WO2020199528A1 - 图像处理方法及装置、电子设备和存储介质 - Google Patents

图像处理方法及装置、电子设备和存储介质 Download PDF

Info

Publication number
WO2020199528A1
WO2020199528A1 PCT/CN2019/107844 CN2019107844W WO2020199528A1 WO 2020199528 A1 WO2020199528 A1 WO 2020199528A1 CN 2019107844 W CN2019107844 W CN 2019107844W WO 2020199528 A1 WO2020199528 A1 WO 2020199528A1
Authority
WO
WIPO (PCT)
Prior art keywords
result
segmentation
processing
convolution
feature map
Prior art date
Application number
PCT/CN2019/107844
Other languages
English (en)
French (fr)
Inventor
夏清
黄宁
Original Assignee
北京市商汤科技开发有限公司
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 北京市商汤科技开发有限公司 filed Critical 北京市商汤科技开发有限公司
Priority to JP2021539065A priority Critical patent/JP2022517571A/ja
Priority to SG11202106290TA priority patent/SG11202106290TA/en
Publication of WO2020199528A1 publication Critical patent/WO2020199528A1/zh
Priority to US17/356,398 priority patent/US20210319560A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T1/00—General purpose image data processing
    • G06T1/20—Processor architectures; Processor configuration, e.g. pipelining
    • 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
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/143—Segmentation; Edge detection involving probabilistic approaches, e.g. Markov random field [MRF] modelling
    • 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
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • 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
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
    • 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
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B2576/00—Medical imaging apparatus involving image processing or analysis
    • A61B2576/02—Medical imaging apparatus involving image processing or analysis specially adapted for a particular organ or body part
    • A61B2576/023—Medical imaging apparatus involving image processing or analysis specially adapted for a particular organ or body part for the heart
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20076—Probabilistic image processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20112—Image segmentation details
    • G06T2207/20132—Image cropping
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30048—Heart; Cardiac
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30096—Tumor; Lesion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images

Definitions

  • the present disclosure relates to the field of image processing technology, and in particular to an image processing method and device, electronic equipment, and storage medium.
  • segmentation of regions of interest or target regions is the basis for image analysis and target recognition. For example, through segmentation in medical images, the boundaries between one or more organs or lesions are clearly identified. Accurate segmentation of 3D medical images is essential for many clinical applications.
  • the present disclosure proposes an image processing technical solution.
  • an image processing method including: performing a stepwise convolution process on an image to be processed to obtain a convolution result; obtaining a positioning result through positioning processing according to the convolution result; As a result, deconvolution processing is performed step by step to obtain a deconvolution result; segmentation processing is performed on the deconvolution result to segment the target object from the image to be processed.
  • the stepwise convolution processing of the image to be processed to obtain the convolution result includes: stepwise convolution processing of the image to be processed to obtain at least one feature map with gradually decreasing resolution, as The convolution result.
  • the stepwise convolution processing is performed on the image to be processed to obtain at least one feature map with a gradually decreasing resolution, as the convolution result, including: performing convolution processing on the image to be processed, The obtained feature map is used as the feature map to be convolved; when the resolution of the feature map to be convolved does not reach the first threshold, convolution processing is performed on the feature map to be convolved, and the obtained result is used as the feature map to be convolved again.
  • Convolution feature map when the resolution of the feature map to be convolved reaches the first threshold, all the obtained feature maps with gradually decreasing resolution are used as the convolution result.
  • the obtaining the positioning result through positioning processing according to the convolution result includes: performing segmentation processing according to the convolution result to obtain the segmentation result; according to the segmentation result, the The convolution result is subjected to positioning processing to obtain the positioning result.
  • the performing segmentation processing according to the convolution result to obtain the segmentation result includes: performing segmentation processing on the feature map with the lowest resolution in the convolution result to obtain the segmentation result.
  • the performing positioning processing on the convolution result according to the segmentation result to obtain the positioning result includes: determining that the target object is in the convolution result according to the segmentation result According to the position information corresponding to the position information, perform positioning processing on the convolution result to obtain the positioning result.
  • the determining the position information corresponding to the target object in the convolution result according to the segmentation result includes: reading the coordinate position of the segmentation result; taking the coordinate position as The center of the area is respectively determined in the convolution result, and the area position of the target object in the feature map at each resolution can be completely covered as the corresponding position information of the target object in the convolution result.
  • the performing positioning processing on the convolution result according to the position information to obtain the positioning result includes: performing, according to the position information, performing a determination of each resolution in the convolution result The following feature maps are cut separately to obtain the positioning results.
  • the step-by-step deconvolution processing on the positioning result to obtain the deconvolution result includes: the feature map with the lowest resolution in all the feature maps included in the positioning result As the feature map to be deconvolved; when the resolution of the feature map to be deconvolved does not reach the second threshold, perform deconvolution processing on the feature map to be deconvolved to obtain the deconvolution processing result; In order of increasing resolution, determine the next feature map of the feature map to be deconvolved in the positioning result; fuse the deconvolution processing result with the next feature map, and merge the fusion result As the feature map to be deconvolved again; when the resolution of the feature map to be deconvolved reaches the second threshold, the feature map to be deconvolved is used as the deconvolution result.
  • the segmentation processing includes: regressing the object to be segmented through softmax to obtain a regression result; and completing the segmentation processing of the object to be segmented by comparing the maximum value of the regression result.
  • the method is implemented by a neural network.
  • the neural network includes a first segmentation sub-network and a second segmentation sub-network.
  • the first segmentation sub-network is used to The image is subjected to stepwise convolution processing and segmentation processing, and the second segmentation sub-network is used to perform stepwise deconvolution processing and segmentation processing on the positioning result.
  • the training process of the neural network includes: training the first segmentation sub-network according to a preset training set; according to the preset training set and the trained first segmentation Sub-network, training the second segmentation sub-network.
  • the method before the stepwise convolution processing is performed on the image to be processed to obtain the convolution result, the method further includes: adjusting the image to be processed to a preset resolution.
  • the image to be processed is a three-dimensional medical image.
  • an image processing device including: a convolution module, configured to perform stepwise convolution processing on an image to be processed to obtain a convolution result; and a positioning module, configured to obtain a convolution result according to the convolution result, The positioning result is obtained through positioning processing; the deconvolution module is used to perform stepwise deconvolution processing on the positioning result to obtain the deconvolution result; the target object acquisition module is used to perform segmentation processing on the deconvolution result , Segmenting the target object from the image to be processed.
  • the convolution module is configured to: perform stepwise convolution processing on the image to be processed to obtain at least one feature map with a gradually decreasing resolution as the convolution result.
  • the convolution module is further configured to: perform convolution processing on the image to be processed, and the obtained feature map is used as the feature map to be convolved; When the first threshold is reached, perform convolution processing on the feature map to be convolved, and use the obtained result as the feature map to be convolved again; when the resolution of the feature map to be convolved reaches the first threshold, All feature maps with gradually decreasing resolution are obtained as the convolution result.
  • the positioning module includes: a segmentation sub-module for performing segmentation processing according to the convolution result to obtain a segmentation result; and a positioning sub-module for performing segmentation on the The convolution result is subjected to positioning processing to obtain the positioning result.
  • the segmentation submodule is configured to perform segmentation processing on the feature map with the lowest resolution in the convolution result to obtain a segmentation result.
  • the positioning sub-module is configured to: determine the corresponding position information of the target object in the convolution result according to the segmentation result; Perform positioning processing on the product result to obtain the positioning result.
  • the positioning sub-module is further configured to: read the coordinate position of the segmentation result; use the coordinate position as the center of the area to respectively determine that each resolution in the convolution result
  • the following feature map can completely cover the area position of the target object as the corresponding position information of the target object in the convolution result.
  • the positioning sub-module is further configured to: according to the position information, perform cutting processing on the feature maps at each resolution in the convolution result to obtain the positioning result.
  • the deconvolution module is configured to: use the feature map with the lowest resolution in all feature maps included in the positioning result as the feature map to be deconvolved; When the resolution of the product feature map does not reach the second threshold, perform deconvolution processing on the feature map to be deconvolved to obtain the deconvolution processing result; determine the location result in the order of gradually increasing resolution. The next feature map of the feature map to be deconvolved; the deconvolution processing result is fused with the next feature map, and the fusion result is again used as the feature map to be deconvolved; When the resolution of the convolution feature map reaches the second threshold, the feature map to be deconvolved is used as the deconvolution result.
  • the segmentation processing includes: regressing the object to be segmented through softmax to obtain a regression result; and completing the segmentation processing of the object to be segmented by comparing the maximum value of the regression result.
  • the device is implemented by a neural network
  • the neural network includes a first segmentation sub-network and a second segmentation sub-network, wherein the first segmentation sub-network is used for the processing of the The image is subjected to stepwise convolution processing and segmentation processing, and the second segmentation sub-network is used to perform stepwise deconvolution processing and segmentation processing on the positioning result.
  • the device further includes a training module, configured to: train the first segmentation sub-network according to a preset training set; according to the preset training set and the trained first The segmentation sub-network, and the second segmentation sub-network is trained.
  • a training module configured to: train the first segmentation sub-network according to a preset training set; according to the preset training set and the trained first The segmentation sub-network, and the second segmentation sub-network is trained.
  • a resolution adjustment module is further included, configured to adjust the image to be processed to a preset resolution.
  • the image to be processed is a three-dimensional medical image.
  • an electronic device including: a processor; a memory for storing executable instructions of the processor; wherein the processor is configured to execute the above-mentioned image processing method.
  • a computer-readable storage medium having computer program instructions stored thereon, and the computer program instructions implement the above-mentioned image processing method when executed by a processor.
  • the segmentation result is obtained by performing stepwise convolution processing and segmentation processing on the image to be processed, and the positioning result is obtained based on the segmentation result, and then the segmentation processing is performed after stepwise deconvolution processing is performed on the positioning result.
  • the target object can be segmented from the image to be processed. Through the above process, the positioning and segmentation of the target object can be realized at the same time in the process of one image processing, which improves the accuracy of image processing while ensuring the speed of image processing.
  • Fig. 1 shows a flowchart of an image processing method according to an embodiment of the present disclosure.
  • Fig. 2 shows a flowchart of an image processing method according to an embodiment of the present disclosure.
  • Fig. 3 shows a flowchart of an image processing method according to an embodiment of the present disclosure.
  • Fig. 4 shows a flowchart of an image processing method according to an embodiment of the present disclosure.
  • Fig. 5 shows a flowchart of an image processing method according to an embodiment of the present disclosure.
  • Fig. 6 shows a flowchart of an image processing method according to an embodiment of the present disclosure.
  • Fig. 7 shows a flowchart of an image processing method according to an embodiment of the present disclosure.
  • Fig. 8 shows a schematic diagram of an application example according to the present disclosure.
  • Fig. 9 shows a block diagram of an image processing device according to an embodiment of the present disclosure.
  • Fig. 10 shows a block diagram of an electronic device according to an embodiment of the present disclosure.
  • Fig. 11 shows a block diagram of an electronic device according to an embodiment of the present disclosure.
  • Fig. 1 shows a flowchart of an image processing method according to an embodiment of the present disclosure.
  • the method may be applied to an image processing apparatus, which may be a terminal device, a server, or other processing equipment.
  • terminal devices can be User Equipment (UE), mobile devices, user terminals, terminals, cellular phones, cordless phones, personal digital assistants (PDAs), handheld devices, computing devices, vehicle-mounted devices, and portable devices. Wearable equipment, etc.
  • UE User Equipment
  • PDAs personal digital assistants
  • the image processing method can be implemented by a processor calling computer-readable instructions stored in the memory.
  • the image processing method may include:
  • step S11 the image to be processed is subjected to stepwise convolution processing to obtain a convolution result.
  • Step S12 According to the convolution result, the positioning result is obtained through positioning processing.
  • Step S13 Perform stepwise deconvolution processing on the positioning result to obtain a deconvolution result.
  • Step S14 performing segmentation processing on the deconvolution result, segmenting the target object from the image to be processed.
  • the target object in the image to be processed is preliminarily segmented through stepwise convolution processing and segmentation processing, thereby obtaining a positioning result reflecting the basic distribution position of the target object in the image to be processed, based on this
  • the positioning result can be further processed by stepwise deconvolution and segmentation to achieve high-precision segmentation of the target object in the image to be processed.
  • the segmentation of the target object is achieved on the basis of the positioning result, and the processing is directly processed Compared with the target segmentation of the image, the accuracy of image processing can be effectively improved; at the same time, the above method can realize the target location and segmentation of the image in one image processing process. Because the target location and segmentation process of the image can be combined and analyzed, Therefore, the time consumption of image processing is reduced, and the storage consumption that may exist in the image processing process is also reduced.
  • the image processing method of the embodiment of the present disclosure can be applied to the processing of three-dimensional medical images, for example, to identify a target area in a medical image, and the target area may be an organ, a lesion, a tissue, and so on.
  • the image to be processed may be a three-dimensional medical image of a heart organ. That is to say, the image processing method of the embodiment of the present disclosure may be applied to the treatment of heart disease.
  • the image The processing method can be applied to the treatment of atrial fibrillation. Through the precise segmentation of atrial images, the etiology of atrial fibrosis can be understood and analyzed, and then targeted surgical ablation treatment plans for atrial fibrillation can be formulated to improve the treatment of atrial fibrillation effect.
  • image processing method of the embodiment of the present disclosure is not limited to being applied to three-dimensional medical image processing, and can be applied to any image processing, which is not limited in the present disclosure.
  • the image to be processed may include multiple pictures, and one or more three-dimensional organs can be identified based on the multiple pictures.
  • step S11 is not limited, and any method that can obtain a feature map for segmentation processing can be used as the implementation of step S11.
  • step S11 may include: performing stepwise convolution processing on the image to be processed to obtain at least one feature map with gradually decreasing resolution as the convolution result.
  • FIG. 2 shows a flowchart of an image processing method according to an embodiment of the present disclosure, as shown in the figure. It is shown that in a possible implementation manner, the image to be processed is subjected to stepwise convolution processing to obtain at least one feature map with a gradually decreasing resolution. As a result of the convolution, it may include:
  • Step S111 Perform convolution processing on the image to be processed, and the obtained feature map is used as the feature map to be convolved.
  • step S112 when the resolution of the feature map to be convolved does not reach the first threshold, the feature map to be convolved is subjected to convolution processing, and the obtained result is used as the feature map to be convolved again.
  • step S113 when the resolution of the feature map to be convolved reaches the first threshold, all the feature maps with gradually decreasing resolutions are used as the convolution result.
  • the feature map at the initial resolution can be obtained, and then the feature map at the initial resolution is subjected to another convolution process.
  • the feature map at the next resolution can be obtained, and so on, by performing multiple convolution processing on the image to be processed, a series of feature maps with gradually decreasing resolution can be obtained, and these feature maps can be used as the convolution result for subsequent steps
  • the progress The number of iterations of this process is not limited, and can be stopped when the obtained feature map with the minimum resolution reaches the first threshold.
  • the first threshold can be set according to requirements and actual conditions, and the specific value is not limited here. Since the first threshold is not limited, the number of feature maps and the resolution of each feature map included in the obtained convolution result are not limited, and specific selections can be made according to actual conditions.
  • the process and implementation manner of convolution processing are not limited.
  • the process of convolution processing may include convolution, pooling, and batch normalization (Batch Normalization) or one or more of Parametric Rectified Linear Unit (PReLU, Parametric Rectified Linear Unit).
  • the encoder structure in the 3D U-Net full convolutional neural network can be used to implement, and in one example, it can also be implemented through the encoder structure in the V-Net full roll machine neural network.
  • the present disclosure does not limit the specific method of convolution processing.
  • step S12 may include:
  • Step S121 Perform segmentation processing according to the convolution result to obtain the segmentation result.
  • Step S122 Perform positioning processing on the convolution result according to the segmentation result to obtain the positioning result.
  • step S121 is also not limited. It can be known from the above disclosed embodiment that the convolution result can contain multiple feature maps, so the segmentation result is obtained by segmenting which feature map in the convolution result. , Can be determined according to the actual situation.
  • step S121 may include: performing segmentation processing on the feature map with the lowest resolution in the convolution result to obtain the segmentation result.
  • the processing method of the segmentation processing is not limited, and any method that can segment the target from the feature map can be used as the method of segmentation processing in the example of the present disclosure.
  • the segmentation process can be implemented by the softmax layer to achieve image segmentation.
  • the specific process can include: regression of the object to be segmented through softmax to obtain the regression result; by comparing the maximum value of the regression results, the segmentation is completed Object segmentation processing.
  • the above-mentioned specific process of performing segmentation processing on the object to be segmented by comparing the maximum value of the regression result may be: the form of the regression result may be output data with the same resolution as the object to be segmented, and the output data may be There is a one-to-one correspondence between the pixel position of the object.
  • the output data contains a probability value to indicate the probability of the object to be segmented as the segmentation target at this pixel position.
  • Perform the maximum value comparison to determine whether each pixel position is the segmentation target position, and then realize the operation of extracting the segmentation target from the object to be segmented.
  • the specific method of the maximum value comparison is not limited, and can be set to a value with greater probability
  • the represented pixel position corresponds to the segmentation target, or can be set to correspond to the segmentation target at the pixel position represented by a value with a lower probability, which can be set according to the actual situation, which is not limited here.
  • the process of obtaining the segmentation result may be: pass the feature map with the lowest resolution in the convolution result through the softmax layer, and compare the obtained results with the maximum value to obtain the segmentation result .
  • step S122 Based on the segmentation result, the positioning result can be obtained by performing positioning processing on the convolution result in step S122.
  • the implementation process of step S122 is not limited.
  • FIG. 4 shows a flowchart of an image processing method according to an embodiment of the present disclosure, as shown in FIG. As shown, in a possible implementation manner, step S122 may include:
  • Step S1221 according to the segmentation result, determine the corresponding position information of the target object in the convolution result.
  • Step S1222 Perform positioning processing on the convolution result according to the position information to obtain the positioning result.
  • the location information is information that can indicate the location of the target object in each feature map of the convolution result.
  • the specific form of expression is not limited.
  • the location information can exist in the form of a collection of location coordinates.
  • the position information can exist in the form of coordinates + area, and the expression form of the position information can be flexibly selected according to actual conditions. Since the expression form of the location information is not limited, the specific process of step S1221 can also be flexibly determined according to the expression form of the location information.
  • Fig. 5 shows a flowchart of an image processing method according to an embodiment of the present disclosure. As shown in the figure, in a possible implementation manner, step S1221 may include:
  • Step S12211 Read the coordinate position of the segmentation result.
  • Step S12212 using the coordinate position as the center of the area, respectively determine the area position of the target object in the feature map at each resolution within the convolution result, as the corresponding position information of the target object in the convolution result.
  • the coordinate position of the segmentation result read in step S12211 may be any coordinate indicating the position of the segmentation result.
  • this coordinate may be the coordinate value of a fixed position on the segmentation result; in one example, this The coordinates may be the coordinate values of certain fixed positions on the segmentation result; in one example, this coordinate may be the coordinate value of the center of gravity of the segmentation result.
  • the target object Based on the read coordinate position, the target object can be located at the corresponding position under each feature map in the convolution result through step S12212, and then the position of the area that completely covers the target object is obtained.
  • this area position is the same without limitation, in an example, the expression form of this area position may be the coordinate collection of all vertices of the area, in an example, the expression form of this area position may be the center coordinates of the area position and the coverage area of the area position set.
  • the specific process of step S12212 can be flexibly changed according to the different manifestations of the location of the region.
  • the process of step S12212 can be: based on the barycenter coordinates of the segmentation result in the feature map, and according to the feature map and The resolution ratio of the remaining feature maps in the convolution result can be determined respectively to determine the barycentric coordinates of the target object in each feature map in the convolution result; using the barycentric coordinates as the center, in each feature map, it is determined that the target can be completely covered
  • the area of the object, the vertex coordinates of this area are used as the corresponding position information of the target object in the convolution result. Since there are differences in resolution between the feature maps in the convolution result, there may also be differences in resolution between the regions covering the target object in the feature maps in the convolution result.
  • the convolution result There may be two feature maps A and B.
  • the area covering the target object in the feature map A is marked as area A
  • the area covering the target object in the feature map B is marked as area B.
  • the resolution of the feature map A is the feature If the image resolution is twice that of B, the area of area A is twice that of area B.
  • step S1222 may include: according to the position information, the feature map at each resolution in the convolution result is respectively cropped to obtain the positioning result.
  • the position information may be a set of coordinates of the vertices of each feature map in the convolution result that can cover the target object. Based on this coordinate set, each feature map in the convolution result can be cropped to retain each feature The area covering the target object in the figure is taken as the new feature map, and the set of these new feature maps is the positioning result.
  • the positioning result can be obtained.
  • This process can effectively perform rough positioning of the target object in the feature map at each resolution in the convolution result. Based on this rough positioning, the original The convolution result is processed as the positioning result. Since most of the image information that does not contain the target object is removed from the feature map at each resolution in the positioning result, the storage consumption in the image processing process can be greatly reduced, and the calculation speed can be accelerated. Improve the efficiency and speed of image processing.
  • the effect of target object segmentation based on the positioning result is compared to the effect of directly using the image to be processed for target object segmentation Better, which can improve the accuracy of image processing.
  • the target object can be segmented based on the positioning result.
  • the specific implementation form of segmentation is not limited and can be flexibly selected according to actual conditions.
  • a certain feature map can be selected from the positioning result, and further segmentation processing can be performed to obtain the target object.
  • the positioning result can be used to restore a feature map containing more target object information, and then the feature map can be used for further segmentation processing to obtain the target object.
  • step S13 may include:
  • step S131 the feature map with the lowest resolution in all the feature maps included in the positioning result is used as the feature map to be deconvolved.
  • Step S132 when the resolution of the feature map to be deconvolved does not reach the second threshold, perform deconvolution processing on the feature map to be deconvolved to obtain a deconvolution processing result.
  • Step S133 Determine the next feature map of the feature map to be deconvolved in the positioning result in the order of gradually increasing resolution.
  • Step S134 fusing the deconvolution processing result with the next feature map, and using the fusion result as the feature map to be deconvolved again.
  • Step S135 When the resolution of the feature map to be deconvolved reaches the second threshold, the feature map to be deconvolved is used as the deconvolution result.
  • the deconvolution processing result is the processing result of the deconvolution processing on the feature map to be deconvolved
  • the next feature map is the feature map obtained from the positioning result, that is, the positioning result satisfies
  • a feature map with a resolution greater than the current deconvolution feature map level one condition can be used as the next feature map to be fused with the deconvolution processing result. Therefore, the step-by-step deconvolution process can start from the feature map with the lowest resolution in the positioning result, and through the deconvolution process, the feature map with the resolution increased by one level is obtained. At this time, the resolution can be increased The feature map obtained after the first level is used as the result of the deconvolution processing.
  • the two feature maps contain the effective target object. Therefore, the two feature maps can be fused.
  • the fused feature map contains the effective information of all target objects contained in the two feature maps. Therefore, the fused feature map can be used again as a new Convolution feature map, perform deconvolution processing on this new feature map to be deconvolved, and fuse the processing result with the feature map of the corresponding resolution in the positioning result again until the resolution of the fused feature map reaches the first
  • the threshold is two
  • the deconvolution process is stopped.
  • the final fusion result obtained at this time contains the effective information of the target object contained in each feature map in the positioning result, so it can be used as the deconvolution result.
  • the second threshold is flexibly determined according to the original resolution of the image to be processed, and the specific value is not limited herein.
  • the deconvolution result is obtained by stepwise deconvolution processing on the positioning result, and the deconvolution result is used for the final target object segmentation, so the final result obtained is due to the existence of the target object.
  • Positioning basis so it can effectively contain the global information of the target object, with higher accuracy; and there is no need to segment the image to be processed, but the image processing as a whole, so the processing process is also more efficient; It can be seen from the above process that in a single image processing process, the segmentation of the target object is realized based on the positioning result of the target object.
  • stepwise deconvolution process can help the effective information contained in the feature map at each resolution remain in the final deconvolution result. Since the deconvolution result is used for the final image segmentation, it can greatly Improve the accuracy of the final result.
  • the deconvolution result can be segmented, and the result can be used as the target object segmented from the image to be processed.
  • the segmentation process of the deconvolution result is the same as the above
  • the process of dividing the product result is the same, except that the objects to be divided are different. Therefore, you can refer to the process of the above-mentioned disclosed embodiment, which will not be repeated here.
  • the image processing method of the embodiment of the present disclosure may be implemented through a neural network.
  • the image processing method of the embodiment of the present disclosure mainly includes two segmentation processes, the first is the rough segmentation of the image to be processed, and the second is the update based on the positioning result obtained by the rough segmentation.
  • High-precision segmentation so the second segmentation and the first segmentation can be realized by a neural network, and the two share a set of parameters. Therefore, the two segmentation can be regarded as two sub-neural networks under one neural network.
  • the neural network may include a first segmentation sub-network and a second segmentation sub-network, where the first segmentation sub-network is used to perform stepwise convolution processing and segmentation processing on the image to be processed, and the second segmentation The sub-network is used to perform stepwise deconvolution processing and segmentation processing on the positioning results.
  • the specific network structure adopted by the neural network is not limited.
  • the V-Net and 3D-U-Net mentioned in the above disclosed embodiment can be used as specific implementations of the neural network. Any neural network that can realize the functions of the first segmentation sub-network and the second segmentation sub-network can be used as an implementation form of the neural network.
  • Fig. 7 shows a flowchart of an image processing method according to an embodiment of the present disclosure.
  • the method of the embodiment of the present disclosure may also include a neural network training process, which is denoted as step S15, where step S15 may include:
  • Step S151 training the first segmentation sub-network according to the preset training set.
  • Step S152 training the second segmentation sub-network according to the preset training set and the trained first segmentation sub-network.
  • the preset training set may be a plurality of picture sets that are divided into multiple picture sets after preprocessing such as manual clipping of the sample pictures.
  • Split into multiple picture sets two adjacent picture sets can include a part of the same pictures, for example, taking medical images as an example, multiple samples can be collected from a hospital, and multiple sample pictures included in one sample can be Continuously collected pictures of an organ of the human body.
  • the three-dimensional structure of the organ can be obtained from the multiple sample pictures, which can be split in one direction.
  • the first picture set can include pictures from frames 1-30, and the second picture The set can include pictures from frames 16 to 45..., so that 15 frames of pictures in two adjacent picture sets are the same. Through this overlapping splitting method, the accuracy of the splitting can be improved.
  • the preset training set can be used as input to train the first segmentation sub-network.
  • the pictures in the training set can be located Processing, the training set after positioning processing can be used as the training data of the second segmentation sub-network and input into the second segmentation sub-network for training.
  • the first segmentation sub-network and the second segmentation sub-network after training can be finally obtained. Split the sub-network.
  • the function used to determine the network loss of the neural network is not specifically limited.
  • the dice loss function can be used to determine the network loss of the neural network.
  • the cross-entropy function can be used to determine the neural network loss.
  • the network loss of the network in one example, the network loss of the neural network can also be determined by other available loss functions.
  • the loss functions used by the first segmentation sub-network and the second segmentation sub-network may be the same or different, and are not limited here.
  • the complete training process of the neural network may be: input a preset training set into the network model of the first segmentation sub-network, and the preset training set contains multiple images to be segmented With the mask corresponding to the image to be segmented, the loss between the data output by the network model of the first segmentation sub-network and the corresponding Mask is calculated through any loss function, and then the first segmentation is updated through the back propagation algorithm The network model parameters of the sub-network until the first segmentation sub-network model converges, indicating that the first segmentation sub-network model has completed training.
  • the preset training set is passed through the trained first segmentation sub-network model again to obtain multiple segmentation results.
  • each segment in the first segmentation sub-network is The feature map of the resolution is processed for positioning, and the positioned and cropped feature maps and the mask of the corresponding position are input into the network model of the second segmentation sub-network for training, and the positioning processed image is calculated by any loss function
  • the network model parameters of the second segmentation sub-network are updated through the back propagation algorithm, and the first segmentation sub-network and the second segmentation sub-network are updated alternately. Split the sub-network model parameters until the entire network model converges, and the neural network completes training.
  • the neural network in this disclosure includes two sub-neural networks
  • the two sub-neural networks share the same set of parameters.
  • the method of the embodiment of the present disclosure may further include: adjusting the image to be processed to a preset resolution.
  • the implementation method of adjusting the image to be processed to the preset resolution is not specifically limited.
  • the image to be processed can be adjusted to the preset resolution by a method of center cropping and expansion.
  • the specific resolution value of the preset resolution is also not limited, and can be flexibly set according to the actual situation.
  • the training pictures included in the preset training set can also be unified to a preset resolution before being used for neural network training.
  • the method of the embodiment of the present disclosure may further include: restoring the segmented target object to a space of the same size as the image to be processed to obtain the final segmentation result. Since the resolution of the image to be processed may be adjusted before step S11, the segmentation result obtained may actually be based on the segmentation content of the image after resolution adjustment, so the segmentation result can be restored to the same size as the image to be processed In the space, the segmentation result based on the original image to be processed is obtained.
  • the space of the same size as the image to be processed is not limited. It is determined according to the image nature of the image to be processed and is not limited here.
  • the image to be processed may be a three-dimensional image, so it is of the same size as the image to be processed.
  • the space can be a three-dimensional space.
  • preprocessing the image to be processed may further include: preprocessing the image to be processed.
  • This preprocessing process is not limited, and any processing method that can improve the segmentation accuracy can be used as a process included in the preprocessing.
  • preprocessing the image to be processed may include equalizing the brightness value of the image to be processed.
  • the processing efficiency of subsequent convolution processing, segmentation processing, and deconvolution processing of the subsequent processing image can be improved, and the time of the entire image processing process can be shortened.
  • the accuracy of image segmentation can be improved, thereby improving the accuracy of the image processing result.
  • Heart diseases are currently one of the diseases with the highest mortality rate.
  • atrial fibrillation is one of the most common heart rate disorders.
  • the probability of occurrence in the general population has reached 2%, and the incidence rate in the elderly is even higher. It is high and has a certain fatality rate, which seriously threatens human health.
  • the precise segmentation of the atria is the key to understanding and analyzing atrial fibrosis, and is often used to assist in the development of targeted surgical ablation treatment plans for atrial fibrillation.
  • the segmentation of other cavities of the heart is equally important for the treatment and surgical planning of other types of heart diseases.
  • the method for segmentation of heart cavity in medical images still faces shortcomings such as low accuracy and low computational efficiency.
  • a segmentation method with high accuracy, high efficiency and low space-time consumption can greatly reduce the workload of doctors, improve the quality of heart segmentation, and thereby improve the treatment effect of heart-related diseases.
  • Fig. 8 shows a schematic diagram of an application example according to the present disclosure.
  • an embodiment of the present disclosure proposes an image processing method, which is implemented based on a trained set of neural networks.
  • the specific training process of this neural network can be:
  • the preset training data is processed.
  • the preset training data contains multiple input images and corresponding masks.
  • the resolution of multiple input images is unified to the same size through the method of center cropping and expansion, which is unified in this example
  • the final resolution is 576 ⁇ 576 ⁇ 96.
  • these input images can be used to train the first segmentation sub-network.
  • the specific training process can be:
  • the encoder structure in a three-dimensional full convolutional neural network based on V-Net or 3D-U-Net is adopted to perform multiple convolution processing on the input image.
  • the process of convolution processing in this example can include convolution and pooling , Batch norm and PRelu, through multiple convolution processing, the input of each convolution processing uses the result of the last convolution processing.
  • a total of 4 convolution processing is performed, so the resolution size can be generated respectively 576 ⁇ 576 ⁇ 96, 288 ⁇ 288 ⁇ 48, 144 ⁇ 144 ⁇ 24, and 72 ⁇ 72 ⁇ 12 feature maps, and the feature channels of the input image are increased from 8 to 128;
  • the feature map with the smallest resolution in this example, it is the feature map of 72 ⁇ 72 ⁇ 12.
  • two resolutions of 72 ⁇ 72 ⁇ can be obtained.
  • the probability output of 12 these two probability outputs respectively represent the probability of whether the relevant position of the pixel is the target cavity, these two probability outputs can be used as the output result of the first segmentation sub-network, using dice loss, cross entropy or other loss functions,
  • the loss between the output result and the mask that is directly downsampled to 72 ⁇ 72 ⁇ 12 can be calculated.
  • the backpropagation algorithm can be used to update the network parameters of the first segmentation sub-network until the first segmentation sub-network
  • the convergence of the network model can represent the completion of the first segmentation sub-network training.
  • multiple input images with a uniform resolution can be passed through the first segmentation sub-network that has been trained to obtain a resolution of 576 ⁇ 576 ⁇ 96, 288 ⁇ 288 ⁇ 48, 144 ⁇ Four feature maps of 144 ⁇ 24 and 72 ⁇ 72 ⁇ 12, and two probability outputs with a resolution of 72 ⁇ 72 ⁇ 12. According to the probability output of low resolution, the roughness of the heart cavity can be obtained by comparing with the maximum value.
  • the segmentation result has a resolution of 72 ⁇ 72 ⁇ 12.
  • the center of gravity coordinates of the heart cavity can be calculated and cropped out at 576 ⁇ 576 ⁇ 96, 288 ⁇ 288 ⁇ 48, 144 ⁇ 144 ⁇ 24 and 72 ⁇ 72 ⁇ 12 these four feature maps are enough to completely cover the fixed size area of the target cavity.
  • the 72 ⁇ 72 ⁇ 12 feature map can be cropped by 30 ⁇ 20 ⁇ 12 Large and small area
  • 144 ⁇ 144 ⁇ 24 feature map can be cropped 60 ⁇ 40 ⁇ 24 size area
  • 288 ⁇ 288 ⁇ 48 feature map can be cropped 120 ⁇ 80 ⁇ 48 size area
  • 576 ⁇ 576 ⁇ 96 feature map Can crop 240 ⁇ 160 ⁇ 96 area.
  • these regional images can be used to train the second segmentation sub-network.
  • the specific training process can be:
  • the regional image can be gradually restored to 240 ⁇ 160 ⁇ 96 resolution.
  • the specific process can be: the 30 ⁇ 20 ⁇ 12 size area cropped out of the 72 ⁇ 72 ⁇ 12 feature map is passed
  • the deconvolution process obtains a feature map with a resolution of 60 ⁇ 40 ⁇ 24, and merges this feature map with the 60 ⁇ 40 ⁇ 24 area cropped out of the previous 144 ⁇ 144 ⁇ 24 feature map to obtain the fusion
  • the fusion of the 120 ⁇ 80 ⁇ 48 size area obtained is fused to obtain a feature map with a resolution of 120 ⁇ 80 ⁇ 48.
  • the fused feature map is subjected to deconvolution processing to obtain a resolution of 240 ⁇ 160 ⁇ 96 Then merge it with the 240 ⁇ 160 ⁇ 96 area cropped out of the 576 ⁇ 576 ⁇ 96 feature map to obtain the final image after the stepwise deconvolution processing.
  • This final image contains the heart
  • the local and global information of the cavity the final image is passed through a softmax layer, and two probabilistic outputs with a resolution of 576 ⁇ 576 ⁇ 96 can be obtained. These two probabilistic outputs respectively represent whether the pixel-related position is the target cavity Probability, these two probability outputs can be used as the output result of the second segmentation sub-network, and then using dice loss, cross entropy or other loss functions, the loss between the output result and the mask can be calculated.
  • a trained neural network for heart cavity segmentation can be obtained.
  • the positioning and segmentation of the heart cavity can be completed in the same neural network at the same time, which can be directly obtained from the image input through the network. Therefore, the segmentation process of the heart cavity by the neural network based on this training can be specifically:
  • the image to be segmented in the trained neural network undergoes a process similar to the training process, that is, firstly generate 4 feature maps with resolution size through convolution processing, and then obtain the rough segmentation result. Based on this rough segmentation result, the feature maps of the above 4 resolutions are cropped, and then the cropping result is deconvolved to obtain the deconvolution result.
  • This deconvolution result is then segmented to obtain the segmentation result of the target cavity
  • the segmentation result is output as the output result of the neural network, and then the output segmentation result is mapped to the same dimension as the input image to be segmented, and the final heart cavity segmentation result is obtained.
  • a three-dimensional network can be used to simultaneously perform the positioning and segmentation of the heart cavity.
  • the positioning and segmentation share the same set of parameters, and the positioning and segmentation of the heart cavity are unified into the same network.
  • the segmentation results can be obtained directly in one step, which has a faster speed, can save more storage space, and can obtain a smoother 3D model segmentation surface.
  • the image processing method of the embodiment of the present disclosure is not limited to be applied to the foregoing cardiac cavity image processing, and can be applied to any image processing, which is not limited in the present disclosure.
  • the writing order of the steps does not mean a strict execution order but constitutes any limitation on the implementation process.
  • the specific execution order of each step should be based on its function and possibility.
  • the inner logic is determined.
  • Fig. 9 shows a block diagram of an image processing apparatus according to an embodiment of the present disclosure.
  • the image processing apparatus may be a terminal device, a server, or other processing equipment.
  • terminal devices can be User Equipment (UE), mobile devices, user terminals, terminals, cellular phones, cordless phones, personal digital assistants (PDAs), handheld devices, computing devices, vehicle-mounted devices, and portable devices. Wearable equipment, etc.
  • UE User Equipment
  • PDAs personal digital assistants
  • handheld devices computing devices
  • vehicle-mounted devices vehicle-mounted devices
  • portable devices wearable equipment, etc.
  • the image processing apparatus may be implemented by a processor invoking computer-readable instructions stored in a memory.
  • the image processing device may include: a convolution module 21, which is used to perform stepwise convolution processing on the image to be processed to obtain a convolution result; and a positioning module 22, which is used to perform positioning processing according to the convolution result Obtain the positioning result; the deconvolution module 23 is used to deconvolve the positioning result step by step to obtain the deconvolution result; the target object acquisition module 24 is used to segment the deconvolution result, from the image to be processed The target object is segmented in.
  • a convolution module 21 which is used to perform stepwise convolution processing on the image to be processed to obtain a convolution result
  • a positioning module 22 which is used to perform positioning processing according to the convolution result Obtain the positioning result
  • the deconvolution module 23 is used to deconvolve the positioning result step by step to obtain the deconvolution result
  • the target object acquisition module 24 is used to segment the deconvolution result, from the image to be processed The target object is segmented in.
  • the convolution module is configured to: perform stepwise convolution processing on the image to be processed to obtain at least one feature map with a gradually decreasing resolution as a convolution result.
  • the convolution module is further configured to: perform convolution processing on the image to be processed, and the obtained feature map is used as the feature map to be convolved; when the resolution of the feature map to be convolved does not reach the first threshold When, perform convolution processing on the feature map to be convolved, and use the result as the feature map to be convolved again; when the resolution of the feature map to be convolved reaches the first threshold, all the features whose resolution gradually decreases are obtained The graph is the result of convolution.
  • the positioning module includes: a segmentation sub-module for performing segmentation processing according to the convolution result to obtain a segmentation result; and a positioning sub-module for performing positioning processing on the convolution result according to the segmentation result to obtain Positioning results.
  • the segmentation sub-module is used to perform segmentation processing on the feature map with the lowest resolution in the convolution result to obtain the segmentation result.
  • the positioning sub-module is used to: determine the corresponding position information of the target object in the convolution result according to the segmentation result; perform positioning processing on the convolution result according to the position information to obtain the positioning result.
  • the positioning sub-module is further used to: read the coordinate position of the segmentation result; use the coordinate position as the center of the area to determine the convolution result separately, and the feature map at each resolution can be fully covered
  • the area position of the target object is used as the corresponding position information of the target object in the convolution result.
  • the positioning sub-module is further configured to: according to the position information, perform cutting processing on the feature map at each resolution in the convolution result separately to obtain the positioning result.
  • the deconvolution module is used to: use the feature map with the lowest resolution in all feature maps included in the positioning result as the feature map to be deconvolved; When the second threshold is not reached, perform deconvolution processing on the feature map to be deconvolved to obtain the deconvolution processing result; determine the next feature map of the feature map to be deconvolved in the positioning result in the order of increasing resolution; Fuse the deconvolution processing result with the next feature map, and use the fusion result as the feature map to be deconvolved again; when the resolution of the feature map to be deconvolved reaches the second threshold, the feature map to be deconvolved is taken as Deconvolution result.
  • the segmentation processing includes: returning the object to be segmented through softmax to obtain a regression result; and completing the segmentation processing of the object to be segmented by comparing the maximum value of the regression results.
  • the device is implemented by a neural network.
  • the neural network includes a first segmentation sub-network and a second segmentation sub-network, wherein the first segmentation sub-network is used to perform stepwise convolution processing and segmentation of the image to be processed Processing, the second segmentation sub-network is used to perform stepwise deconvolution processing and segmentation processing on the positioning result.
  • the device further includes a training module for training the first segmentation sub-network according to a preset training set; training the first segmentation sub-network according to the preset training set and the trained first segmentation sub-network Two-division sub-network.
  • a resolution adjustment module is further included, configured to adjust the image to be processed to a preset resolution.
  • the embodiments of the present disclosure also provide a computer-readable storage medium on which computer program instructions are stored, and the computer program instructions implement the above-mentioned method when executed by a processor.
  • the computer-readable storage medium may be a non-volatile computer-readable storage medium.
  • An embodiment of the present disclosure also provides an electronic device, including: a processor; a memory for storing executable instructions of the processor; wherein the processor is configured as the above method.
  • the electronic device can be provided as a terminal, server or other form of device.
  • FIG. 10 is a block diagram of an electronic device 800 according to an embodiment of the present disclosure.
  • the electronic device 800 may be a mobile phone, a computer, a digital broadcasting terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, and other terminals.
  • the electronic device 800 may include one or more of the following components: a processing component 802, a memory 804, a power component 806, a multimedia component 808, an audio component 810, an input/output (I/O) interface 812, and a sensor component 814 , And communication component 816.
  • the processing component 802 generally controls the overall operations of the electronic device 800, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations.
  • the processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the foregoing method.
  • the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components.
  • the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.
  • the memory 804 is configured to store various types of data to support operations in the electronic device 800. Examples of these data include instructions for any application or method operating on the electronic device 800, contact data, phone book data, messages, pictures, videos, etc.
  • the memory 804 can be implemented by any type of volatile or nonvolatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), Magnetic Memory, Flash Memory, Magnetic Disk or Optical Disk.
  • SRAM static random access memory
  • EEPROM electrically erasable programmable read-only memory
  • EPROM erasable Programmable Read Only Memory
  • PROM Programmable Read Only Memory
  • ROM Read Only Memory
  • Magnetic Memory Flash Memory
  • Magnetic Disk Magnetic Disk or Optical Disk.
  • the power supply component 806 provides power for various components of the electronic device 800.
  • the power supply component 806 may include a power management system, one or more power supplies, and other components associated with the generation, management, and distribution of power for the electronic device 800.
  • the multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user.
  • the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user.
  • the touch panel includes one or more touch sensors to sense touch, sliding, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure related to the touch or slide operation.
  • the multimedia component 808 includes a front camera and/or a rear camera. When the electronic device 800 is in an operation mode, such as a shooting mode or a video mode, the front camera and/or the rear camera can receive external multimedia data. Each front camera and rear camera can be a fixed optical lens system or have focal length and optical zoom capabilities.
  • the audio component 810 is configured to output and/or input audio signals.
  • the audio component 810 includes a microphone (MIC).
  • the microphone is configured to receive external audio signals.
  • the received audio signal may be further stored in the memory 804 or transmitted via the communication component 816.
  • the audio component 810 further includes a speaker for outputting audio signals.
  • the I/O interface 812 provides an interface between the processing component 802 and a peripheral interface module.
  • the peripheral interface module may be a keyboard, a click wheel, a button, and the like. These buttons may include but are not limited to: home button, volume button, start button, and lock button.
  • the sensor component 814 includes one or more sensors for providing the electronic device 800 with various aspects of state evaluation.
  • the sensor component 814 can detect the on/off status of the electronic device 800 and the relative positioning of the components.
  • the component is the display and the keypad of the electronic device 800.
  • the sensor component 814 can also detect the electronic device 800 or the electronic device 800.
  • the position of the component changes, the presence or absence of contact between the user and the electronic device 800, the orientation or acceleration/deceleration of the electronic device 800, and the temperature change of the electronic device 800.
  • the sensor component 814 may include a proximity sensor configured to detect the presence of nearby objects when there is no physical contact.
  • the sensor component 814 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications.
  • the sensor component 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor or a temperature sensor.
  • the communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices.
  • the electronic device 800 can access a wireless network based on a communication standard, such as WiFi, 2G, or 3G, or a combination thereof.
  • the communication component 816 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel.
  • the communication component 816 further includes a near field communication (NFC) module to facilitate short-range communication.
  • the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.
  • RFID radio frequency identification
  • IrDA infrared data association
  • UWB ultra-wideband
  • Bluetooth Bluetooth
  • the electronic device 800 can be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field A programmable gate array (FPGA), controller, microcontroller, microprocessor, or other electronic components are implemented to implement the above methods.
  • ASIC application specific integrated circuits
  • DSP digital signal processors
  • DSPD digital signal processing devices
  • PLD programmable logic devices
  • FPGA field A programmable gate array
  • controller microcontroller, microprocessor, or other electronic components are implemented to implement the above methods.
  • a non-volatile computer-readable storage medium such as a memory 804 including computer program instructions, which can be executed by the processor 820 of the electronic device 800 to complete the foregoing method.
  • FIG. 11 is a block diagram of an electronic device 1900 according to an embodiment of the present disclosure.
  • the electronic device 1900 may be provided as a server. 11
  • the electronic device 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932, for storing instructions that can be executed by the processing component 1922, such as application programs.
  • the application program stored in the memory 1932 may include one or more modules each corresponding to a set of instructions.
  • the processing component 1922 is configured to execute instructions to perform the above-described methods.
  • the electronic device 1900 may also include a power supply component 1926 configured to perform power management of the electronic device 1900, a wired or wireless network interface 1950 configured to connect the electronic device 1900 to a network, and an input output (I/O) interface 1958 .
  • the electronic device 1900 can operate based on an operating system stored in the memory 1932, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or the like.
  • a non-volatile computer-readable storage medium such as the memory 1932 including computer program instructions, which can be executed by the processing component 1922 of the electronic device 1900 to complete the foregoing method.
  • the present disclosure may be a system, method, and/or computer program product.
  • the computer program product may include a computer-readable storage medium loaded with computer-readable program instructions for enabling a processor to implement various aspects of the present disclosure.
  • the computer-readable storage medium may be a tangible device that can hold and store instructions used by the instruction execution device.
  • the computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing.
  • Computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) Or flash memory), static random access memory (SRAM), portable compact disk read-only memory (CD-ROM), digital versatile disk (DVD), memory stick, floppy disk, mechanical encoding device, such as a printer with instructions stored thereon
  • RAM random access memory
  • ROM read-only memory
  • EPROM erasable programmable read-only memory
  • flash memory flash memory
  • SRAM static random access memory
  • CD-ROM compact disk read-only memory
  • DVD digital versatile disk
  • memory stick floppy disk
  • mechanical encoding device such as a printer with instructions stored thereon
  • the computer-readable storage medium used here is not interpreted as a transient signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (for example, light pulses through fiber optic cables), or through wires Transmission of electrical signals.
  • the computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to various computing/processing devices, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and/or a wireless network.
  • the network may include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers, and/or edge servers.
  • the network adapter card or network interface in each computing/processing device receives computer-readable program instructions from the network, and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing/processing device .
  • the computer program instructions used to perform the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, status setting data, or in one or more programming languages.
  • Source code or object code written in any combination, the programming language includes object-oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as "C" language or similar programming languages.
  • Computer-readable program instructions can be executed entirely on the user's computer, partly on the user's computer, executed as a stand-alone software package, partly on the user's computer and partly executed on a remote computer, or entirely on the remote computer or server carried out.
  • the remote computer can be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to access the Internet connection).
  • LAN local area network
  • WAN wide area network
  • an electronic circuit such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), can be customized by using the status information of the computer-readable program instructions.
  • the computer-readable program instructions are executed to realize various aspects of the present disclosure.
  • These computer-readable program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine such that when these instructions are executed by the processor of the computer or other programmable data processing device , A device that implements the functions/actions specified in one or more blocks in the flowchart and/or block diagram is produced. It is also possible to store these computer-readable program instructions in a computer-readable storage medium. These instructions make computers, programmable data processing apparatuses, and/or other devices work in a specific manner, so that the computer-readable medium storing instructions includes An article of manufacture, which includes instructions for implementing various aspects of the functions/actions specified in one or more blocks in the flowchart and/or block diagram.
  • each block in the flowchart or block diagram may represent a module, program segment, or part of an instruction, and the module, program segment, or part of an instruction contains one or more functions for implementing the specified logical function.
  • Executable instructions may also occur in a different order from the order marked in the drawings. For example, two consecutive blocks can actually be executed in parallel, or they can sometimes be executed in the reverse order, depending on the functions involved.
  • each block in the block diagram and/or flowchart, and the combination of the blocks in the block diagram and/or flowchart can be implemented by a dedicated hardware-based system that performs the specified functions or actions Or it can be realized by a combination of dedicated hardware and computer instructions.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Mathematical Physics (AREA)
  • Quality & Reliability (AREA)
  • Radiology & Medical Imaging (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Physiology (AREA)
  • Fuzzy Systems (AREA)
  • Psychiatry (AREA)
  • Signal Processing (AREA)
  • Pathology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Public Health (AREA)
  • Veterinary Medicine (AREA)
  • Probability & Statistics with Applications (AREA)
  • Image Analysis (AREA)

Abstract

一种图像处理方法及装置、电子设备和存储介质，所述方法包括：对待处理图像进行逐级卷积处理，得到卷积结果（S11）；根据所述卷积结果，通过定位处理得到定位结果（S12）；对所述定位结果进行逐级反卷积处理，得到反卷积结果（S13）；对所述反卷积结果进行分割处理，从所述待处理图像中分割出目标对象（S14）。该方法可在一次图像处理的过程中同时实现目标对象的定位和分割，提高图像处理精度的同时保障图像处理的速度。

Description

图像处理方法及装置、电子设备和存储介质
本申请要求在2019年4月1日提交中国专利局、申请号为201910258038.1、发明名称为“图像处理方法及装置、电子设备和存储介质”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
本公开涉及图像处理技术领域，尤其涉及一种图像处理方法及装置、电子设备和存储介质。
背景技术
在图像技术领域，对感兴趣区域或目标区域进行分割，是进行图像分析和目标识别的基础。例如，在医学图像中通过分割，清晰地识别一个或多个器官或病灶之间的边界。准确地分割三维医学图像对于许多临床应用而言是至关重要的。
发明内容
本公开提出了一种图像处理技术方案。
根据本公开的一方面，提供了一种图像处理方法，包括：对待处理图像进行逐级卷积处理，得到卷积结果；根据所述卷积结果，通过定位处理得到定位结果；对所述定位结果进行逐级反卷积处理，得到反卷积结果；对所述反卷积结果进行分割处理，从所述待处理图像中分割出目标对象。
在一种可能的实现方式中，所述对待处理图像进行逐级卷积处理，得到卷积结果，包括：对待处理图像进行逐级卷积处理，得到至少一个分辨率逐渐递减的特征图，作为所述卷积结果。
在一种可能的实现方式中，所述对待处理图像进行逐级卷积处理，得到至少一个分辨率逐渐递减的特征图，作为所述卷积结果，包括：对待处理图像进行卷积处理，所得到的特征图作为待卷积特征图；在所述待卷积特征图的分辨率未达到第一阈值时，对所述待卷积特征图进行卷积处理，并将得到的结果再次作为待卷积特征图；在所述待卷积特征图的分辨率达到第一阈值时，将得到的分辨率逐渐递减的所有特征图作为所述卷积结果。
在一种可能的实现方式中，所述根据所述卷积结果，通过定位处理得到定位结果，包括：根据所述卷积结果进行分割处理，得到分割结果；根据所述分割结果，对所述卷积结果进行定位处理，得到定位结果。
在一种可能的实现方式中，所述根据所述卷积结果进行分割处理，得到分割结果，包括：对所述卷积结果中分辨率最低的特征图进行分割处理，得到分割结果。
在一种可能的实现方式中，所述根据所述分割结果，对所述卷积结果进行定位处理，得到定位结果，包括：根据所述分割结果，确定所述目标对象在所述卷积结果中对应的位置信息；根据所述位置信息，对所述卷积结果进行定位处理，得到定位结果。
在一种可能的实现方式中，所述根据所述分割结果，确定所述目标对象在卷积结果中对应的位置信息，包括：读取所述分割结果的坐标位置；将所述坐标位置作为区域中心，分别确定所述卷积结果内，每个分辨率下的特征图中可全部覆盖所述目标对象的区域位置，作为所述目标对象在卷积结果中对应的位置信息。
在一种可能的实现方式中，所述根据所述位置信息，对所述卷积结果进行定位处理，得到定位结果，包括：根据所述位置信息，对所述卷积结果中每个分辨率下的特征图分别进行裁切处理，得到定位结果。
在一种可能的实现方式中，所述对所述定位结果进行逐级反卷积处理，得到反卷积结果，包括：将所述定位结果 中包含的所有特征图中分辨率最低的特征图作为待反卷积特征图；在所述待反卷积特征图的分辨率未达到第二阈值时，对所述待反卷积特征图进行反卷积处理，得到反卷积处理结果；按照分辨率逐渐递增的顺序，确定所述定位结果中所述待反卷积特征图的下一特征图；将所述反卷积处理结果与所述下一特征图进行融合，将所述融合结果再次作为待反卷积特征图；在所述待反卷积特征图的分辨率达到第二阈值时，将所述待反卷积特征图作为反卷积结果。
在一种可能的实现方式中，所述分割处理包括：将待分割对象通过softmax回归，得到回归结果；通过对所述回归结果进行最大值比较，完成对所述待分割对象的分割处理。
在一种可能的实现方式中，所述方法通过神经网络实现，所述神经网络包括第一分割子网络及第二分割子网络，其中，所述第一分割子网络用于对所述待处理图像进行逐级卷积处理及分割处理，所述第二分割子网络用于对所述定位结果进行逐级反卷积处理及分割处理。
在一种可能的实现方式中，所述神经网络的训练过程，包括：根据预设的训练集，训练所述第一分割子网络；根据所述预设的训练集以及已训练的第一分割子网络，训练所述第二分割子网络。
在一种可能的实现方式中，所述对所述待处理图像进行逐级卷积处理，得到卷积结果之前，还包括：将所述待处理图像调整至预设分辨率。
在一种可能的实现方式中，所述待处理图像为三维医学图像。
根据本公开的一方面，提供了一种图像处理装置，包括：卷积模块，用于对待处理图像进行逐级卷积处理，得到卷积结果；定位模块，用于根据所述卷积结果，通过定位处理得到定位结果；反卷积模块，用于对所述定位结果进行逐级反卷积处理，得到反卷积结果；目标对象获取模块，用于对所述反卷积结果进行分割处理，从所述待处理图像中分割出目标对象。
在一种可能的实现方式中，所述卷积模块用于：对待处理图像进行逐级卷积处理，得到至少一个分辨率逐渐递减的特征图，作为所述卷积结果。
在一种可能的实现方式中，所述卷积模块进一步用于：对待处理图像进行卷积处理，所得到的特征图作为待卷积特征图；在所述待卷积特征图的分辨率未达到第一阈值时，对所述待卷积特征图进行卷积处理，并将得到的结果再次作为待卷积特征图；在所述待卷积特征图的分辨率达到第一阈值时，将得到的分辨率逐渐递减的所有特征图作为所述卷积结果。
在一种可能的实现方式中，所述定位模块包括：分割子模块，用于根据所述卷积结果进行分割处理，得到分割结果；定位子模块，用于根据所述分割结果，对所述卷积结果进行定位处理，得到定位结果。
在一种可能的实现方式中，所述分割子模块用于：对所述卷积结果中分辨率最低的特征图进行分割处理，得到分割结果。
在一种可能的实现方式中，所述定位子模块用于：根据所述分割结果，确定所述目标对象在所述卷积结果中对应的位置信息；根据所述位置信息，对所述卷积结果进行定位处理，得到定位结果。
在一种可能的实现方式中，所述定位子模块进一步用于：读取所述分割结果的坐标位置；将所述坐标位置作为区域中心，分别确定所述卷积结果内，每个分辨率下的特征图中可全部覆盖所述目标对象的区域位置，作为所述目标对象在卷积结果中对应的位置信息。
在一种可能的实现方式中，所述定位子模块进一步用于：根据所述位置信息，对所述卷积结果中每个分辨率下的特征图分别进行裁切处理，得到定位结果。
在一种可能的实现方式中，所述反卷积模块用于：将所述定位结果中包含的所有特征图中分辨率最低的特征图作 为待反卷积特征图；在所述待反卷积特征图的分辨率未达到第二阈值时，对所述待反卷积特征图进行反卷积处理，得到反卷积处理结果；按照分辨率逐渐递增的顺序，确定所述定位结果中所述待反卷积特征图的下一特征图；将所述反卷积处理结果与所述下一特征图进行融合，将所述融合结果再次作为待反卷积特征图；在所述待反卷积特征图的分辨率达到第二阈值时，将所述待反卷积特征图作为反卷积结果。
在一种可能的实现方式中，所述分割处理包括：将待分割对象通过softmax回归，得到回归结果；通过对所述回归结果进行最大值比较，完成对所述待分割对象的分割处理。
在一种可能的实现方式中，所述装置通过神经网络实现，所述神经网络包括第一分割子网络及第二分割子网络，其中，所述第一分割子网络用于对所述待处理图像进行逐级卷积处理及分割处理，所述第二分割子网络用于对所述定位结果进行逐级反卷积处理及分割处理。
在一种可能的实现方式中，所述装置还包括训练模块，用于：根据预设的训练集，训练所述第一分割子网络；根据所述预设的训练集以及已训练的第一分割子网络，训练所述第二分割子网络。
在一种可能的实现方式中，所述卷积模块之前还包括分辨率调整模块，用于：将所述待处理图像调整至预设分辨率。
在一种可能的实现方式中，所述待处理图像为三维医学图像。
根据本公开的一方面，提供了一种电子设备，包括：处理器；用于存储处理器可执行指令的存储器；其中，所述处理器被配置为：执行上述图像处理方法。
根据本公开的一方面，提供了一种计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令被处理器执行时实现上述图像处理方法。
在本公开实施例中，通过对待处理图像进行逐级卷积处理和分割处理得到分割结果，并基于分割结果得到定位结果，再通过对定位结果进行逐级反卷积处理后再进行分割处理，可以从待处理图像中分割出目标对象。通过上述过程可以在一次图像处理的过程中同时实现目标对象的定位和分割，提高图像处理精度的同时保障图像处理的速度。
应当理解的是，以上的一般描述和后文的细节描述仅是示例性和解释性的，而非限制本公开。根据下面参考附图对示例性实施例的详细说明，本公开的其它特征及方面将变得清楚。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分，这些附图示出了符合本公开的实施例，并与说明书一起用于说明本公开的技术方案。
图1示出根据本公开一实施例的图像处理方法的流程图。
图2示出根据本公开一实施例的图像处理方法的流程图。
图3示出根据本公开一实施例的图像处理方法的流程图。
图4示出根据本公开一实施例的图像处理方法的流程图。
图5示出根据本公开一实施例的图像处理方法的流程图。
图6示出根据本公开一实施例的图像处理方法的流程图。
图7示出根据本公开一实施例的图像处理方法的流程图。
图8示出根据本公开一应用示例的示意图。
图9示出根据本公开一实施例的图像处理装置的框图。
图10示出根据本公开实施例的一种电子设备的框图。
图11示出根据本公开实施例的一种电子设备的框图。
具体实施方式
以下将参考附图详细说明本公开的各种示例性实施例、特征和方面。附图中相同的附图标记表示功能相同或相似的元件。尽管在附图中示出了实施例的各种方面，但是除非特别指出，不必按比例绘制附图。
在这里专用的词“示例性”意为“用作例子、实施例或说明性”。这里作为“示例性”所说明的任何实施例不必解释为优于或好于其它实施例。
本文中术语“和/或”，仅仅是一种描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。另外，本文中术语“至少一种”表示多种中的任意一种或多种中的至少两种的任意组合，例如，包括A、B、C中的至少一种，可以表示包括从A、B和C构成的集合中选择的任意一个或多个元素。
另外，为了更好地说明本公开，在下文的具体实施方式中给出了众多的具体细节。本领域技术人员应当理解，没有某些具体细节，本公开同样可以实施。在一些实例中，对于本领域技术人员熟知的方法、手段、元件和电路未作详细描述，以便于凸显本公开的主旨。
图1示出根据本公开一实施例的图像处理方法的流程图，该方法可以应用于图像处理装置，图像处理装置可以为终端设备、服务器或者其他处理设备等。其中，终端设备可以为用户设备(User Equipment，UE)、移动设备、用户终端、终端、蜂窝电话、无绳电话、个人数字处理(Personal Digital Assistant，PDA)、手持设备、计算设备、车载设备、可穿戴设备等。
在一些可能的实现方式中，该图像处理方法可以通过处理器调用存储器中存储的计算机可读指令的方式来实现。
如图1所示，所述图像处理方法可以包括：
步骤S11，对待处理图像进行逐级卷积处理，得到卷积结果。
步骤S12，根据卷积结果，通过定位处理得到定位结果。
步骤S13，对定位结果进行逐级反卷积处理，得到反卷积结果。
步骤S14，对反卷积结果进行分割处理，从待处理图像中分割出目标对象。
本公开实施例的图像处理方法，通过逐级卷积处理和分割处理，对待处理图像中的目标对象进行初步分割，从而得到反映目标对象在待处理图像的基本分布位置的定位结果，基于这一定位结果，可以再通过逐级反卷积处理和分割处理，实现待处理图像内目标对象的高精度分割，通过这一过程，在定位结果的基础上实现对目标对象的分割，与直接对待处理图像进行目标分割相比，可以有效提升图像处理的精度；同时，上述方法可以在一次图像处理过程中，先后实现对图像的目标定位和分割，由于可以将图像的目标定位和分割过程结合分析，因此减少了图像处理的耗时，也降低了图像处理过程中可能存在的存储消耗。
其中，本公开实施例的图像处理方法可以应用于三维医学图像的处理，例如，用于识别医学图像中的目标区域，该目标区域可以是器官、病灶、组织等等。在一种可能的实现方式中，待处理图像可以是心脏器官的三维医学图像，也就是说，本公开实施例的图像处理方法可以应用于心脏病的治疗过程中，在一个示例中，该图像处理方法可以应用于心房纤维化颤动治疗过程，通过精确分割心房图像，从而理解和分析心房纤维化的病因，继而制定针对性的心房纤 维化颤动的手术消融治疗方案，提升心房纤维化颤动的治疗效果。
需要说明的是，本公开实施例的图像处理方法不限于应用在三维医学图像处理，可以应用于任意的图像处理，本公开对此不作限定。
在一种可能的实现方式中，待处理图像可以包括多张图片，根据该多张图片可以识别出一个或多个三维的器官。
步骤S11的实现方式不受限定，任何可以得到用于进行分割处理的特征图的方式都可以作为步骤S11的实现方式。在一种可能的实现方式中，步骤S11可以包括：对待处理图像进行逐级卷积处理，得到至少一个分辨率逐渐递减的特征图，作为卷积结果。
如何通过逐级卷积处理来得到至少一个分辨率逐渐递减的特征图，其具体的处理过程同样不受限定，图2示出根据本公开一实施例的图像处理方法的流程图，如图所示，在一种可能的实现方式中，对待处理图像进行逐级卷积处理，得到至少一个分辨率逐渐递减的特征图，作为卷积结果，可以包括：
步骤S111，对待处理图像进行卷积处理，所得到的特征图作为待卷积特征图。
步骤S112，在待卷积特征图的分辨率未达到第一阈值时，对待卷积特征图进行卷积处理，并将得到的结果再次作为待卷积特征图。
步骤S113，在待卷积特征图的分辨率达到第一阈值时，将得到的分辨率逐渐递减的所有特征图作为卷积结果。
通过上述步骤可以看出，在本公开实施例中，通过对待处理图像进行一次卷积处理，可以得到初始分辨率下的特征图，再对初始分辨率下的特征图再进行一次卷积处理，可以得到下一分辨率下的特征图，以此类推，通过对待处理图像进行多次卷积处理，可以得到一系列分辨率逐渐递减的特征图，这些特征图可以作为卷积结果用于后续步骤的进行。这一过程的迭代次数不受限制，可以在得到的最小分辨率的特征图达到第一阈值时停止，第一阈值可以根据需求和实际情况进行设定，在此不限定具体值。由于第一阈值不受限定，因此得到的卷积结果中包含的特征图的个数和每一张特征图的分辨率均不受限定，可以根据实际情况进行具体选择。
在一种可能的实现方式中，卷积处理的过程和实现方式不受限定，在一个示例中，卷积处理的过程可以包括将待处理对象通过卷积、池化、批量归一化(Batch Normalization)或是参数线性整流单元(PReLU，Parametric Rectified Linear Unit)中的一个或多个。在一个示例中，可以采用3D U-Net全卷积神经网络中的编码器结构来实现，在一个示例中，也可以通过V-Net全卷机神经网络中的编码器结构来实现。本公开对卷积处理的具体方式不作限制。
根据卷积结果，通过定位处理来得到定位结果的过程可以存在多种实现方式，图3示出根据本公开一实施例的图像处理方法的流程图，如图所示，在一种可能的实现方式中，步骤S12可以包括：
步骤S121，根据卷积结果进行分割处理，得到分割结果。
步骤S122，根据分割结果，对卷积结果进行定位处理，得到定位结果。
步骤S121的过程同样不受限定，通过上述公开实施例可以得知，卷积结果中可以包含多张特征图，因此分割结果是通过对卷积结果中的哪一特征图进行分割处理来得到的，可以根据实际情况进行确定。在一种可能的实现方式中，步骤S121可以包括：对卷积结果中分辨率最低的特征图进行分割处理，得到分割结果。
分割处理的处理方式不受限定，任何可以从特征图中分割出目标的方式均可以作为本公开示例中分割处理的方法。
在一种可能的实现方式中，分割处理可以为通过softmax层来实现图像分割，具体过程可以包括：将待分割对象通 过softmax回归，得到回归结果；通过对回归结果进行最大值比较，完成对待分割对象的分割处理。在一个示例中，上述通过回归结果进行最大值比较来实现对待分割对象进行分割处理的具体过程可以为：回归结果的形式可以为与待分割对象具有相同分辨率的输出数据，输出数据与待分割对象的像素位置一一对应，在每个对应的像素位置处，输出数据包含一个概率值，用以表明待分割对象在这一像素位置处作为分割目标的概率，基于输出数据中包含的概率可以进行最大值比较，从而确定每一像素位置是否为分割目标位置，继而实现从待分割对象中提取出分割目标的操作，最大值比较的具体方式不受限定，可以设定为概率较大的值所代表的像素位置处对应分割目标，也可以设定为概率较小的值所代表的的像素位置处对应分割目标，根据实际情况进行设定即可，在此不做限定。基于上述各公开实施例可知，在一个示例中，分割结果的得到过程可以为：将卷积结果中分辨率最低的特征图通过softmax层，并将得到的结果进行最大值比较，从而得到分割结果。
基于分割结果，可以通过步骤S122对卷积结果进行定位处理，来得到定位结果，步骤S122的实现过程不受限定，图4示出根据本公开一实施例的图像处理方法的流程图，如图所示，在一种可能的实现方式中，步骤S122可以包括：
步骤S1221，根据分割结果，确定目标对象在卷积结果中对应的位置信息。
步骤S1222，根据位置信息，对卷积结果进行定位处理，得到定位结果。
其中，位置信息为可以表明在卷积结果的各特征图内，目标对象所处位置的信息，其具体表现形式不受限定，在一个示例中，位置信息可以通过位置坐标集合的形式存在，在一个示例中，位置信息可以通过坐标+面积的形式存在，可以根据实际情况灵活选择位置信息的表现形式。由于位置信息的表现形式不受限定，因此步骤S1221的具体过程也可以随着位置信息的表现形式灵活确定。图5示出根据本公开一实施例的图像处理方法的流程图，如图所示，在一种可能的实现方式中，步骤S1221可以包括：
步骤S12211，读取分割结果的坐标位置。
步骤S12212，将坐标位置作为区域中心，分别确定卷积结果内，每个分辨率下的特征图中可全部覆盖目标对象的区域位置，作为目标对象在卷积结果中对应的位置信息。
其中，步骤S12211读取的分割结果的坐标位置，可以是表明分割结果位置的任意坐标，在一个示例中，这一坐标可以是分割结果上某固定位置的坐标值；在一个示例中，这一坐标可以是分割结果上某几个固定位置的坐标值；在一个示例中，这一坐标可以是分割结果重心位置的坐标值。基于读取的坐标位置，可以通过步骤S12212，在卷积结果中每张特征图下对应的位置处，定位到目标对象，继而得到完全覆盖目标对象的区域位置，这一区域位置的表现形式同样不受限定，在一个示例中，这一区域位置的表现形式可以是区域所有顶点的坐标集合，在一个示例中，这一区域位置的表现形式可以是区域位置的中心坐标与区域位置的覆盖面积集合。步骤S12212的具体过程可以根据区域位置的表现形式不同而随之灵活改变，在一个示例中，步骤S12212的过程可以为：基于分割结果在所在特征图内的重心坐标，依据分割结果所在特征图与卷积结果中其余特征图的分辨率比例关系，可以分别确定卷积结果中每张特征图内目标对象的重心坐标；以此重心坐标为中心，在每张特征图中，确定可以完全覆盖目标对象的区域，将此区域的顶点坐标作为目标对象在卷积结果中对应的位置信息。由于卷积结果中各特征图之间存在分辨率的差异，因此卷积结果中各特征图内覆盖目标对象的区域之间也可能存在分辨率的差异。在一个示例中，不同特征图确定的覆盖目标对象的区域之间可以存在比例关系，这一比例关系可以与特征图之间的分辨率比例关系一致，举例说明，在一个示例中，卷积结果中可能存在两个特征图A和B，特征图A中覆盖目标对象的区域被记为区域A，特征图B中覆盖目标对象的区域被记为区域B，其中特征图A的分辨率为特征图分辨率B的2倍，则区域A的面积为区域B的2倍。
基于步骤S1221得到的位置信息，可以通过步骤S1222来得到定位结果，上述公开实施例已表明，位置信息可以存在多种不同的表现形式，随着位置信息表现形式的不同，步骤S1222的具体实施过程也可能存在不同。在一种可能的实现方式中，步骤S1222可以包括：根据位置信息，对卷积结果中每个分辨率下的特征图分别进行裁切处理，得到定位结果。在一个示例中，位置信息可以为卷积结果内各特征图可以覆盖目标对象的区域顶点的坐标集合，基于此坐标集合，可以对卷积结果中的各特征图进行裁切，保留每个特征图中覆盖目标对象的区域作为新的特征图，则这些新的特征图的集合即为定位结果。
通过上述各公开实施例的任意形式组合，可以得到定位结果，这一过程可以有效的对卷积结果中各分辨率下的特征图内的目标对象进行粗略定位，基于此粗略定位可以将原有的卷积结果处理为定位结果，由于定位结果中各分辨率下的特征图内去掉了大部分不包含目标对象的图片信息，因此可以大大减小图像处理过程中的存储消耗，加快计算速度，提升图像处理的效率和速度，同时，由于目标对象在定位结果中所占的信息比例更大，因此基于定位结果进行目标对象分割的效果，相比于直接利用待处理图像进行目标对象分割的效果更好，从而可以提高图像处理的精度。
在得到了定位结果后，可以基于此定位结果来实现目标对象的分割，分割的具体实现形式不受限定，可以根据实际情况灵活选择。在一种可能的实现方式中，可以从定位结果中选择某一特征图，再进行进一步的分割处理，来得到目标对象。在另一种可能的实现方式中，可以利用定位结果还原出包含更多目标对象信息的特征图，再利用此特征图进行进一步的分割处理，来得到目标对象。
通过上述步骤中可以看出，在一种可能的实现方式中，利用定位结果实现目标对象分割的过程可以通过步骤S13和S14实现，即先对定位结果进行逐级反卷积处理，来得到包含更多目标对象信息的反卷积结果，再基于此反卷积结果进行分割处理，来得到目标对象。逐级反卷积的过程可以被看作是逐级卷积过程的逆向操作过程，因此其实现过程也如步骤S11一样具有多种可能的实现形式。图6示出根据本公开一实施例的图像处理方法的流程图，如图所示，在一种可能的实现方式中，步骤S13可以包括：
步骤S131，将定位结果中包含的所有特征图中分辨率最低的特征图作为待反卷积特征图。
步骤S132，在待反卷积特征图的分辨率未达到第二阈值时，对待反卷积特征图进行反卷积处理，得到反卷积处理结果。
步骤S133，按照分辨率逐渐递增的顺序，确定定位结果中待反卷积特征图的下一特征图。
步骤S134，将反卷积处理结果与下一特征图进行融合，将融合结果再次作为待反卷积特征图。
步骤S135，在待反卷积特征图的分辨率达到第二阈值时，将待反卷积特征图作为反卷积结果。
上述步骤中，反卷积处理结果是对待反卷积特征图进行反卷积处理得到的处理结果，而下一特征图，则是从定位结果中得到的特征图，即在定位结果中，满足分辨率大于当前反卷积特征图一级这一条件的特征图，可以作为下一特征图，与反卷积处理结果进行融合。因此逐级反卷积处理的过程，可以是从定位结果中分辨率最低的特征图开始，通过反卷积处理，得到分辨率提升一级后的特征图，此时可以将这一分辨率提升一级后得到的特征图作为反卷积处理结果，由于在定位结果中，本身也存在与反卷积处理结果的分辨率相同的特征图，这两张特征图之间均包含目标对象的有效信息，因此可以将这两张特征图进行融合，融合后的特征图包含了这两张特征图内所包含的所有目标对象的有效信息，因此可以将融合后的特征图再次作为新的待反卷积特征图，对这一新的待反卷积特征图进行反卷积处理，并将处理结果再次与定位结果内对应分辨率的特征图进行融合，直至融合后的特征图分辨率达到第二阈值时，停止反卷积处理，此时得到的最终的融合结果，包含了定位结果内每一张特征图中所含有的目标对象的有效信息，因此可以将其 作为反卷积结果，用于后续的目标对象分割。在本公开实施例中，第二阈值根据待处理图像原有的分辨率灵活决定，在此并不限定具体值。
通过上述过程中，反卷积结果是通过对定位结果进行逐级反卷积处理来得到的，且反卷积结果用于最终的目标对象分割，因此得到的最终结果，由于存在了目标对象的定位基础，因此可以有效包含目标对象的全局信息，具有更高的准确率；而且也无需将待处理图像进行分割，而是作为整体进行图像处理，因此处理过程也具有更高的效率；同时通过上述过程可以看出，在一次的图像处理过程中，对于目标对象的分割是基于目标对象的定位结果来实现的，无需通过两个独立的过程分别实现目标对象定位和目标对象分割，因此可以大大减小数据的存储、消耗和计算量，继而提升图像处理的速度和效率，减小时间和空间上的消耗。而且逐级反卷积过程可以有利于每一个分辨率下特征图所包含的有效信息均保留在了最终得到的反卷积结果内，由于反卷积结果用于最终的图像分割，因此可以大大提升最终得到结果的精度。
在得到了反卷积结果后，可以对反卷积结果进行分割处理，得到的结果可以作为从待处理图像中分割出的目标对象，对反卷积结果进行分割处理的过程，与上述对卷积结果进行分割处理的过程一致，只是被分割处理的对象存在差异，因此可以参考上述公开实施例的过程，在此不再赘述。
在一种可能的实现方式中，可以通过神经网络实现本公开实施例的图像处理方法。通过上述过程中可以看出，本公开实施例的图像处理方法主要包含了两次分割过程，第一次是对待处理图像的粗略分割，第二次是基于粗略分割得到的定位结果来进行的更高精度的分割，因此第二次分割与第一次分割可以通过一个神经网络实现，二者共享一套参数，因此，可以将两次分割看作为一个神经网络下的两个子神经网络，因此，在一种可能的实现方式中，神经网络可以包括第一分割子网络及第二分割子网络，其中，第一分割子网络用于对待处理图像进行逐级卷积处理及分割处理，第二分割子网络用于对定位结果进行逐级反卷积处理及分割处理。神经网络所采用的具体网络结构不受限定，在一个示例中，上述公开实施例中提到的V-Net和3D-U-Net均可以作为神经网络的具体实现方式。任何可以实现第一分割子网络和第二分割子网络功能的神经网络，均可以作为神经网络的实现形式。
图7示出根据本公开实施例的图像处理方法的流程图。在一种可能的实现方式中，如图所示，本公开实施例的方法还可以包括神经网络的训练过程，记为步骤S15，其中，步骤S15可以包括：
步骤S151，根据预设的训练集，训练第一分割子网络。
步骤S152，根据预设的训练集以及已训练的第一分割子网络，训练第二分割子网络。
其中，预设的训练集可以是对样本图片进行手动剪裁等预处理后，并拆分成的多个图片集。拆分成的多个图片集中，相邻的两个图片集可以包括一部分相同的图片，例如，以医学图像为例，可以从医院采集多个样本，一个样本中包括的多个样本图片可以是连续采集的人体某一器官的图片，通过该多个样本图片可以得到器官的三维立体结构，可以沿一个方向进行拆分，第一个图片集可以包括第1-30帧图片，第二个图片集可以包括第16-45帧图片……，这样相邻的两个图片集中有15帧图片是相同的。通过这种重叠拆分的方式，可以提高分割的精确度。
如图7所示，在神经网络的训练过程中，首先可以将预设的训练集作为输入，训练第一分割子网络，根据第一分割子网络的输出结果，可以对训练集中的图片进行定位处理，经由定位处理后的训练集，可以作为第二分割子网络的训练数据，输入到第二分割子网络中进行训练，通过上述过程，最终可以得到训练完成的第一分割子网络和第二分割子网络。
在训练的过程中，确定神经网络的网络损失所使用的函数不受具体限定，在一个示例中，可以通过dice loss函数确定神经网络的网络损失，在一个示例中，可以通过交叉熵函数确定神经网络的网络损失，在一个示例中，也可以通过其他可用的损失函数确定神经网络的网络损失。第一分割子网络和第二分割子网络使用的损失函数可以相同，也可以不同，在此不受限定。
基于上述公开的实施例，在一个示例中，神经网络的完整训练过程可以为：将预设的训练集输入到第一分割子网络的网络模型中，预设的训练集中包含多张待分割图像和与待分割图像对应的掩模Mask，通过任一损失函数计算出图像经过第一分割子网络的网络模型输出的数据与对应的Mask之间的损失，然后通过反向传播算法更新第一分割子网络的网络模型参数，直至第一分割子网络模型收敛，表明第一分割子网络模型完成训练。在第一分割子网络模型完成训练后，将预设的训练集再次通过训练好的第一分割子网络模型，得到多张分割结果，基于这多张分割结果，对第一分割子网络中各分辨率的特征图进行定位处理，将这些定位并裁剪后的特征图与对应的位置的Mask输入到第二分割子网络的网络模型中进行训练，通过任一损失函数计算出定位处理后的图像经过第二分割子网络的网络模型输出的数据与对应的Mask之间的损失，然后通过反向传播算法更新第二分割子网络的网络模型参数，同时交替地更新第一分割子网络和第二分割子网络模型参数，直至整个网络模型收敛，神经网络完成训练。
通过上述各公开实施例可以看出，本公开中的神经网络虽然包含两个子神经网络，但是在训练过程中，只需要通过一套训练集数据即可完成训练，两个子神经网络共享同一套参数，可以节省更多的存储空间。由于训练的两个子神经网络共享同一套参数，因此在该神经网络应用于图像处理方法时，输入的待处理图像直接依次通过两个子神经网络即可得到输出结果，而不是分别输入到两个子神经网络分别得到输出结果后再进行计算，因此本公开中提出的图像处理方法具有更快的处理速度，同时也具有更低的空间消耗和时间消耗。
在一种可能的实现方式中，本公开实施例的方法，在步骤S11之前还可以包括：将待处理图像调整至预设分辨率。将待处理图像调整至预设分辨率的实现方法不受具体限定，在一个示例中，可以通过中心裁剪和扩充的方法，将待处理图像调整到预设分辨率。预设分辨率的具体分辨率数值也不受限定，可以根据实际情况进行灵活设定。
基于这一步骤，在本公开实施例的图像处理方法通过神经网络实现时，也可以将预设的训练集中包含的各训练图片均统一至预设分辨率后，再用于神经网络的训练。
与之相应的，在一种可能的实现方式中，本公开实施例的方法还可以包括：将分割出的目标对象还原至与待处理图像同样大小的空间中，得到最终的分割结果。由于在步骤S11之前可能对待处理图像进行了分辨率的调整，得到的分割结果实际上可能是基于分辨率调整后的图像的分割内容，因此可以将分割的结果还原至与待处理图像同样大小的空间中，得到基于最原始待处理图像的分割结果。与待处理图像同样大小的空间不受限定，根据待处理图像本身的图像性质所决定，在此不受限定，在一个示例中，待处理图像可能是三维图像，因此与待处理图像同样大小的空间可以是三维空间。
在一种可能的实现方式中，在步骤S11之前还可以包括：对待处理图像进行预处理。这一预处理过程不受限定，任何可以提高分割精度的处理方式均可以作为预处理包含的过程，在一个示例中，对待处理图像进行预处理可以包括将待处理图像进行亮度值均衡化。
通过采取同一分辨率的待处理图像作为输入来进行图像处理，可以提高后续对待处理图像依次执行卷积处理、分割处理和逐级反卷积处理的处理效率，缩短整个图像处理过程的时间。通过对待处理图像进行预处理，可以提升图像分割的准确程度，从而提高图像处理结果的精度。
应用场景示例
心脏类疾病是当前致死率最高的疾病之一，比如心房纤维化颤动是当前最为常见的心率紊乱病症之一，在一般人群中出现的概率达到了2％，而在老年人群中的发病率更高并且具有一定的致死率，严重威胁到了人类的健康。而对心房的精确分割是理解和分析心房纤维化的关键，常常被用来辅助制定针对性的心房纤维化颤动的手术消融治疗方案。而心脏的其他腔体的分割对于其他类型的心脏病的治疗和手术规划也具有同样重要的意义。然而针对医学图像中心脏腔体分割的方法仍然面临着准确率不高、计算效率低下等缺点，虽然已经有部分方法实现了较高的准确率，但是仍然存在着一些实际问题，比如缺乏三维信息，分割结果不够平滑；缺乏全局信息，计算效率低下；或是需要分成两个网络进行分割训练，时间空间上都有一定程度的冗余等。
因此，一个精度高、效率高且时空消耗低的分割方法能够极大减少医生的工作量，提高心脏分割的质量，从而提高心脏相关疾病的治疗效果。
图8示出根据本公开一应用示例的示意图，如图所示，本公开实施例提出了一种图像处理方法，这一处理方法是基于训练好的一套神经网络来实现的。从图中可以看出，这套神经网络的具体训练过程可以为：
首先对预设的训练数据进行处理，预设的训练数据包含多张输入图像和对应的Mask，通过中心裁剪和扩充的方法将多张输入图像的分辨率统一为同样大小，在本示例中统一后的分辨率为576×576×96。
在将多张输入图像统一分辨率后，可以利用这些输入图像对第一分割子网络进行训练，具体的训练过程可以为：
采用类似基于V-Net或者3D-U-Net的三维全卷积神经网络中的编码器结构，对输入图像进行多次卷积处理，本示例中卷积处理的过程可以包括卷积，池化，batch norm以及PRelu，通过多次卷积处理，每次卷积处理的输入均采用上次卷积处理得到的结果，本示例中共执行了4次卷积处理，因此可以分别生成分辨率大小为576×576×96，288×288×48，144×144×24，以及72×72×12的特征图，并且输入图像的特征通道从8个提升到128个；
在得到了上述4个特征图后，针对其中分辨率最小的特征图，本示例中为72×72×12的特征图，将其通过一个softmax层，可以得到两个分辨率为72×72×12的概率输出，这两个概率输出分别代表像素相关位置是否为目标腔体的概率，这两个概率输出可以作为第一分割子网络的输出结果，利用dice loss、交叉熵或者其他损失函数，可以计算该输出结果与直接降采样为72×72×12的mask之间的损失，基于计算出的损失，可以利用反向传播算法更新第一分割子网络的网络参数，直到第一分割子网络的网络模型收敛，此时可以代表第一分割子网络训练完成。
在第一分割子网络训练完成后，可以将统一分辨率后的多张输入图像通过训练完成的第一分割子网络，得到分辨率大小为576×576×96，288×288×48，144×144×24，以及72×72×12的4个特征图，以及2个分辨率为72×72×12的概率输出，根据低分辨率的概率输出，利用最大值比较可以得到心脏腔体的粗略分割结果，其分辨率为72×72×12，基于这一粗略分割结果，可以计算心脏腔体的重心坐标，并以此为中心裁剪出在576×576×96，288×288×48，144×144×24，以及72×72×12这4个特征图中，足够完全覆盖目标腔体的固定大小区域，在一个示例中，72×72×12的特征图中可以裁剪30×20×12大小区域，144×144×24的特征图中可以裁剪60×40×24大小区域，288×288×48的特征图中可以裁剪120×80×48大小区域，576×576×96的特征图中可以裁剪240×160×96大小区域。
得到上述四个裁剪后的区域图像后，可以利用这些区域图像对第二分割子网络进行训练，具体的训练过程可以为：
利用逐级反卷积处理，可以将区域图像逐步还原到240×160×96分辨率，具体过程可以为：将72×72×12的特征图中裁剪出的30×20×12大小区域，通过反卷积处理得到分辨率为60×40×24的特征图，并将这一特征图与之前144×144×24的特征图中裁剪出的60×40×24大小区域进行融合，得到融合后的分辨率60×40×24的特征图，再将这一特 征图进行反卷积处理，得到分辨率为120×80×48的特征图，将其余之前288×288×48的特征图中裁剪出的120×80×48大小区域进行融合，得到融合后的分辨率为120×80×48的特征图，将融合后的特征图再进行反卷积处理，得到分辨率为240×160×96的特征图，将其再与576×576×96的特征图中裁剪出的240×160×96大小区域进行融合，得到逐级反卷积处理后的最终图像，则这一最终图像中包含心脏腔体的局部和全局信息，将这一最终图像通过一个softmax层，可以得到两个分辨率为576×576×96的概率输出，这两个概率输出分别代表像素相关位置是否为目标腔体的概率，这两个概率输出可以作为第二分割子网络的输出结果，然后利用dice loss、交叉熵或者其他损失函数，可以计算该输出结果与mask之间的损失，基于计算的损失，可以利用反向传播算法更新第二分割子网络的网络参数，直到第二分割子网络的网络模型收敛，此时可以代表第二分割子网络训练完成。
经过以上步骤，可以得到一个训练完成的用于心脏腔体分割的神经网络，对于心脏腔体的定位和分割可以在这同一个神经网络中同时完成，可以从图像输入经过网络后直接得到。因此，基于这一训练完成的神经网络对心脏腔体的分割过程具体可以为：
首先利用中心裁剪和扩充的方法将待进行心脏腔体分割的待分割图像的分辨率调整为神经网络的预设大小，在本示例中为576×576×96，然后将这一待分割图像数据输入上述训练好的神经网络中，待分割图像在训练好的神经网络中，经历与训练过程相似的过程，即先通过卷积处理生成4个分辨率大小的特征图，然后得到粗略分割结果，基于此粗略分割结果对上述4个分辨率大小的特征图进行裁剪，再对裁剪结果进行反卷积处理得到反卷积结果，这一反卷积结果再通过分割处理得到目标腔体的分割结果，这一分割结果即作为神经网络的输出结果进行输出，再将这一输出的分割结果映射到与输入的待分割图像同样的维度大小上，即得到最终的心脏腔体分割结果。
采用本公开的图像处理方法，可以利用一个三维网络同时进行心脏腔体的定位和分割，定位和分割共享同一套参数，将心脏腔体的定位和分割统一到同一个网络当中，因此可以从输入一步直接得到分割结果，具有更快的速度，也可以节省更多的存储空间，同时可以得到更加平滑的三维模型分割表面。
需要说明的是，本公开实施例的图像处理方法不限于应用在上述心脏腔体图像处理，可以应用于任意的图像处理，本公开对此不作限定。
可以理解，本公开提及的上述各个方法实施例，在不违背原理逻辑的情况下，均可以彼此相互结合形成结合后的实施例，限于篇幅，本公开不再赘述。
本领域技术人员可以理解，在具体实施方式的上述方法中，各步骤的撰写顺序并不意味着严格的执行顺序而对实施过程构成任何限定，各步骤的具体执行顺序应当以其功能和可能的内在逻辑确定。
图9示出根据本公开实施例的图像处理装置的框图。该图像处理装置可以为终端设备、服务器或者其他处理设备等。其中，终端设备可以为用户设备(User Equipment，UE)、移动设备、用户终端、终端、蜂窝电话、无绳电话、个人数字处理(Personal Digital Assistant，PDA)、手持设备、计算设备、车载设备、可穿戴设备等。
在一些可能的实现方式中，该图像处理装置可以通过处理器调用存储器中存储的计算机可读指令的方式来实现。
如图9所示，所述图像处理装置可以包括：卷积模块21，用于对待处理图像进行逐级卷积处理，得到卷积结果；定位模块22，用于根据卷积结果，通过定位处理得到定位结果；反卷积模块23，用于对定位结果进行逐级反卷积处理，得到反卷积结果；目标对象获取模块24，用于对反卷积结果进行分割处理，从待处理图像中分割出目标对象。
在一种可能的实现方式中，卷积模块用于：对待处理图像进行逐级卷积处理，得到至少一个分辨率逐渐递减的特征图，作为卷积结果。
在一种可能的实现方式中，卷积模块进一步用于：对待处理图像进行卷积处理，所得到的特征图作为待卷积特征图；在待卷积特征图的分辨率未达到第一阈值时，对待卷积特征图进行卷积处理，并将得到的结果再次作为待卷积特征图；在待卷积特征图的分辨率达到第一阈值时，将得到的分辨率逐渐递减的所有特征图作为卷积结果。
在一种可能的实现方式中，定位模块包括：分割子模块，用于根据卷积结果进行分割处理，得到分割结果；定位子模块，用于根据分割结果，对卷积结果进行定位处理，得到定位结果。
在一种可能的实现方式中，分割子模块用于：对卷积结果中分辨率最低的特征图进行分割处理，得到分割结果。
在一种可能的实现方式中，定位子模块用于：根据分割结果，确定目标对象在卷积结果中对应的位置信息；根据位置信息，对卷积结果进行定位处理，得到定位结果。
在一种可能的实现方式中，定位子模块进一步用于：读取分割结果的坐标位置；将坐标位置作为区域中心，分别确定卷积结果内，每个分辨率下的特征图中可全部覆盖目标对象的区域位置，作为目标对象在卷积结果中对应的位置信息。
在一种可能的实现方式中，定位子模块进一步用于：根据位置信息，对卷积结果中每个分辨率下的特征图分别进行裁切处理，得到定位结果。
在一种可能的实现方式中，反卷积模块用于：将定位结果中包含的所有特征图中分辨率最低的特征图作为待反卷积特征图；在待反卷积特征图的分辨率未达到第二阈值时，对待反卷积特征图进行反卷积处理，得到反卷积处理结果；按照分辨率逐渐递增的顺序，确定定位结果中待反卷积特征图的下一特征图；将反卷积处理结果与下一特征图进行融合，将融合结果再次作为待反卷积特征图；在待反卷积特征图的分辨率达到第二阈值时，将待反卷积特征图作为反卷积结果。
在一种可能的实现方式中，分割处理包括：将待分割对象通过softmax回归，得到回归结果；通过对回归结果进行最大值比较，完成对待分割对象的分割处理。
在一种可能的实现方式中，装置通过神经网络实现，神经网络包括第一分割子网络及第二分割子网络，其中，第一分割子网络用于对待处理图像进行逐级卷积处理及分割处理，第二分割子网络用于对定位结果进行逐级反卷积处理及分割处理。
在一种可能的实现方式中，装置还包括训练模块，用于：根据预设的训练集，训练第一分割子网络；根据预设的训练集以及已训练的第一分割子网络，训练第二分割子网络。
在一种可能的实现方式中，卷积模块之前还包括分辨率调整模块，用于：将待处理图像调整至预设分辨率。
本公开实施例还提出一种计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令被处理器执行时实现上述方法。计算机可读存储介质可以是非易失性计算机可读存储介质。
本公开实施例还提出一种电子设备，包括：处理器；用于存储处理器可执行指令的存储器；其中，所述处理器被配置为上述方法。
电子设备可以被提供为终端、服务器或其它形态的设备。
图10是根据本公开实施例的一种电子设备800的框图。例如，电子设备800可以是移动电话，计算机，数字广播终端，消息收发设备，游戏控制台，平板设备，医疗设备，健身设备，个人数字助理等终端。
参照图10，电子设备800可以包括以下一个或多个组件：处理组件802，存储器804，电源组件806，多媒体组件808，音频组件810，输入/输出(I/O)的接口812，传感器组件814，以及通信组件816。
处理组件802通常控制电子设备800的整体操作，诸如与显示，电话呼叫，数据通信，相机操作和记录操作相关联 的操作。处理组件802可以包括一个或多个处理器820来执行指令，以完成上述的方法的全部或部分步骤。此外，处理组件802可以包括一个或多个模块，便于处理组件802和其他组件之间的交互。例如，处理组件802可以包括多媒体模块，以方便多媒体组件808和处理组件802之间的交互。
存储器804被配置为存储各种类型的数据以支持在电子设备800的操作。这些数据的示例包括用于在电子设备800上操作的任何应用程序或方法的指令，联系人数据，电话簿数据，消息，图片，视频等。存储器804可以由任何类型的易失性或非易失性存储设备或者它们的组合实现，如静态随机存取存储器(SRAM)，电可擦除可编程只读存储器(EEPROM)，可擦除可编程只读存储器(EPROM)，可编程只读存储器(PROM)，只读存储器(ROM)，磁存储器，快闪存储器，磁盘或光盘。
电源组件806为电子设备800的各种组件提供电力。电源组件806可以包括电源管理系统，一个或多个电源，及其他与为电子设备800生成、管理和分配电力相关联的组件。
多媒体组件808包括在所述电子设备800和用户之间的提供一个输出接口的屏幕。在一些实施例中，屏幕可以包括液晶显示器(LCD)和触摸面板(TP)。如果屏幕包括触摸面板，屏幕可以被实现为触摸屏，以接收来自用户的输入信号。触摸面板包括一个或多个触摸传感器以感测触摸、滑动和触摸面板上的手势。所述触摸传感器可以不仅感测触摸或滑动动作的边界，而且还检测与所述触摸或滑动操作相关的持续时间和压力。在一些实施例中，多媒体组件808包括一个前置摄像头和/或后置摄像头。当电子设备800处于操作模式，如拍摄模式或视频模式时，前置摄像头和/或后置摄像头可以接收外部的多媒体数据。每个前置摄像头和后置摄像头可以是一个固定的光学透镜系统或具有焦距和光学变焦能力。
音频组件810被配置为输出和/或输入音频信号。例如，音频组件810包括一个麦克风(MIC)，当电子设备800处于操作模式，如呼叫模式、记录模式和语音识别模式时，麦克风被配置为接收外部音频信号。所接收的音频信号可以被进一步存储在存储器804或经由通信组件816发送。在一些实施例中，音频组件810还包括一个扬声器，用于输出音频信号。
I/O接口812为处理组件802和外围接口模块之间提供接口，上述外围接口模块可以是键盘，点击轮，按钮等。这些按钮可包括但不限于：主页按钮、音量按钮、启动按钮和锁定按钮。
传感器组件814包括一个或多个传感器，用于为电子设备800提供各个方面的状态评估。例如，传感器组件814可以检测到电子设备800的打开/关闭状态，组件的相对定位，例如所述组件为电子设备800的显示器和小键盘，传感器组件814还可以检测电子设备800或电子设备800一个组件的位置改变，用户与电子设备800接触的存在或不存在，电子设备800方位或加速/减速和电子设备800的温度变化。传感器组件814可以包括接近传感器，被配置用来在没有任何的物理接触时检测附近物体的存在。传感器组件814还可以包括光传感器，如CMOS或CCD图像传感器，用于在成像应用中使用。在一些实施例中，该传感器组件814还可以包括加速度传感器，陀螺仪传感器，磁传感器，压力传感器或温度传感器。
通信组件816被配置为便于电子设备800和其他设备之间有线或无线方式的通信。电子设备800可以接入基于通信标准的无线网络，如WiFi，2G或3G，或它们的组合。在一个示例性实施例中，通信组件816经由广播信道接收来自外部广播管理系统的广播信号或广播相关信息。在一个示例性实施例中，所述通信组件816还包括近场通信(NFC)模块，以促进短程通信。例如，在NFC模块可基于射频识别(RFID)技术，红外数据协会(IrDA)技术，超宽带(UWB)技术，蓝牙(BT)技术和其他技术来实现。
在示例性实施例中，电子设备800可以被一个或多个应用专用集成电路(ASIC)、数字信号处理器(DSP)、数字 信号处理设备(DSPD)、可编程逻辑器件(PLD)、现场可编程门阵列(FPGA)、控制器、微控制器、微处理器或其他电子元件实现，用于执行上述方法。
在示例性实施例中，还提供了一种非易失性计算机可读存储介质，例如包括计算机程序指令的存储器804，上述计算机程序指令可由电子设备800的处理器820执行以完成上述方法。
图11是根据本公开实施例的一种电子设备1900的框图。例如，电子设备1900可以被提供为一服务器。参照图11，电子设备1900包括处理组件1922，其进一步包括一个或多个处理器，以及由存储器1932所代表的存储器资源，用于存储可由处理组件1922的执行的指令，例如应用程序。存储器1932中存储的应用程序可以包括一个或一个以上的每一个对应于一组指令的模块。此外，处理组件1922被配置为执行指令，以执行上述方法。
电子设备1900还可以包括一个电源组件1926被配置为执行电子设备1900的电源管理，一个有线或无线网络接口1950被配置为将电子设备1900连接到网络，和一个输入输出(I/O)接口1958。电子设备1900可以操作基于存储在存储器1932的操作系统，例如Windows ServerTM，Mac OS XTM，UnixTM,LinuxTM，FreeBSDTM或类似。
在示例性实施例中，还提供了一种非易失性计算机可读存储介质，例如包括计算机程序指令的存储器1932，上述计算机程序指令可由电子设备1900的处理组件1922执行以完成上述方法。
本公开可以是系统、方法和/或计算机程序产品。计算机程序产品可以包括计算机可读存储介质，其上载有用于使处理器实现本公开的各个方面的计算机可读程序指令。
计算机可读存储介质可以是可以保持和存储由指令执行设备使用的指令的有形设备。计算机可读存储介质例如可以是――但不限于――电存储设备、磁存储设备、光存储设备、电磁存储设备、半导体存储设备或者上述的任意合适的组合。计算机可读存储介质的更具体的例子(非穷举的列表)包括：便携式计算机盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、静态随机存取存储器(SRAM)、便携式压缩盘只读存储器(CD-ROM)、数字多功能盘(DVD)、记忆棒、软盘、机械编码设备、例如其上存储有指令的打孔卡或凹槽内凸起结构、以及上述的任意合适的组合。这里所使用的计算机可读存储介质不被解释为瞬时信号本身，诸如无线电波或者其他自由传播的电磁波、通过波导或其他传输媒介传播的电磁波(例如，通过光纤电缆的光脉冲)、或者通过电线传输的电信号。
这里所描述的计算机可读程序指令可以从计算机可读存储介质下载到各个计算/处理设备，或者通过网络、例如因特网、局域网、广域网和/或无线网下载到外部计算机或外部存储设备。网络可以包括铜传输电缆、光纤传输、无线传输、路由器、防火墙、交换机、网关计算机和/或边缘服务器。每个计算/处理设备中的网络适配卡或者网络接口从网络接收计算机可读程序指令，并转发该计算机可读程序指令，以供存储在各个计算/处理设备中的计算机可读存储介质中。
用于执行本公开操作的计算机程序指令可以是汇编指令、指令集架构(ISA)指令、机器指令、机器相关指令、微代码、固件指令、状态设置数据、或者以一种或多种编程语言的任意组合编写的源代码或目标代码，所述编程语言包括面向对象的编程语言—诸如Smalltalk、C++等，以及常规的过程式编程语言—诸如“C”语言或类似的编程语言。计算机可读程序指令可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或服务器上执行。在涉及远程计算机的情形中，远程计算机可以通过任意种类的网络—包括局域网(LAN)或广域网(WAN)—连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。在一些实施例中，通过利用计算机可读程序指令的状态信息来个性化定制电子电路，例如可编程逻辑电路、现场可编程门阵列(FPGA)或可编程逻辑阵列(PLA)，该电子电 路可以执行计算机可读程序指令，从而实现本公开的各个方面。
这里参照根据本公开实施例的方法、装置(系统)和计算机程序产品的流程图和/或框图描述了本公开的各个方面。应当理解，流程图和/或框图的每个方框以及流程图和/或框图中各方框的组合，都可以由计算机可读程序指令实现。
这些计算机可读程序指令可以提供给通用计算机、专用计算机或其它可编程数据处理装置的处理器，从而生产出一种机器，使得这些指令在通过计算机或其它可编程数据处理装置的处理器执行时，产生了实现流程图和/或框图中的一个或多个方框中规定的功能/动作的装置。也可以把这些计算机可读程序指令存储在计算机可读存储介质中，这些指令使得计算机、可编程数据处理装置和/或其他设备以特定方式工作，从而，存储有指令的计算机可读介质则包括一个制造品，其包括实现流程图和/或框图中的一个或多个方框中规定的功能/动作的各个方面的指令。
也可以把计算机可读程序指令加载到计算机、其它可编程数据处理装置、或其它设备上，使得在计算机、其它可编程数据处理装置或其它设备上执行一系列操作步骤，以产生计算机实现的过程，从而使得在计算机、其它可编程数据处理装置、或其它设备上执行的指令实现流程图和/或框图中的一个或多个方框中规定的功能/动作。
附图中的流程图和框图显示了根据本公开的多个实施例的系统、方法和计算机程序产品的可能实现的体系架构、功能和操作。在这点上，流程图或框图中的每个方框可以代表一个模块、程序段或指令的一部分，所述模块、程序段或指令的一部分包含一个或多个用于实现规定的逻辑功能的可执行指令。在有些作为替换的实现中，方框中所标注的功能也可以以不同于附图中所标注的顺序发生。例如，两个连续的方框实际上可以基本并行地执行，它们有时也可以按相反的顺序执行，这依所涉及的功能而定。也要注意的是，框图和/或流程图中的每个方框、以及框图和/或流程图中的方框的组合，可以用执行规定的功能或动作的专用的基于硬件的系统来实现，或者可以用专用硬件与计算机指令的组合来实现。
在不违背逻辑的情况下，本申请不同实施例之间可以相互结合，不同实施例描述有所侧重，为侧重描述的部分可以参见其他实施例的记载。以上已经描述了本公开的各实施例，上述说明是示例性的，并非穷尽性的，并且也不限于所披露的各实施例。在不偏离所说明的各实施例的范围和精神的情况下，对于本技术领域的普通技术人员来说许多修改和变更都是显而易见的。本文中所用术语的选择，旨在最好地解释各实施例的原理、实际应用或对市场中的技术改进，或者使本技术领域的其它普通技术人员能理解本文披露的各实施例。

Claims (31)

  1. 一种图像处理方法，其特征在于，包括：
    对待处理图像进行逐级卷积处理，得到卷积结果；
    根据所述卷积结果，通过定位处理得到定位结果；
    对所述定位结果进行逐级反卷积处理，得到反卷积结果；
    对所述反卷积结果进行分割处理，从所述待处理图像中分割出目标对象。
  2. 根据权利要求1所述的方法，其特征在于，所述对待处理图像进行逐级卷积处理，得到卷积结果，包括：
    对待处理图像进行逐级卷积处理，得到至少一个分辨率逐渐递减的特征图，作为所述卷积结果。
  3. 根据权利要求2所述的方法，其特征在于，所述对待处理图像进行逐级卷积处理，得到至少一个分辨率逐渐递减的特征图，作为所述卷积结果，包括：
    对待处理图像进行卷积处理，所得到的特征图作为待卷积特征图；
    在所述待卷积特征图的分辨率未达到第一阈值时，对所述待卷积特征图进行卷积处理，并将得到的结果再次作为待卷积特征图；
    在所述待卷积特征图的分辨率达到第一阈值时，将得到的分辨率逐渐递减的所有特征图作为所述卷积结果。
  4. 根据权利要求1-3中任意一项所述的方法，其特征在于，所述根据所述卷积结果，通过定位处理得到定位结果，包括：
    根据所述卷积结果进行分割处理，得到分割结果；
    根据所述分割结果，对所述卷积结果进行定位处理，得到定位结果。
  5. 根据权利要求4所述的方法，其特征在于，所述根据所述卷积结果进行分割处理，得到分割结果，包括：
    对所述卷积结果中分辨率最低的特征图进行分割处理，得到分割结果。
  6. 根据权利要求4或5所述的方法，其特征在于，所述根据所述分割结果，对所述卷积结果进行定位处理，得到定位结果，包括：
    根据所述分割结果，确定所述目标对象在所述卷积结果中对应的位置信息；
    根据所述位置信息，对所述卷积结果进行定位处理，得到定位结果。
  7. 根据权利要求6所述的方法，其特征在于，所述根据所述分割结果，确定所述目标对象在卷积结果中对应的位置信息，包括：
    读取所述分割结果的坐标位置；
    将所述坐标位置作为区域中心，分别确定所述卷积结果内，每个分辨率下的特征图中可全部覆盖所述目标对象的区域位置，作为所述目标对象在卷积结果中对应的位置信息。
  8. 根据权利要求6或7所述的方法，其特征在于，所述根据所述位置信息，对所述卷积结果进行定位处理，得到定位结果，包括：
    根据所述位置信息，对所述卷积结果中每个分辨率下的特征图分别进行裁切处理，得到定位结果。
  9. 根据权利要求1-7中任意一项所述的方法，其特征在于，所述对所述定位结果进行逐级反卷积处理，得到反卷积结果，包括：
    将所述定位结果中包含的所有特征图中分辨率最低的特征图作为待反卷积特征图；
    在所述待反卷积特征图的分辨率未达到第二阈值时，对所述待反卷积特征图进行反卷积处理，得到反卷积处理结果；
    按照分辨率逐渐递增的顺序，确定所述定位结果中所述待反卷积特征图的下一特征图；
    将所述反卷积处理结果与所述下一特征图进行融合，将所述融合结果再次作为待反卷积特征图；
    在所述待反卷积特征图的分辨率达到第二阈值时，将所述待反卷积特征图作为反卷积结果。
  10. 根据权利要求1-9中任意一项所述的方法，其特征在于，所述分割处理包括：
    将待分割对象通过softmax回归，得到回归结果；
    通过对所述回归结果进行最大值比较，完成对所述待分割对象的分割处理。
  11. 根据权利要求1-10中任意一项所述的方法，其特征在于，所述方法通过神经网络实现，所述神经网络包括第一分割子网络及第二分割子网络，
    其中，所述第一分割子网络用于对所述待处理图像进行逐级卷积处理及分割处理，所述第二分割子网络用于对所述定位结果进行逐级反卷积处理及分割处理。
  12. 根据权利要求11所述的方法，其特征在于，所述神经网络的训练过程，包括：
    根据预设的训练集，训练所述第一分割子网络；
    根据所述预设的训练集以及已训练的第一分割子网络，训练所述第二分割子网络。
  13. 根据权利要求1-12中任意一项所述的方法，其特征在于，所述对所述待处理图像进行逐级卷积处理，得到卷积结果之前，还包括：
    将所述待处理图像调整至预设分辨率。
  14. 根据权利要求1-13中任意一项所述的方法，其特征在于，所述待处理图像为三维医学图像。
  15. 一种图像处理装置，其特征在于，包括：
    卷积模块，用于对待处理图像进行逐级卷积处理，得到卷积结果；
    定位模块，用于根据所述卷积结果，通过定位处理得到定位结果；
    反卷积模块，用于对所述定位结果进行逐级反卷积处理，得到反卷积结果；
    目标对象获取模块，用于对所述反卷积结果进行分割处理，从所述待处理图像中分割出目标对象。
  16. 根据权利要求15所述的装置，其特征在于，所述卷积模块用于：
    对待处理图像进行逐级卷积处理，得到至少一个分辨率逐渐递减的特征图，作为所述卷积结果。
  17. 根据权利要求16所述的装置，其特征在于，所述卷积模块进一步用于：
    对待处理图像进行卷积处理，所得到的特征图作为待卷积特征图；
    在所述待卷积特征图的分辨率未达到第一阈值时，对所述待卷积特征图进行卷积处理，并将得到的结果再次作为待卷积特征图；
    在所述待卷积特征图的分辨率达到第一阈值时，将得到的分辨率逐渐递减的所有特征图作为所述卷积结果。
  18. 根据权利要求15-17中任意一项所述的装置，其特征在于，所述定位模块包括：
    分割子模块，用于根据所述卷积结果进行分割处理，得到分割结果；
    定位子模块，用于根据所述分割结果，对所述卷积结果进行定位处理，得到定位结果。
  19. 根据权利要求18所述的装置，其特征在于，所述分割子模块用于：
    对所述卷积结果中分辨率最低的特征图进行分割处理，得到分割结果。
  20. 根据权利要求18或19所述的装置，其特征在于，所述定位子模块用于：
    根据所述分割结果，确定所述目标对象在所述卷积结果中对应的位置信息；
    根据所述位置信息，对所述卷积结果进行定位处理，得到定位结果。
  21. 根据权利要求20所述的装置，其特征在于，所述定位子模块进一步用于：
    读取所述分割结果的坐标位置；
    将所述坐标位置作为区域中心，分别确定所述卷积结果内，每个分辨率下的特征图中可全部覆盖所述目标对象的区域位置，作为所述目标对象在卷积结果中对应的位置信息。
  22. 根据权利要求20或21所述的装置，其特征在于，所述定位子模块进一步用于：
    根据所述位置信息，对所述卷积结果中每个分辨率下的特征图分别进行裁切处理，得到定位结果。
  23. 根据权利要求15-22中任意一项所述的装置，其特征在于，所述反卷积模块用于：
    将所述定位结果中包含的所有特征图中分辨率最低的特征图作为待反卷积特征图；
    在所述待反卷积特征图的分辨率未达到第二阈值时，对所述待反卷积特征图进行反卷积处理，得到反卷积处理结果；
    按照分辨率逐渐递增的顺序，确定所述定位结果中所述待反卷积特征图的下一特征图；
    将所述反卷积处理结果与所述下一特征图进行融合，将所述融合结果再次作为待反卷积特征图；
    在所述待反卷积特征图的分辨率达到第二阈值时，将所述待反卷积特征图作为反卷积结果。
  24. 根据权利要求15-23中任意一项所述的装置，其特征在于，所述分割处理包括：
    将待分割对象通过softmax回归，得到回归结果；
    通过对所述回归结果进行最大值比较，完成对所述待分割对象的分割处理。
  25. 根据权利要求15-24中任意一项所述的装置，其特征在于，所述装置通过神经网络实现，所述神经网络包括第一分割子网络及第二分割子网络，
    其中，所述第一分割子网络用于对所述待处理图像进行逐级卷积处理及分割处理，所述第二分割子网络用于对所述定位结果进行逐级反卷积处理及分割处理。
  26. 根据权利要求25所述的装置，其特征在于，所述装置还包括训练模块，用于：
    根据预设的训练集，训练所述第一分割子网络；
    根据所述预设的训练集以及已训练的第一分割子网络，训练所述第二分割子网络。
  27. 根据权利要求15-26中任意一项所述的方法，其特征在于，所述卷积模块之前还包括分辨率调整模块，用于：
    将所述待处理图像调整至预设分辨率。
  28. 根据权利要求15-27中任意一项所述的装置，其特征在于，所述待处理图像为三维医学图像。
  29. 一种电子设备，其特征在于，包括：
    处理器；
    用于存储处理器可执行指令的存储器；
    其中，所述处理器被配置为：执行权利要求1至14中任意一项所述的方法。
  30. 一种计算机可读存储介质，其上存储有计算机程序指令，其特征在于，所述计算机程序指令被处理器执行时实现权利要求1至14中任意一项所述的方法。
  31. 一种计算机程序，其特征在于，所述计算机程序包括计算机可读代码，当所述计算机可读代码在电子设备中运行时，所述电子设备中的处理器执行用于实现权利要求1至14中的任意一项所述的方法。
PCT/CN2019/107844 2019-04-01 2019-09-25 图像处理方法及装置、电子设备和存储介质 WO2020199528A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
JP2021539065A JP2022517571A (ja) 2019-04-01 2019-09-25 画像処理方法及び装置、電子機器並びに記憶媒体
SG11202106290TA SG11202106290TA (en) 2019-04-01 2019-09-25 Image processing method and apparatus, electronic device, and storage medium
US17/356,398 US20210319560A1 (en) 2019-04-01 2021-06-23 Image processing method and apparatus, and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910258038.1A CN109978886B (zh) 2019-04-01 2019-04-01 图像处理方法及装置、电子设备和存储介质
CN201910258038.1 2019-04-01

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/356,398 Continuation US20210319560A1 (en) 2019-04-01 2021-06-23 Image processing method and apparatus, and storage medium

Publications (1)

Publication Number Publication Date
WO2020199528A1 true WO2020199528A1 (zh) 2020-10-08

Family

ID=67082222

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/107844 WO2020199528A1 (zh) 2019-04-01 2019-09-25 图像处理方法及装置、电子设备和存储介质

Country Status (6)

Country Link
US (1) US20210319560A1 (zh)
JP (1) JP2022517571A (zh)
CN (1) CN109978886B (zh)
SG (1) SG11202106290TA (zh)
TW (3) TWI758233B (zh)
WO (1) WO2020199528A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113012178A (zh) * 2021-05-07 2021-06-22 西安智诊智能科技有限公司 一种肾脏肿瘤图像分割方法
US11841923B2 (en) 2020-07-06 2023-12-12 Alibaba Group Holding Limited Processing method, model training method, means, and storage medium for spinal images

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109978886B (zh) * 2019-04-01 2021-11-09 北京市商汤科技开发有限公司 图像处理方法及装置、电子设备和存储介质
CN110807463B (zh) * 2019-09-17 2022-10-11 珠海格力电器股份有限公司 图像分割方法、装置、计算机设备和存储介质
US11335045B2 (en) * 2020-01-03 2022-05-17 Gyrfalcon Technology Inc. Combining feature maps in an artificial intelligence semiconductor solution
CN113706548B (zh) * 2020-05-09 2023-08-22 北京康兴顺达科贸有限公司 基于ct影像对胸部前纵隔病灶自动分割的方法
CN112150449B (zh) * 2020-09-29 2022-11-25 太原理工大学 一种脑卒中病灶分割方法和系统
CN112233194B (zh) * 2020-10-15 2023-06-02 平安科技（深圳）有限公司 医学图片优化方法、装置、设备及计算机可读存储介质
CN112308867B (zh) * 2020-11-10 2022-07-22 上海商汤智能科技有限公司 牙齿图像的处理方法及装置、电子设备和存储介质
US11461989B2 (en) * 2020-12-04 2022-10-04 Himax Technologies Limited Monitor method and monitor system thereof wherein mask is used to cover image for detecting object
TWI768759B (zh) * 2021-03-11 2022-06-21 瑞昱半導體股份有限公司 具有超解析放大機制的影像放大裝置及方法
CN113225226B (zh) * 2021-04-30 2022-10-21 上海爱数信息技术股份有限公司 一种基于信息熵的云原生系统观测方法及系统
US11475158B1 (en) * 2021-07-26 2022-10-18 Netskope, Inc. Customized deep learning classifier for detecting organization sensitive data in images on premises
CN114708608B (zh) * 2022-06-06 2022-09-16 浙商银行股份有限公司 一种银行票据全自动化特征工程方法及装置

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018048575A1 (en) * 2016-09-07 2018-03-15 Elekta, Inc. System and method for learning models of radiotherapy treatment plans to predict radiotherapy dose distributions
CN108986891A (zh) * 2018-07-24 2018-12-11 北京市商汤科技开发有限公司 医疗影像处理方法及装置、电子设备及存储介质
CN109063609A (zh) * 2018-07-18 2018-12-21 电子科技大学 一种基于光流特征与全卷积语义分割特征结合的异常行为检测方法
CN109191476A (zh) * 2018-09-10 2019-01-11 重庆邮电大学 基于U-net网络结构的生物医学图像自动分割新方法
CN109493317A (zh) * 2018-09-25 2019-03-19 哈尔滨理工大学 基于级联卷积神经网络的3d多椎骨分割方法
CN109978886A (zh) * 2019-04-01 2019-07-05 北京市商汤科技开发有限公司 图像处理方法及装置、电子设备和存储介质

Family Cites Families (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7058210B2 (en) * 2001-11-20 2006-06-06 General Electric Company Method and system for lung disease detection
US7899514B1 (en) * 2006-01-26 2011-03-01 The United States Of America As Represented By The Secretary Of The Army Medical image processing methodology for detection and discrimination of objects in tissue
WO2016172612A1 (en) * 2015-04-23 2016-10-27 Cedars-Sinai Medical Center Automated delineation of nuclei for three dimensional (3-d) high content screening
WO2017158575A1 (en) * 2016-03-17 2017-09-21 Imagia Cybernetics Inc. Method and system for processing a task with robustness to missing input information
CN106530320B (zh) * 2016-09-30 2019-12-17 深圳大学 一种端到端的图像分割处理方法及系统
US20190295260A1 (en) * 2016-10-31 2019-09-26 Konica Minolta Laboratory U.S.A., Inc. Method and system for image segmentation using controlled feedback
TWI624804B (zh) * 2016-11-07 2018-05-21 盾心科技股份有限公司 利用超解析重建法生成高解析度影像的方法與系統
JP6787196B2 (ja) * 2017-03-09 2020-11-18 コニカミノルタ株式会社 画像認識装置及び画像認識方法
CN107016681B (zh) * 2017-03-29 2023-08-25 浙江师范大学 基于全卷积网络的脑部mri肿瘤分割方法
JP6972757B2 (ja) * 2017-08-10 2021-11-24 富士通株式会社 制御プログラム、制御方法、及び情報処理装置
CN108776969B (zh) * 2018-05-24 2021-06-22 复旦大学 基于全卷积网络的乳腺超声图像肿瘤分割方法
CN108682015B (zh) * 2018-05-28 2021-10-19 安徽科大讯飞医疗信息技术有限公司 一种生物图像中的病灶分割方法、装置、设备及存储介质
CN108765422A (zh) * 2018-06-13 2018-11-06 云南大学 一种视网膜图像血管自动分割方法
CN108986115B (zh) * 2018-07-12 2020-12-18 佛山生物图腾科技有限公司 医学图像分割方法、装置及智能终端
CN109145769A (zh) * 2018-08-01 2019-01-04 辽宁工业大学 融合图像分割特征的目标检测网络设计方法
CN109166130B (zh) * 2018-08-06 2021-06-22 北京市商汤科技开发有限公司 一种图像处理方法及图像处理装置
CN109035261B (zh) * 2018-08-09 2023-01-10 北京市商汤科技开发有限公司 医疗影像处理方法及装置、电子设备及存储介质
CN109493343A (zh) * 2018-12-29 2019-03-19 上海鹰瞳医疗科技有限公司 医疗图像异常区域分割方法及设备

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018048575A1 (en) * 2016-09-07 2018-03-15 Elekta, Inc. System and method for learning models of radiotherapy treatment plans to predict radiotherapy dose distributions
CN109063609A (zh) * 2018-07-18 2018-12-21 电子科技大学 一种基于光流特征与全卷积语义分割特征结合的异常行为检测方法
CN108986891A (zh) * 2018-07-24 2018-12-11 北京市商汤科技开发有限公司 医疗影像处理方法及装置、电子设备及存储介质
CN109191476A (zh) * 2018-09-10 2019-01-11 重庆邮电大学 基于U-net网络结构的生物医学图像自动分割新方法
CN109493317A (zh) * 2018-09-25 2019-03-19 哈尔滨理工大学 基于级联卷积神经网络的3d多椎骨分割方法
CN109978886A (zh) * 2019-04-01 2019-07-05 北京市商汤科技开发有限公司 图像处理方法及装置、电子设备和存储介质

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11841923B2 (en) 2020-07-06 2023-12-12 Alibaba Group Holding Limited Processing method, model training method, means, and storage medium for spinal images
CN113012178A (zh) * 2021-05-07 2021-06-22 西安智诊智能科技有限公司 一种肾脏肿瘤图像分割方法

Also Published As

Publication number Publication date
TWI758233B (zh) 2022-03-11
TW202207156A (zh) 2022-02-16
CN109978886B (zh) 2021-11-09
CN109978886A (zh) 2019-07-05
JP2022517571A (ja) 2022-03-09
TWI758234B (zh) 2022-03-11
SG11202106290TA (en) 2021-07-29
TW202209343A (zh) 2022-03-01
TWI750518B (zh) 2021-12-21
TW202038188A (zh) 2020-10-16
US20210319560A1 (en) 2021-10-14

Similar Documents

Publication Publication Date Title
WO2020199528A1 (zh) 图像处理方法及装置、电子设备和存储介质
CN110047078B (zh) 图像处理方法及装置、电子设备和存储介质
WO2021164469A1 (zh) 目标对象的检测方法、装置、设备和存储介质
CN111310616B (zh) 图像处理方法及装置、电子设备和存储介质
WO2021051650A1 (zh) 人脸和人手关联检测方法及装置、电子设备和存储介质
JP7262659B2 (ja) 目標対象物マッチング方法及び装置、電子機器並びに記憶媒体
WO2021047267A1 (zh) 图像处理方法及装置、电子设备、存储介质和计算机程序
WO2021056808A1 (zh) 图像处理方法及装置、电子设备和存储介质
JP2022537974A (ja) ニューラルネットワーク訓練方法及び装置、電子機器並びに記憶媒体
KR20200113195A (ko) 이미지 클러스터링 방법 및 장치, 전자 기기 및 저장 매체
CN112767329B (zh) 图像处理方法及装置、电子设备
JP2022515722A (ja) 画像セグメンテーション方法及び装置、電子機器並びに記憶媒体
JP7178514B2 (ja) 場面深度とカメラ運動を予測する方法及び装置、機器、媒体並びにプログラム
CN112541928A (zh) 网络训练方法及装置、图像分割方法及装置和电子设备
US20210279892A1 (en) Image processing method and device, and network training method and device
JP2022522551A (ja) 画像処理方法及び装置、電子機器並びに記憶媒体
WO2020192113A1 (zh) 图像处理方法及装置、电子设备和存储介质
CN111724361B (zh) 实时显示病灶的方法及装置、电子设备和存储介质
TW202121343A (zh) 圖像處理方法及裝置、電子設備和電腦可讀儲存媒體
TW202226049A (zh) 關鍵點檢測方法、電子設備和儲存媒體
CN112613447A (zh) 关键点检测方法及装置、电子设备和存储介质
WO2023155393A1 (zh) 特征点匹配方法、装置、电子设备、存储介质和计算机程序产品
CN111738998A (zh) 病灶位置动态检测方法及装置、电子设备和存储介质
CN117152128B (zh) 神经影像的病灶识别方法、装置、电子设备和存储介质
CN111723715B (zh) 一种视频显著性检测方法及装置、电子设备和存储介质

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: 19923059

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2021539065

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 07.02.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19923059

Country of ref document: EP

Kind code of ref document: A1