练特征样本图像之前，还需要预先对待训练图像集合中所有的样本图像进行预处理，即先获取至少一张目标输入图像，在该目标输入图像依次经过卷积计算和非线性激活函数之后，可以得到多个样本图像。通过上述方式，多个样本图像是经过了非线性激活函数计算而得到的，而采用非线性激活函数能够把有价值的激活信号都过滤处理，使得分析起来更加方便，且非线性激活函数本身也具有计算速度快，能够减轻梯度消失问题，以及激活率具有稀疏性的优势。
在上述图5对应的第三个实施例的基础上，本发明实施例提供的图像处理方法的第四个可选实施例中，判断待训练图像集合中的样本图像是否满足预设图像特征值提取条件，可以包括：
获取待训练图像集合中的样本图像的感兴趣区域ROI亮度值；
判断ROI亮度值是否小于或等于第一预置门限，若是，则确定样本图像满足预设图像特征值提取条件。
本实施例中，图像处理装置可以获取至少一个待训练图像集合中每个样本图像的感兴趣区域(英文全称：region of interest，英文缩写：ROI)，其中，样本图像的ROI是由用户预先设定的，可以从样本图像的背景中选取，也可以是从图层中选取，且不对选取区域的形状进行限定。
由于样本图像是经过ReLU层计算得到的，根据ReLU层查找样本图像中ROI被激活的部分，其中，如图7所示，背景较亮的则是被ReLU函数激活的，然后将这些背景较亮的样本图像的权重值和偏置值都设置为0，代入式(1)中就不难发现，这个样本图像不满足预设图像特征值提取条件。
在判断亮度时，具体可以以一个第一预置门限进行判断，如果ROI亮度值小于或 等于第一预置门限，若是，则确定样本图像满足预设图像特征值提取条件，反之，如果ROI亮度值大于第一预置门限，那么这个样本图像就不满足预设图像特征值提取条件，从而将其剔除。其中，第一预置门限为预先设定的一个亮度预置值，通常是根据一系列经验数据确定的一个值，既可以是人为设定的，也可以是设备自动生成的，此处不作限定。
进一步地，本发明实施例中，图像处理装置还可以预先获取待训练图像集合中的样本图像的ROI亮度值，通过判断ROI亮度值是否小于或等于第一预置门限来确定是否满足预设图像特征值提取条件。通过上述方式，采用基于非线性激活函数来激活样本图像，通过获取ROI的亮度值来选择满足预设图像特征值提取条件的样本图像，亮度较高的ROI则是被非线性激活函数来激活的部分，这部分会造成图像输出不够干净，因此在训练模型的过程中将不采用这些样本图像，以此提升输出图像的白净化效果。
在上述图5对应的第四个实施例的基础上，本发明实施例提供的图像处理方法的第五个可选实施例中，判断ROI亮度值是否小于或等于第一预置门限，可以包括：
接收用户触发的样本提取指令，样本提取指令用于指示样本图像的ROI亮度值小于或等于第一预置门限；
根据样本提取指令确定样本图片的ROI亮度值小于或等于第一预置门限。
本实施例中，主要通过手动选取样本图像中符合亮度值小于或等于第一预置门限的样本图像，这些样本图像即为特征样本图像。
为了保持目标图像输出的背景白净化，需要训练得到一个较优的预置训练模型。请继续参阅图7，根据ReLU层计算得到的样本图像，用户可以一一进行查找，寻找背景被点亮的样本图像，其中，被点亮则是背景被激活的部分，也是造成图像输出背景不干净的原因，因此对症下药，将背景较亮的特征图的权重值wi,j和偏置值wb设置为0，即将该样本图像抹杀掉，因此也达到了参数减少和模型压缩的效果。将图7背景较亮的样本图像(例如第0号样本图像，第1号样本图像以及第7号样本图像)设置为0，最后输出的样本图像如图8所示，图8为本发明实施例中半自动化选取样本图像的示意图，用户通过手动选择图7中ROI亮度值小于或等于第一预置门限的样本图片，由此触发样本提取指令，使得图像处理装置可以根据该样本提取指令知道哪些样本图片小于或者等于第一预置门限，即可确定这些样本图像满足预设图像特征值提取条件。
更进一步地，本发明实施例中，提供了一种半自动选取样本图像的方式，图像处 理装置接收用户手动选择的样本图像，并将这些样本图像作为特征样本图像进行模型训练。通过上述方式，用户可以根据需预置训练模型的需求选取符合要求的特征样本图像，从而提升方案的灵活性和可行性。
在上述图5对应的第四个实施例的基础上，本发明实施例提供的图像处理方法的第六个可选实施例中，判断ROI亮度值是否小于或等于第一预置门限，可以包括：
获取ROI亮度值与目标ROI亮度值之间的亮度差值，其中，目标ROI亮度值为预先设定的；
判断亮度差值是否大于或等于第二预置门限，若是，则确定ROI亮度值小于或等于第一预置门限。
本实施例中，在图像处理装置自动选择样本图像中的满足条件的特征样本图像时，需要预先选取一个标准模板，请参阅图9，图9为本发明实施例中目标感兴趣区域的模板示意图，在图9的左图中选择目标ROI区域，即人脸上方的区域，然后做成相应的目标ROI模板，该目标ROI模板的亮度值为Fm，具体如图9的右图所示，若设定的一个第二预置门限为T，并获取样本图像ROI的亮度值Fn，将ROI亮度值Fn与目标ROI亮度值Fm相减，从而得到亮度差值D，如果D小于T，则将该样本图像的权重值和偏置值设置为0，从而自动完成背景白净化处理和自动压缩的任务。
为便于理解，下面可以以一个具体应用场景对本发明中自动选取样本图像流程过程进行详细描述，请参阅图10，图10所示为根据本发明实施例中自动选取样本图像的流程示意图，具体为：
步骤201中，图像处理装置开始从待训练图像集合中进行样本图像的自动选取；
步骤202中，Fn表示每层的第n个样本图像，Fm表示与ROI对应的目标ROI模板，N表示该层样本图像的总个数，首先判断Fn-Fm是否小于T，T为预先设定的第二预置门限，如果是，则进入步骤203，反之，则跳转至步骤206；
步骤203中，当Fn-Fm＜T成立时，则将这张样本图像Fn的权重值和偏置值均设置为0；
步骤204中，在将样本图像Fn的权重值和偏置值均设置为0之后，再进入对下一张样本图像的判断过程，即在n变为n+1时，可以进入下一张样本图像筛选；
步骤205中，判断n是否小于N，若是，则进入步骤202，开始新一轮的循环判断和计算，若否，则跳转至步骤209，完成对样本图像的自动选取；
步骤206中，图像处理装置对这张样本图像不执行任何操作；
步骤207中，继续对下一张样本图像进行判断，即n变为n+1，从而进入下一张样本图像筛选；
步骤208中，判断n是否小于N，若是，则进入步骤202，开始新一轮的循环判断和计算，若否，则跳转至步骤209，完成对样本图像的自动选取；
步骤209中，图像处理装置结束样本图像的自动选取。
更进一步地，本发明实施例中，提供了一种自动选取样本图像的方式，即图像处理装置获取ROI亮度值与目标ROI亮度值之间的亮度差值，然后若亮度差值是否大于或等于第二预置门限，则确定样本图像满足预设图像特征值提取条件。通过上述方式，可以由图像处理装置自动提取满足要求的特征样本图像，从而提升了方案的便利性和实用性。
为便于理解，下面还可以以一个具体应用场景对本发明中图像处理的过程进行详细描述，本方案主要是基于样本图像进行调参，对背景影响较大而对其他的局部特征不影响或者影响较小的样本图像进行权重值和偏置值置零处理，从而达到背景白净化和参数减少的目标，请参阅图11，图11所示为根据本发明应用场景中特征样本图像调参模块的流程设计图，具体为：
模型文件加载模块301用于加载图片数据库中至少一张目标输入图像，在图像输入模块302中提取其中一张目标输入图像，然后在特征图可视化模块303中，将目标输入图像进行可视化处理，具体处理方式为先通过Conv层计算出卷积后的样本图像，在利用Relu算法对卷积后的样本图像进行计算，得到可视化的样本图像。于是，用户可以通过兴趣区域定义模块304来选取样本图像中的ROI，最后，有观察者决定模块305筛选出符合要求的特征样本图像，并采用特征样本图像进行预置训练模型的训练。
其中，在特征图可视化模块303中，可以采用基于Relu算法的可视化分析技术，请参阅图12，图12为应用场景中基于非线性激活函数的可视化技术分析示意图，如图所示，可视化技术的分析主要包括四个步骤，分别为兴趣区域定义、基于Relu算法进行查找、迭代优化策略结果生成和观察者决定。如图12所示，黑色边框表示图像的感兴趣区域，即ROI，中间图为基于Relu算法进行查找，确定出样本图像中被激活ROI部分，然后将该样本图像的权重值和偏置值均设置为0，之后对下一个样本图像执行同样的操作，进而生成优化结果。
下面对本发明中的图像处理装置进行详细描述，请参阅图13，本发明实施例中的图像处理装置为应用于图像处理方法中的图像处理装置，该图像处理装置40包括：
第一获取模块401，用于获取待处理图像；
处理模块402，用于采用预置训练模型对所述第一获取模块401获取的所述待处理图像进行处理，其中，所述预置训练模型为特征样本图像与所述特征样本图像的激活函数的函数关系模型，所述特征样本图像中包含满足预设图像特征值提取条件的图像；
第二获取模块403，用于根据所述处理模块402通过所述预置训练模型处理得到的处理结果，获取所述待处理图像所对应的目标图像。
本实施例中，第一获取模块401获取待处理图像，处理模块402采用预置训练模型对所述第一获取模块401获取的所述待处理图像进行处理，其中，所述预置训练模型为特征样本图像与所述特征样本图像的激活函数的函数关系模型，所述特征样本图像中包含满足预设图像特征值提取条件的图像，第二获取模块403根据所述处理模块402通过所述预置训练模型处理得到的处理结果，获取所述待处理图像所对应的目标图像。
本发明实施例提供的图像处理装置可以实现以下功能：先获取待处理图像，然后采用预置训练模型对待处理图像进行处理，其中，预置训练模型为特征样本图像与特征样本图像的激活函数的函数关系模型，且特征样本图像中包含满足预设图像特征值提取条件的图像，最后图像处理装置根据预置训练模型的处理结果，获取待处理图像所对应的目标图像。通过上述装置，将一些不满足预设图像特征值提取条件的样本图像在训练模型时进行剔除，可以有效地减少计算冗余和和空间浪费，同时这类模型的压缩率较高，能够应用于移动终端的应用程序，从而提升方案的实用性。
在上述图13所对应的实施例的基础上，请参阅图14，图14所示为根据本发明实施例的另一种图像处理装置的示意图，在该图像处理装置中，所述图像处理装置40还包括：
第三获取模块404，用于所述处理模块402采用预置训练模型对所述待处理图像进行处理之前，获取待训练图像集合，其中，所述待训练图像集合中包含多个样本图像；
判断模块405，用于判断第三获取模块404获取的所述待训练图像集合中的样本图像是否满足所述预设图像特征值提取条件；
确定模块406，用于若所述判断模块405判断得到所述样本图像满足所述预设图像 特征值提取条件，则将所述样本图像确定为所述特征样本图像，所述特征样本图像用于进行所述预置训练模型的训练。
其次，本发明实施例中，说明了图像处理装置预先选取特征样本图像的方法，主要是先获取包含多个样本图像的待训练图像集合，然后根据预设图像特征值提取条件来提取可以作为训练样本的特征样本图像，只需要采用这部分特征样本图像即可训练得到预置训练模型。通过上述方式，限定了在训练预置训练模型的过程中，并非将所有的样本图像都进行训练，而且选择一部分价值较高的样本图像，以此提升了样本训练的准确性，同时也减少了样本训练的资源，进而提升方案的实用性。
在上述图14所对应的实施例的基础上，请参阅图15，图15所示为根据本发明实施例的另一种图像处理装置的示意图，如图15所示，该图像处理装置在图14所示装置的基础上，所述图像处理装置40还包括：
删除模块407，用于所述判断模块405判断所述待训练图像集合中的样本图像是否满足所述预设图像特征值提取条件之后，若所述样本图像不满足所述预设图像特征值提取条件，则从所述待训练图像集合中删除所述样本图像。
再次，本发明实施例中，在图像处理装置根据预设图像特征值提取条件提取特征样本图像的过程中，还可以把没有通过筛选的样本图像从待训练图像集合中删除。通过上述方式，能够剔除待训练图像集合中不满足要求的样本图像，以此来压缩待训练图像集合，从而降低样本训练过程中的计算复杂度以及节省网络中的计算资源。
在上述图14所对应的实施例的基础上，请参阅图16，图16所示为根据本发明实施例的另一种图像处理装置的示意图，在该图像处理装置中，所述第三获取模块404包括：
第一获取单元4041，用于获取目标输入图像；
卷积单元4042，用于采用线性滤波器对所述第一获取单元4041获取的所述目标输入图像进行卷积处理，并获取到多个卷积样本图像；
计算单元4043，用于采用非线性激活函数对经过所述卷积单元4042卷积处理得到的所述多个卷积样本图像进行计算，并获取到所述待训练图像集合中的多个样本图像。
再次，本发明实施例中，在图像处理装置训练特征样本图像之前，还需要预先对待训练图像集合中所有的样本图像进行预处理，即先获取至少一张目标输入图像，在该目标输入图像依次经过卷积计算和非线性激活函数之后，可以得到多个样本图像。 通过上述方式，多个样本图像是经过了非线性激活函数计算而得到的，而采用非线性激活函数能够把有价值的激活信号都过滤处理，使得分析起来更加方便，且非线性激活函数本身也具有计算速度快，能够减轻梯度消失问题，以及激活率具有稀疏性的优势。
在上述图16所对应的实施例的基础上，请参阅图17，图17所示为根据本发明实施例的另一种图像处理装置的示意图，在该图像处理装置中，所述判断模块405包括：
第二获取单元4051，用于获取所述待训练图像集合中的所述样本图像的感兴趣区域ROI亮度值；
判断单元4052，用于判断所述第二获取单元4051获取的所述ROI亮度值是否小于或等于第一预置门限，若是，则确定所述样本图像满足所述预设图像特征值提取条件。
进一步地，本发明实施例中，图像处理装置还可以预先获取待训练图像集合中的样本图像的ROI亮度值，通过判断ROI亮度值是否小于或等于第一预置门限来确定是否满足预设图像特征值提取条件。通过上述方式，采用基于非线性激活函数来激活样本图像，通过获取ROI的亮度值来选择满足预设图像特征值提取条件的样本图像，亮度较高的ROI则是被非线性激活函数来激活的部分，这部分会造成图像输出不够干净，因此在训练模型的过程中将不采用这些样本图像，以此提升输出图像的白净化效果。
可选地，在上述图17所对应的实施例的基础上，请参阅图18，图18所示为根据本发明实施例的另一种图像处理装置的示意图，在该图像处理装置中
所述判断单元4052包括：
接收子单元40521，用于接收用户触发的样本提取指令，所述样本提取指令用于指示所述样本图像的ROI亮度值小于或等于所述第一预置门限；
确定子单元40522，用于根据所述接收子单元40521接收的所述样本提取指令确定所述样本图片的ROI亮度值小于或等于所述第一预置门限。
更进一步地，本发明实施例中，图像处理装置接收用户手动选择的样本图像，并将这些样本图像作为特征样本图像进行模型训练。通过上述方式，用户可以根据需预置训练模型的需求选取符合要求的特征样本图像，从而提升方案的灵活性和可行性。
可选地，在上述图17所对应的实施例的基础上，请参阅图19，图19所示为根据本发明实施例中另一种图像处理装置的示意图，在该图像处理装置中
所述判断单元4052包括：
获取子单元40523，用于获取ROI亮度值与目标RIO亮度值之间的亮度差值，其中，所述目标RIO亮度值为预先设定的；
判断子单元40524，判断所述获取子单元40523获取的所述亮度差值是否大于或等于第二预置门限，若是，则确定所述ROI亮度值小于或等于所述第一预置门限。
更进一步地，本发明实施例中，提供了一种自动选取样本图像的方式，即图像处理装置获取ROI亮度值与目标ROI亮度值之间的亮度差值，然后若亮度差值是否大于或等于第二预置门限，则确定样本图像满足预设图像特征值提取条件。通过上述方式，可以由图像处理装置自动提取满足要求的特征样本图像，从而提升了方案的便利性和实用性。
另外，本发明还提供一种图像处理装置，该装置包括：
处理器以及存储器；
所述存储器，用于存储程序代码，并将所述程序代码传输给所述处理器；
所述处理器，用于根据所述程序代码中的指令执行上文描述的图5所示的方法实施例中的图像处理方法。
此外，本发明实施例还提供一种存储介质，该存储介质用于存储程序代码，所述程序代码用于执行上文描述的图5所示的方法实施例中的图像处理方法。
另一方面，本发明实施例还提供了一种包括指令的计算机程序产品，当其在计算机上运行时，使得计算机执行上文描述的图5所示的方法实施例中的图像处理方法。
图像处理方法本发明实施例还提供了另一种图像处理装置，如图20所示，为了便于说明，仅示出了与本发明实施例相关的部分，具体技术细节未揭示的，请参照本发明实施例方法部分。该图像处理装置可以为包括手机、平板电脑、个人数字助理(英文全称：Personal Digital Assistant，英文缩写：PDA)、销售终端(英文全称：Point of Sales，英文缩写：POS)、车载电脑等任意终端设备。以手机为例对本发明实施例提供的图像处理装置的硬件结构进行解释说明。
图20示出的是与本发明实施例提供的图像处理装置的部分结构的框图。参考图20，手机包括：射频(英文全称：Radio Frequency，英文缩写：RF)电路510、存储器520、输入单元530、显示单元540、传感器550、音频电路560、无线保真(英文全称：wireless fidelity，英文缩写：WiFi)模块570、处理器580、以及电源590等部件。 本领域技术人员可以理解，图20中示出的手机结构并不构成对图像处理装置的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。
下面结合图20对手机的各个构成部件进行具体的介绍：
RF电路510可用于收发信息或通话过程中，信号的接收和发送，特别地，将基站的下行信息接收后，给处理器580处理；另外，将设计上行的数据发送给基站。通常，RF电路510包括但不限于天线、至少一个放大器、收发信机、耦合器、低噪声放大器(英文全称：Low Noise Amplifier，英文缩写：LNA)、双工器等。此外，RF电路510还可以通过无线通信与网络和其他设备通信。上述无线通信可以使用任一通信标准或协议，包括但不限于全球移动通讯***(英文全称：Global System of Mobile communication，英文缩写：GSM)、通用分组无线服务(英文全称：General Packet Radio Service，GPRS)、码分多址(英文全称：Code Division Multiple Access，英文缩写：CDMA)、宽带码分多址(英文全称：Wideband Code Division Multiple Access,英文缩写：WCDMA)、长期演进(英文全称：Long Term Evolution，英文缩写：LTE)、电子邮件、短消息服务(英文全称：Short Messaging Service，SMS)等。
存储器520可用于存储软件程序以及模块，处理器580通过运行存储在存储器520的软件程序以及模块，从而执行手机的各种功能应用以及数据处理。存储器520可主要包括存储程序区和存储数据区，其中，存储程序区可存储操作***、至少一个功能所需的应用程序(比如声音播放功能、图像播放功能等)等；存储数据区可存储根据手机的使用所创建的数据(比如音频数据、电话本等)等。此外，存储器520可以包括高速随机存取存储器，还可以包括非易失性存储器，例如至少一个磁盘存储器件、闪存器件、或其他易失性固态存储器件。
输入单元530可用于接收输入的数字或字符信息，以及产生与手机的用户设置以及功能控制有关的键信号输入。具体地，输入单元530可包括触控面板531以及其他输入设备532。触控面板531，也称为触摸屏，可收集用户在其上或附近的触摸操作(比如用户使用手指、触笔等任何适合的物体或附件在触控面板531上或在触控面板531附近的操作)，并根据预先设定的程式驱动相应的连接装置。可选的，触控面板531可包括触摸检测装置和触摸控制器两个部分。其中，触摸检测装置检测用户的触摸方位，并检测触摸操作带来的信号，将信号传送给触摸控制器；触摸控制器从触摸检测装置上接收触摸信息，并将它转换成触点坐标，再送给处理器580，并能接收处理器580 发来的命令并加以执行。此外，可以采用电阻式、电容式、红外线以及表面声波等多种类型实现触控面板531。除了触控面板531，输入单元530还可以包括其他输入设备532。具体地，其他输入设备532可以包括但不限于物理键盘、功能键(比如音量控制按键、开关按键等)、轨迹球、鼠标、操作杆等中的一种或多种。
显示单元540可用于显示由用户输入的信息或提供给用户的信息以及手机的各种菜单。显示单元540可包括显示面板541，可选的，可以采用液晶显示器(英文全称：Liquid Crystal Display，英文缩写：LCD)、有机发光二极管(英文全称：Organic Light-Emitting Diode，英文缩写：OLED)等形式来配置显示面板541。进一步的，触控面板531可覆盖显示面板541，当触控面板531检测到在其上或附近的触摸操作后，传送给处理器580以确定触摸事件的类型，随后处理器580根据触摸事件的类型在显示面板541上提供相应的视觉输出。虽然在图20中，触控面板531与显示面板541是作为两个独立的部件来实现手机的输入和输入功能，但是在某些实施例中，可以将触控面板531与显示面板541集成而实现手机的输入和输出功能。
手机还可包括至少一种传感器550，比如光传感器、运动传感器以及其他传感器。具体地，光传感器可包括环境光传感器及接近传感器，其中，环境光传感器可根据环境光线的明暗来调节显示面板541的亮度，接近传感器可在手机移动到耳边时，关闭显示面板541和/或背光。作为运动传感器的一种，加速计传感器可检测各个方向上(一般为三轴)加速度的大小，静止时可检测出重力的大小及方向，可用于识别手机姿态的应用(比如横竖屏切换、相关游戏、磁力计姿态校准)、振动识别相关功能(比如计步器、敲击)等；至于手机还可配置的陀螺仪、气压计、湿度计、温度计、红外线传感器等其他传感器，在此不再赘述。
音频电路560、扬声器561，传声器562可提供用户与手机之间的音频接口。音频电路560可将接收到的音频数据转换后的电信号，传输到扬声器561，由扬声器561转换为声音信号输出；另一方面，传声器562将收集的声音信号转换为电信号，由音频电路560接收后转换为音频数据，再将音频数据输出处理器580处理后，经RF电路510以发送给比如另一手机，或者将音频数据输出至存储器520以便进一步处理。
WiFi属于短距离无线传输技术，手机通过WiFi模块570可以帮助用户收发电子邮件、浏览网页和访问流式媒体等，它为用户提供了无线的宽带互联网访问。虽然图20示出了WiFi模块570，但是可以理解的是，其并不属于移动设备的必须构成，完全可 以根据需要在不改变发明的本质的范围内而省略。
处理器580是手机的控制中心，利用各种接口和线路连接整个手机的各个部分，通过运行或执行存储在存储器520内的软件程序和/或模块，以及调用存储在存储器520内的数据，执行手机的各种功能和处理数据，从而对手机进行整体监控。可选的，处理器580可包括一个或多个处理单元；优选的，处理器580可集成应用处理器和调制解调处理器，其中，应用处理器主要处理操作***、用户界面和应用程序等，调制解调处理器主要处理无线通信。可以理解的是，上述调制解调处理器也可以不集成到处理器580中。
手机还包括给各个部件供电的电源590(比如电池)，优选的，电源可以通过电源管理***与处理器580逻辑相连，从而通过电源管理***实现管理充电、放电、以及功耗管理等功能。
尽管未示出，手机还可以包括摄像头、蓝牙模块等，在此不再赘述。
在本发明实施例中，该终端所包括的处理器580还具有以下功能：
获取待处理图像；
采用预置训练模型对所述待处理图像进行处理，其中，所述预置训练模型为特征样本图像与所述特征样本图像的激活函数的函数关系模型，所述特征样本图像中包含满足预设图像特征值提取条件的图像；
根据所述预置训练模型的处理结果，获取所述待处理图像所对应的目标图像。
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的***，装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
在本申请所提供的几个实施例中，应该理解到，所揭露的***，装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个***，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例 方案的目的。
另外，在本发明各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本发明的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本发明各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(英文全称：Read-Only Memory，英文缩写：ROM)、随机存取存储器(英文全称：Random Access Memory，英文缩写：RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
在本发明中描述的术语“第一”、“第二”、“第三”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本发明的实施例例如能够以除了在这里图示或描述的那些以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、***、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
以上所述，以上实施例仅用以说明本发明的技术方案，而非对其限制；尽管参照前述实施例对本发明进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本发明各实施例技术方案的精神和范围。

Claims (18)

  1. 一种图像处理方法，应用于图像处理装置中，包括：
    获取待处理图像；
    采用预置训练模型对所述待处理图像进行处理，其中，所述预置训练模型为特征样本图像与所述特征样本图像的激活函数的函数关系模型，所述特征样本图像中包含满足预设图像特征值提取条件的图像；
    根据所述预置训练模型的处理结果，获取所述待处理图像所对应的目标图像。
  2. 根据权利要求1所述的方法，所述采用预置训练模型对所述待处理图像进行处理之前，所述方法还包括：
    获取待训练图像集合，其中，所述待训练图像集合中包含多个样本图像；
    判断所述待训练图像集合中的样本图像是否满足所述预设图像特征值提取条件；
    若所述样本图像满足所述预设图像特征值提取条件，则将所述样本图像确定为所述特征样本图像，所述特征样本图像用于进行所述预置训练模型的训练。
  3. 根据权利要求2所述的方法，所述判断所述待训练图像集合中的样本图像是否满足所述预设图像特征值提取条件之后，所述方法还包括：
    若所述样本图像不满足所述预设图像特征值提取条件，则从所述待训练图像集合中删除所述样本图像。
  4. 根据权利要求2所述的方法，所述获取待训练图像集合，包括：
    获取目标输入图像；
    采用线性滤波器对所述目标输入图像进行卷积处理，并获取到多个卷积样本图像；
    采用非线性激活函数对所述多个卷积样本图像进行计算，并获取到所述待训练图像集合中的多个样本图像。
  5. 根据权利要求2所述的方法，所述判断所述待训练图像集合中的样本图像是否满足所述预设图像特征值提取条件，包括：
    获取所述待训练图像集合中的所述样本图像的感兴趣区域ROI亮度值；
    判断所述ROI亮度值是否小于或等于第一预置门限，若是，则确定所述样本图像满足所述预设图像特征值提取条件。
  6. 根据权利要求5所述的方法，所述判断所述ROI亮度值是否小于或等于第一预置门限，包括：
    接收用户触发的样本提取指令，所述样本提取指令用于指示所述样本图像的ROI亮度值小于或等于所述第一预置门限；
    根据所述样本提取指令确定所述样本图片的ROI亮度值小于或等于所述第一预置门限。
  7. 根据权利要求5所述的方法，所述判断所述ROI亮度值是否小于或等于第一预置门限，包括：
    获取ROI亮度值与目标ROI亮度值之间的亮度差值，其中，所述目标ROI亮度值为预先设定的；
    判断所述亮度差值是否大于或等于第二预置门限，若是，则确定所述ROI亮度值小于或等于所述第一预置门限。
  8. 根据权利要求1所述的方法，其特征在于，所述预置训练模型包括：
    3个卷积层、2个残差层和3个反卷积层。
  9. 一种图像处理装置，包括：
    第一获取模块，用于获取待处理图像；
    处理模块，用于采用预置训练模型对所述第一获取模块获取的所述待处理图像进行处理，其中，所述预置训练模型为特征样本图像与所述特征样本图像的激活函数的函数关系模型，所述特征样本图像中包含满足预设图像特征值提取条件的图像；
    第二获取模块，用于根据所述处理模块通过所述预置训练模型处理得到的处理结果，获取所述待处理图像所对应的目标图像。
  10. 根据权利要求9所述的图像处理装置，所述图像处理装置还包括：
    第三获取模块，用于所述处理模块采用预置训练模型对所述待处理图像进行处理之前，获取待训练图像集合，其中，所述待训练图像集合中包含多个样本图像；
    判断模块，用于判断第三获取模块获取的所述待训练图像集合中的样本图像是否满足所述预设图像特征值提取条件；
    确定模块，用于若所述判断模块判断得到所述样本图像满足所述预设图像特征值提取条件，则将所述样本图像确定为所述特征样本图像，所述特征样本图像用于进行所述预置训练模型的训练。
  11. 根据权利要求10所述的图像处理装置，所述图像处理装置还包括：
    删除模块，用于所述判断模块判断所述待训练图像集合中的样本图像是否满足所 述预设图像特征值提取条件之后，若所述样本图像不满足所述预设图像特征值提取条件，则从所述待训练图像集合中删除所述样本图像。
  12. 根据权利要求10所述的图像处理装置，所述第三获取模块包括：
    第一获取单元，用于获取目标输入图像；
    卷积单元，用于采用线性滤波器对所述第一获取单元获取的所述目标输入图像进行卷积处理，并获取到多个卷积样本图像；
    计算单元，用于采用非线性激活函数对经过所述卷积单元卷积处理得到的所述多个卷积样本图像进行计算，并获取到所述待训练图像集合中的多个样本图像。
  13. 根据权利要求10所述的图像处理装置，所述判断模块包括：
    第二获取单元，用于获取所述待训练图像集合中的所述样本图像的感兴趣区域ROI亮度值；
    判断单元，用于判断所述第二获取单元获取的所述ROI亮度值是否小于或等于第一预置门限，若是，则确定所述样本图像满足所述预设图像特征值提取条件。
  14. 根据权利要求13所述的图像处理装置，所述判断单元包括：
    接收子单元，用于接收用户触发的样本提取指令，所述样本提取指令用于指示所述样本图像的ROI亮度值小于或等于所述第一预置门限；
    确定子单元，用于根据所述接收子单元接收的所述样本提取指令确定所述样本图片的ROI亮度值小于或等于所述第一预置门限。
  15. 根据权利要求13所述的图像处理装置，所述判断单元包括：
    获取子单元，用于获取ROI亮度值与目标RIO亮度值之间的亮度差值，其中，所述目标RIO亮度值为预先设定的；
    判断子单元，判断所述获取子单元获取的所述亮度差值是否大于或等于第二预置门限，若是，则确定所述ROI亮度值小于或等于所述第一预置门限。
  16. 一种存储介质，所述存储介质用于存储程序代码，所述程序代码用于执行权利要求1-8任意一项所述的图像处理方法。
  17. 一种包括指令的计算机程序产品，当其在计算机上运行时，使得所述计算机执行权利要求1-8任意一项所述的图像处理方法。
  18. 一种图像处理装置，包括：
    处理器以及存储器；
    所述存储器，用于存储程序代码，并将所述程序代码传输给所述处理器；
    所述处理器，用于根据所述程序代码中的指令执行权利要求1-8任意一项所述的图像处理方法。
PCT/CN2017/114568 2016-12-21 2017-12-05 图像处理方法以及相关装置 WO2018113512A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/356,346 US10956783B2 (en) 2016-12-21 2019-03-18 Image processing method and apparatus

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201611191518.3 2016-12-21
CN201611191518.3A CN108230232B (zh) 2016-12-21 2016-12-21 一种图像处理的方法以及相关装置

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/356,346 Continuation US10956783B2 (en) 2016-12-21 2019-03-18 Image processing method and apparatus

Publications (1)

Publication Number Publication Date
WO2018113512A1 true WO2018113512A1 (zh) 2018-06-28

Family

ID=62624369

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/114568 WO2018113512A1 (zh) 2016-12-21 2017-12-05 图像处理方法以及相关装置

Country Status (3)

Country Link
US (1) US10956783B2 (zh)
CN (1) CN108230232B (zh)
WO (1) WO2018113512A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110827375A (zh) * 2019-10-31 2020-02-21 湖北大学 一种基于微光图像的红外图像真彩着色方法及***

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10664718B1 (en) * 2017-09-11 2020-05-26 Apple Inc. Real-time adjustment of hybrid DNN style transfer networks
US11694083B2 (en) * 2017-10-15 2023-07-04 Alethio Co. Signal translation system and signal translation method
CN110751283B (zh) * 2018-07-05 2022-11-15 第四范式（北京）技术有限公司 模型解释方法、装置、设备及存储介质
CN110163048B (zh) 2018-07-10 2023-06-02 腾讯科技（深圳）有限公司 手部关键点的识别模型训练方法、识别方法及设备
CN110930302B (zh) * 2018-08-30 2024-03-26 珠海金山办公软件有限公司 一种图片处理方法、装置、电子设备及可读存储介质
CN109685749B (zh) * 2018-09-25 2023-04-18 平安科技（深圳）有限公司 图像风格转换方法、装置、设备和计算机存储介质
US11367163B2 (en) 2019-05-31 2022-06-21 Apple Inc. Enhanced image processing techniques for deep neural networks
CN110675312B (zh) * 2019-09-24 2023-08-29 腾讯科技（深圳）有限公司 图像数据处理方法、装置、计算机设备以及存储介质
CN111598808B (zh) * 2020-05-18 2022-08-23 腾讯科技（深圳）有限公司 图像处理方法、装置、设备及其训练方法
CN113435454B (zh) * 2021-05-21 2023-07-25 厦门紫光展锐科技有限公司 一种数据处理方法、装置及设备
CN113920404A (zh) * 2021-11-09 2022-01-11 北京百度网讯科技有限公司 训练方法、图像处理方法、装置、电子设备以及存储介质

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104215584A (zh) * 2014-08-29 2014-12-17 华南理工大学 一种基于高光谱图像技术区分大米产地的检测方法
US9208567B2 (en) * 2013-06-04 2015-12-08 Apple Inc. Object landmark detection in images
CN106204467A (zh) * 2016-06-27 2016-12-07 深圳市未来媒体技术研究院 一种基于级联残差神经网络的图像去噪方法

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP4257615B2 (ja) * 2006-07-14 2009-04-22 ソニー株式会社 画像処理装置および方法、並びにプログラム
JPWO2012046671A1 (ja) * 2010-10-06 2014-02-24 日本電気株式会社 測位システム
US9396621B2 (en) * 2012-03-23 2016-07-19 International Business Machines Corporation Systems and methods for false alarm reduction during event detection
CN103559504B (zh) * 2013-11-04 2016-08-31 北京京东尚科信息技术有限公司 图像目标类别识别方法及装置
JP6277818B2 (ja) * 2014-03-26 2018-02-14 日本電気株式会社 機械学習装置、機械学習方法、及びプログラム
CN105528638B (zh) * 2016-01-22 2018-04-24 沈阳工业大学 灰色关联分析法确定卷积神经网络隐层特征图个数的方法
CN105809704B (zh) * 2016-03-30 2019-03-15 北京小米移动软件有限公司 识别图像清晰度的方法及装置
KR20180024200A (ko) * 2016-08-29 2018-03-08 오드컨셉 주식회사 영상 검색 정보 제공 방법, 장치 및 컴퓨터 프로그램
US11205103B2 (en) * 2016-12-09 2021-12-21 The Research Foundation for the State University Semisupervised autoencoder for sentiment analysis

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9208567B2 (en) * 2013-06-04 2015-12-08 Apple Inc. Object landmark detection in images
CN104215584A (zh) * 2014-08-29 2014-12-17 华南理工大学 一种基于高光谱图像技术区分大米产地的检测方法
CN106204467A (zh) * 2016-06-27 2016-12-07 深圳市未来媒体技术研究院 一种基于级联残差神经网络的图像去噪方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110827375A (zh) * 2019-10-31 2020-02-21 湖北大学 一种基于微光图像的红外图像真彩着色方法及***
CN110827375B (zh) * 2019-10-31 2023-05-30 湖北大学 一种基于微光图像的红外图像真彩着色方法及***

Also Published As

Publication number Publication date
CN108230232A (zh) 2018-06-29
US10956783B2 (en) 2021-03-23
US20190213444A1 (en) 2019-07-11
CN108230232B (zh) 2021-02-09

Similar Documents

Publication Publication Date Title
WO2018113512A1 (zh) 图像处理方法以及相关装置
CN109784424B (zh) 一种图像分类模型训练的方法、图像处理的方法及装置
CN110210571B (zh) 图像识别方法、装置、计算机设备及计算机可读存储介质
CN111260665B (zh) 图像分割模型训练方法和装置
WO2021135601A1 (zh) 辅助拍照方法、装置、终端设备及存储介质
CN106156807B (zh) 卷积神经网络模型的训练方法及装置
WO2020140772A1 (zh) 一种人脸检测方法、装置、设备以及存储介质
WO2018233438A1 (zh) 人脸特征点跟踪方法、装置、存储介质及设备
CN111813532B (zh) 一种基于多任务机器学习模型的图像管理方法及装置
WO2019020014A1 (zh) 解锁控制方法及相关产品
WO2021098609A1 (zh) 图像检测方法、装置及电子设备
WO2021120875A1 (zh) 搜索方法、装置、终端设备及存储介质
CN111612093A (zh) 一种视频分类方法、视频分类装置、电子设备及存储介质
CN109376781B (zh) 一种图像识别模型的训练方法、图像识别方法和相关装置
CN108921941A (zh) 图像处理方法、装置、存储介质和电子设备
CN109495616B (zh) 一种拍照方法及终端设备
CN112184548A (zh) 图像超分辨率方法、装置、设备及存储介质
CN110162956B (zh) 确定关联账户的方法和装置
CN109086680A (zh) 图像处理方法、装置、存储介质和电子设备
CN107704514A (zh) 一种照片管理方法、装置及计算机可读存储介质
CN109753202B (zh) 一种截屏方法和移动终端
CN114418069A (zh) 一种编码器的训练方法、装置及存储介质
CN110083742B (zh) 一种视频查询方法和装置
CN110097570B (zh) 一种图像处理方法和装置
CN107728920A (zh) 一种复制方法及移动终端

Legal Events

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

Ref document number: 17885300

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17885300

Country of ref document: EP

Kind code of ref document: A1