CN106157319A - The significance detection method that region based on convolutional neural networks and Pixel-level merge - Google Patents

The significance detection method that region based on convolutional neural networks and Pixel-level merge Download PDF

Info

Publication number
CN106157319A
CN106157319A CN201610604732.0A CN201610604732A CN106157319A CN 106157319 A CN106157319 A CN 106157319A CN 201610604732 A CN201610604732 A CN 201610604732A CN 106157319 A CN106157319 A CN 106157319A
Authority
CN
China
Prior art keywords
pixel
significance
level
cnn
image
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201610604732.0A
Other languages
Chinese (zh)
Other versions
CN106157319B (en
Inventor
邬向前
卜巍
唐有宝
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Harbin Institute of Technology
Original Assignee
Harbin Institute of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Harbin Institute of Technology filed Critical Harbin Institute of Technology
Priority to CN201610604732.0A priority Critical patent/CN106157319B/en
Publication of CN106157319A publication Critical patent/CN106157319A/en
Application granted granted Critical
Publication of CN106157319B publication Critical patent/CN106157319B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • 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
    • G06T2207/20212—Image combination
    • G06T2207/20221—Image fusion; Image merging

Abstract

The invention discloses a kind of region based on convolutional neural networks and the significance detection method of Pixel-level fusion, the method research to as if still image, wherein the content of image can be arbitrary, the target of research is the target found out from image and attract human eye attention, and gives different significance value for it.The present invention mainly proposes a kind of adaptive Area growth, and devises two CNN network structures, is respectively used to the prediction of Pixel-level significance and significance merges.The two CNN network model, using image as input, is used for the training of network model, and finally output and input picture Saliency maps of the same size using the legitimate reading of image as supervisory signals.The present invention can effectively carry out region class significance and estimate and Pixel-level significance is predicted, obtains two Saliency maps, finally uses the CNN carrying out significance fusion to carry out merging by two Saliency maps and original image and obtain final Saliency maps.

Description

The significance detection method that region based on convolutional neural networks and Pixel-level merge
Technical field
The present invention relates to a kind of image processing method based on degree of depth study, be specifically related to a kind of based on convolutional neural networks Region and Pixel-level merge significance detection method.
Background technology
Along with development and the rise of degree of depth study, significance detection technique also development based on degree of depth study.Significantly Property detection can be divided into bottom-up data-driven model and the big class of top-down task-driven model two.Bottom-up aobvious The detection of work property refers to for given any piece image, finds out attractive target in figure, and this target can be any classification Things.And top-down significance detection method generally finds out the target of given classification from given picture, and give not Same significance value.At present, the most study to bottom-up significance detection method.
Existing bottom-up significance detection method can be divided into two classes, is respectively method based on hand-designed feature With method based on convolutional neural networks.Owing to method based on hand-designed feature generally utilizes the surface information in image (such as color, texture etc.) carries out feature extraction, and these the manual features extracted can not catch the deep layer of significance target Characteristic and multi-scale information, thus good performance can not be obtained.Recently along with the rise of degree of depth study, part researcher is had to open Begin to use convolutional neural networks to carry out significance target detection.The most existing it is mostly based on convolutional neural networks (CNN) Significance object detection method first divide the image into into multiple region, then to each region CNN model trained Predict its significance value.But the significance that these methods can not obtain accurate Pixel-level predicts the outcome.
Summary of the invention
In order to overcome the problems referred to above, the present invention proposes a kind of new significance detection method based on CNN, i.e. based on volume The significance detection method that the region of long-pending neutral net and Pixel-level merge.The method comprises three phases, respectively region class Significance is estimated, Pixel-level significance is predicted and significance merges, and each stage is directed to a CNN model.The method can Obtain the Saliency maps of accurate Pixel-level, it is thus possible to more effectively promote the development of related application based on significance.
It is an object of the invention to be achieved through the following technical solutions:
The invention provides a kind of region based on convolutional neural networks and the significance detection method of Pixel-level fusion, the method Research to as if still image, wherein the content of image can be arbitrary, and the target of research is to find out attraction from image The target of human eye attention, and give different significance value for it.It is raw that the present invention mainly proposes a kind of adaptive region One-tenth technology, and devise two CNN network structures, it is respectively used to the prediction of Pixel-level significance and significance merges.The two CNN Network model is using image as input, using the legitimate reading of image as supervisory signals for the training of network model, and finally Output and input picture Saliency maps of the same size.The present invention can effectively carry out region class significance and estimate to show with Pixel-level Work is predicted, obtains two Saliency maps, finally uses and carry out the CNN of significance fusion by two Saliency maps and original image Carrying out merging and obtain final Saliency maps, whole system block diagram is as shown in Figure 1.
The present invention to be embodied as step as follows:
One, region class significance is estimated
The first step, use adaptive Area growth that input picture I is split
(1) use SLIC algorithm that input picture I is carried out super-pixel segmentation, obtain n super-pixel；
(2) from each super-pixel, extract a simple characteristic vector, be used for characterizing the characteristic of this super-pixel；
(3) use an agglomerative clustering algorithm based on figure that super-pixel is clustered and obtain different regions；
Second step, use Clarifai network model carry out region significance estimation
(1) around the centrage in each region, m super-pixel is randomly selected；
(2) center m window of center construction as window of m super-pixel is set, and these windows comprise whole image；
(3) by m video in window of structure through CNN model, m significance value is obtained；
(4) average of m significance value is calculated and as the significance value in this region；
Two, Pixel-level significance prediction
(1) using model VGGNet as pre-training model, by last module removal in VGGNet, to the 4th and the 5th mould The output of block carries out operation of deconvoluting, and they is spliced in feature channel direction, learns for Analysis On Multi-scale Features；Then Using size is that the convolution kernel of 1* 1 carries out convolution and obtains a probability graph spliced characteristic pattern；
(2) in the Pixel-level CNN model training stage, the mistake between fork entropy loss function calculating probability figure and legitimate reading figure is used Difference, and carry out returning error to update Pixel-level CNN model parameter；
(3), after Pixel-level CNN model training is complete, it is directly inputted to input picture I in Pixel-level CNN model predict that it is corresponding Pixel-level Saliency maps；
Three, significance merges
(1) fusion CNN network structure is built: CNN network structure comprises a splicing layer, three convolutional layers and a loss layer；
(2) two Saliency maps of input picture I and step one, two are spliced into the image of 5 passages, are then sent to Three convolutional layers；
(3) merge the CNN network training stage, use the fork entropy loss function in loss layer to calculate the defeated of last convolutional layer Go out the error between true Saliency maps, and carry out returning by error to update fusion CNN model parameter；
(4), during test, input picture I is directly inputted in the fusion CNN model trained, last convolutional layer of this model Output be the Saliency maps of final prediction.
Present invention have the advantage that
1, the present invention proposes a kind of new significance detection method based on CNN, and it is notable that the method has given full play to region class Property estimate and Pixel-level significance prediction advantage, and achieve good significance detection performance.
2, the present invention proposes a kind of adaptive Area growth, and this technology can be that different images generates different number The region of amount, and can well keep the edge of object.
3, the present invention devises a CNN network structure, and this network structure can excavate the multiple dimensioned letter in image effectively Breath, in addition to can be used for the prediction of Pixel-level significance, also can carry out the task relevant to pixel classifications, such as image segmentation.
4, the present invention proposes a kind of new significance convergence strategy based on CNN, not only takes full advantage of Saliency maps Between complementary information, also use in original image abundant information, thus improve significance detection to a great extent Performance.
Accompanying drawing explanation
Fig. 1 is whole system block diagram of the present invention；
Fig. 2 is that adaptive region generates result example, a-original image, b-legitimate reading, c-super-pixel segmentation result, d-region Generate result；
Fig. 3 is region class significance estimated result example, a-original image, b-legitimate reading, c-region class result；
Fig. 4 is Pixel-level CNN network structure；
Fig. 5 is that Pixel-level significance predicts the outcome example, a-original image, b-legitimate reading, c-Pixel-level result；
Fig. 6 is for merging CNN network structure；
Fig. 7 is the result that the present invention carries out significance detection, a-original image, b-legitimate reading, c-fusion results, d-Pixel-level As a result, e-region class result.
Detailed description of the invention
Below in conjunction with the accompanying drawings technical scheme is further described, but is not limited thereto, every to this Inventive technique scheme is modified or equivalent, without deviating from the spirit and scope of technical solution of the present invention, all should contain In protection scope of the present invention.
The invention provides a kind of region based on convolutional neural networks and the significance detection method of Pixel-level fusion, tool It is as follows that body implements step:
One, region class significance is estimated
In region class significance estimation procedure, wherein the first step generates substantial amounts of region exactly from input picture.The simplest Method be use super-pixel as region to carry out significance estimation so that how to determine segmentation super-pixel number become Obtain highly difficult.If super-pixel number is very little, may be by less divided so that belong to same significance mesh target area. If super-pixel number is too many, so that the region belonging to significance target or background may be by over-segmentation.Either owe Segmentation or over-segmentation, the significance value that all may make significance target or background is inconsistent.Therefore, for different figures Picture, due to their different qualities, it should be divided into the super-pixel of varying number.In order to solve the problems referred to above, the present invention carries Go out a kind of adaptive Area growth to carry out image segmentation.A given input picture I, this adaptive region generates skill The process of art is as follows:
(1) use SLIC algorithm that I carries out super-pixel segmentation and obtain n super-pixel.Consider effect and the efficiency of method, In the present invention, n=300.
(2) from each super-pixel, extract a simple characteristic vector and (contain the average color on Lab color space With average locus coordinate), it is used for characterizing the characteristic of this super-pixel.
(3) use an agglomerative clustering algorithm based on figure that super-pixel is clustered and obtain different regions.
After said process, in image I color similarity and adjacent super-pixel be generally clustered same district In territory.For different images, the areal obtained after final cluster is also different, and will be far smaller than super-pixel Number n.Fig. 2 provides the result example of three generations that adaptive region generation technique obtains.
After obtaining the region generated, next step is that region significance is estimated.The present invention uses Clarifai network model (it is the CNN model that in ImageNet2013, image classification task obtains top performance) carries out region significance estimation.Tool For body, around the centrage in each region, first randomly select m super-pixel, then the center of this m super-pixel is set Center as window builds m window, and these windows contain whole image.Select the super picture around regional center line Element is that (2) are from the window of zones of different in order to (1) makes the border as far away from region, the center of constructed window In content the most different.In the present invention, when the number of the super-pixel comprised in region is more than 5, m=5 is set, otherwise, The value of m is set to the number of super-pixel.From the foregoing, it will be observed that for each region, m video in window will be constructed, through CNN model After, m significance value will be obtained, calculate their average and as the significance value in this region, so that this region Significance value to noise more robust.Fig. 3 gives the result example that three region class significances are estimated.
Two, Pixel-level significance prediction
Although region class significance is estimated to obtain the consistent and good Saliency maps of edge holding, but can not obtain Pixel-level The Saliency maps of precision.To this, the present invention proposes a kind of CNN network structure (being designated as Pixel-level CNN) and shows for carrying out Pixel-level Work is predicted.This Pixel-level CNN is with original image for input, with the Saliency maps with the size such as original image for output.In order to Obtaining the prediction of accurate significance, this CNN structure should be deep layer and the multiple dimensioned rank having different stride (strides) Section, to such an extent as to learn to the strong Analysis On Multi-scale Features of discriminating power for image pixel.When training sample scale is less, accent to open Beginning effectively to train such a network structure is a very difficult task.The such issues that of in order to overcome, one well Way is exactly model (such as those nets extremely successful on ImageNet using some to train on large-scale dataset Network model VGGNet and GoogleNet) as pre-training model, then in the small data set in needing of task, model is entered Row fine setting, thus can train and obtain a strong model of learning capacity.
It is modified on the basis of VGGNet model by the present invention, thus builds this Pixel-level CNN model.VGGNet Being made up of six modules (block), first five module is made up of convolutional layer (being designated as conv) and pond layer (being designated as pooling), As shown in Figure 4.Last module is made up of a pond layer and two full articulamentums.The present invention is by last in VGGNet Individual module removal.In order to utilize the multi-scale information of image, the present invention merges the output of the 4th and the 5th module and realizes many Scale feature learns.Owing to the output of latter two module varies in size and much smaller than the size of original image, therefore to make This Pixel-level CNN model can learn Analysis On Multi-scale Features automatically to be predicted for Pixel-level significance, and the present invention first will be to last two The output of individual module carries out operation (being designated as deconv) of deconvoluting so that their size keeps consistent with original image, and They are carried out splicing (being designated as concat) by feature channel direction.Then using size is that the convolution kernel of 1* 1 is to spliced spy Levying figure to carry out convolution and obtain a probability graph, in this probability graph, value means the most greatly the most notable.When test, this probability graph It is actually the Saliency maps of input picture.When training, fork entropy loss function (being designated as loss) is used to calculate this probability Error between figure and legitimate reading figure, and carry out returning to update model parameter by error.Arrive this, whole Pixel-level CNN Network structure has the most all built, as shown in Figure 4.In the model training stage, the stochastic gradient descent algorithm of standard by with Minimize loss function.After model training is complete, it is directly inputted to image in model predict that the Pixel-level of its correspondence shows Work property figure.Fig. 5 gives the example that three Pixel-level significances predict the outcome.
Three, significance merges
For a given image, said process can effectively obtain two Saliency maps, be respectively as follows: region class significance Figure and Pixel-level Saliency maps.Owing to they are that the CNN model that make use of the different information in image is calculated, therefore they There is complementarity.If can effectively it be merged, the performance of significance detection must be will further improve.
The present invention designs a simple CNN network structure (be designated as merge with CNN) and learns a kind of nonlinear transformation and fill Divide the complementary information between excavation regions level Saliency maps and Pixel-level Saliency maps, thus reach to put forward high performance purpose.Should CNN network structure contains splicing layer (concat), three convolutional layers (conv) and a loss layer (loss), such as Fig. 6 Shown in.First original image and its two Saliency maps are spliced into the image of 5 passages, are then sent to follow-up Three convolutional layers (concrete configuration is shown in Fig. 6).When test, the output of last convolutional layer is the significance of final prediction Figure.When training, the fork entropy loss function in loss layer is used to calculate the output of last convolutional layer and true significance Error between figure.From the foregoing, it will be observed that in the significance fusion method that the present invention proposes, obtain Saliency maps except using two Outward, original image is also used.This is because the information that introducing original image enriches can be corrected some and cannot only be used significantly Property the mistake corrected when merging of figure.
This fusion CNN can individually train, it is possible to carries out combination learning to obtain lastness with CNN network above The lifting of energy.In region class significance is estimated, start most to need from input picture, generate multiple region, then use region Level CNN carries out significance value estimation to each region.Pixel-level CNN and fusion CNN are then direct using image as input, and directly Connect output and obtain Saliency maps, be therefore a process end to end.Therefore, it is difficult to above three CNN network structure is incorporated In a unified network, and carry out combination learning end to end.In order to simplify this process, finally, three CNN are the most advanced Row individually training, then, Pixel-level CNN and fusion CNN carry out combination learning, such as Fig. 1 institute on the basis of pre-training further Show.During test, image is inputted the framework in Fig. 1, merge the Saliency maps that CNN is output as finally predicting.Fig. 7 gives The testing result example of the significance detection method of four present invention propositions, as shown in Figure 7, the result of present invention detection is with true Result very close to, thus illustrate effectiveness of the invention.

Claims (2)

1. the significance detection method that a region based on convolutional neural networks and Pixel-level merge, it is characterised in that described side Method step is as follows:
One, region class significance is estimated
The first step, use adaptive Area growth that input picture I is split
(1) use SLIC algorithm that input picture I is carried out super-pixel segmentation, obtain n super-pixel；
(2) from each super-pixel, extract a simple characteristic vector, be used for characterizing the characteristic of this super-pixel；
(3) use an agglomerative clustering algorithm based on figure that super-pixel is clustered and obtain different regions；
Second step, use Clarifai network model carry out region significance estimation
(1) around the centrage in each region, m super-pixel is randomly selected；
(2) center m window of center construction as window of m super-pixel is set, and these windows comprise whole image；
(3) by m video in window of structure through CNN model, m significance value is obtained；
(4) average of m significance value is calculated and as the significance value in this region；
Two, Pixel-level significance prediction
(1) using model VGGNet as pre-training model, by last module removal in VGGNet, to the 4th and the 5th mould The output of block carries out operation of deconvoluting, and they is spliced in feature channel direction, learns for Analysis On Multi-scale Features；Then Using size is that the convolution kernel of 1* 1 carries out convolution and obtains a probability graph spliced characteristic pattern；
(2) in the Pixel-level CNN model training stage, the mistake between fork entropy loss function calculating probability figure and legitimate reading figure is used Difference, and carry out returning error to update Pixel-level CNN model parameter；
(3), after Pixel-level CNN model training is complete, it is directly inputted to input picture I in Pixel-level CNN model predict that it is corresponding Pixel-level Saliency maps；
Three, significance merges
(1) fusion CNN network structure is built: CNN network structure comprises a splicing layer, three convolutional layers and a loss layer；
(2) two Saliency maps of input picture I and step one, two are spliced into the image of 5 passages, are then sent to Three convolutional layers；
(3) merge the CNN network training stage, use the fork entropy loss function in loss layer to calculate the defeated of last convolutional layer Go out the error between true Saliency maps, and carry out returning by error to update fusion CNN model parameter；
(4), during test, input picture I is directly inputted in the fusion CNN model trained, last convolutional layer of this model Output be the Saliency maps of final prediction.
The significance detection method that region based on convolutional neural networks the most according to claim 1 and Pixel-level merge, It is characterized in that described n=300.
CN201610604732.0A 2016-07-28 2016-07-28 The conspicuousness detection method in region and Pixel-level fusion based on convolutional neural networks Active CN106157319B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610604732.0A CN106157319B (en) 2016-07-28 2016-07-28 The conspicuousness detection method in region and Pixel-level fusion based on convolutional neural networks

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610604732.0A CN106157319B (en) 2016-07-28 2016-07-28 The conspicuousness detection method in region and Pixel-level fusion based on convolutional neural networks

Publications (2)

Publication Number Publication Date
CN106157319A true CN106157319A (en) 2016-11-23
CN106157319B CN106157319B (en) 2018-11-02

Family

ID=58060262

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610604732.0A Active CN106157319B (en) 2016-07-28 2016-07-28 The conspicuousness detection method in region and Pixel-level fusion based on convolutional neural networks

Country Status (1)

Country Link
CN (1) CN106157319B (en)

Cited By (50)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106709532A (en) * 2017-01-25 2017-05-24 京东方科技集团股份有限公司 Image processing method and device
CN106911930A (en) * 2017-03-03 2017-06-30 深圳市唯特视科技有限公司 It is a kind of that the method for perceiving video reconstruction is compressed based on recursive convolution neutral net
CN106934397A (en) * 2017-03-13 2017-07-07 北京市商汤科技开发有限公司 Image processing method, device and electronic equipment
CN107016409A (en) * 2017-03-20 2017-08-04 华中科技大学 A kind of image classification method and system based on salient region of image
CN107169954A (en) * 2017-04-18 2017-09-15 华南理工大学 A kind of image significance detection method based on parallel-convolution neutral net
CN107194933A (en) * 2017-04-24 2017-09-22 天津大学 With reference to convolutional neural networks and the brain tumor dividing method and device of fuzzy reasoning
CN107368831A (en) * 2017-07-19 2017-11-21 中国人民解放军国防科学技术大学 English words and digit recognition method in a kind of natural scene image
CN107369160A (en) * 2017-06-28 2017-11-21 苏州比格威医疗科技有限公司 A kind of OCT image median nexus film new vessels partitioning algorithm
CN107437246A (en) * 2017-07-05 2017-12-05 浙江大学 A kind of common conspicuousness detection method based on end-to-end full convolutional neural networks
CN107506792A (en) * 2017-08-16 2017-12-22 上海荷福人工智能科技（集团）有限公司 A kind of semi-supervised notable method for checking object
CN107730546A (en) * 2017-08-25 2018-02-23 华北电力大学（保定） A kind of picture depth feature determines method and system
CN107766810A (en) * 2017-10-10 2018-03-06 湖南省测绘科技研究所 A kind of cloud, shadow detection method
CN107767383A (en) * 2017-11-01 2018-03-06 太原理工大学 A kind of Road image segmentation method based on super-pixel
CN107784308A (en) * 2017-10-09 2018-03-09 哈尔滨工业大学 Conspicuousness object detection method based on the multiple dimensioned full convolutional network of chain type
CN107886533A (en) * 2017-10-26 2018-04-06 深圳大学 Vision significance detection method, device, equipment and the storage medium of stereo-picture
CN107945109A (en) * 2017-11-06 2018-04-20 清华大学 Image split-joint method and device based on convolutional network
CN107945204A (en) * 2017-10-27 2018-04-20 西安电子科技大学 A kind of Pixel-level portrait based on generation confrontation network scratches drawing method
CN107967474A (en) * 2017-11-24 2018-04-27 上海海事大学 A kind of sea-surface target conspicuousness detection method based on convolutional neural networks
CN108345850A (en) * 2018-01-23 2018-07-31 哈尔滨工业大学 The scene text detection method of the territorial classification of stroke feature transformation and deep learning based on super-pixel
CN108389182A (en) * 2018-01-24 2018-08-10 北京卓视智通科技有限责任公司 A kind of picture quality detection method and device based on deep neural network
CN108960261A (en) * 2018-07-25 2018-12-07 扬州万方电子技术有限责任公司 A kind of obvious object detection method based on attention mechanism
CN108961220A (en) * 2018-06-14 2018-12-07 上海大学 A kind of image collaboration conspicuousness detection method based on multilayer convolution Fusion Features
CN109086777A (en) * 2018-07-09 2018-12-25 南京师范大学 A kind of notable figure fining method based on global pixel characteristic
WO2018233708A1 (en) * 2017-06-23 2018-12-27 华为技术有限公司 Method and device for detecting salient object in image
CN109389056A (en) * 2018-09-21 2019-02-26 北京航空航天大学 A kind of track surrounding enviroment detection method of space base multi-angle of view collaboration
CN109409222A (en) * 2018-09-20 2019-03-01 中国地质大学（武汉） A kind of multi-angle of view facial expression recognizing method based on mobile terminal
CN109409435A (en) * 2018-11-01 2019-03-01 上海大学 A kind of depth perception conspicuousness detection method based on convolutional neural networks
CN109934241A (en) * 2019-03-28 2019-06-25 南开大学 It can be integrated into Image Multiscale information extracting method and the application in neural network framework
CN110084221A (en) * 2019-05-08 2019-08-02 南京云智控产业技术研究院有限公司 A kind of serializing face critical point detection method of the tape relay supervision based on deep learning
CN110111295A (en) * 2018-02-01 2019-08-09 北京中科奥森数据科技有限公司 A kind of image collaboration conspicuousness detection method and device
CN110166850A (en) * 2019-05-30 2019-08-23 上海交通大学 The method and system of multiple CNN neural network forecast panoramic video viewing location
CN110222704A (en) * 2019-06-12 2019-09-10 北京邮电大学 A kind of Weakly supervised object detection method and device
CN110390327A (en) * 2019-06-25 2019-10-29 北京百度网讯科技有限公司 Foreground extracting method, device, computer equipment and storage medium
CN110472639A (en) * 2019-08-05 2019-11-19 山东工商学院 A kind of target extraction method based on conspicuousness prior information
CN110651300A (en) * 2017-07-14 2020-01-03 欧姆龙株式会社 Object detection device, object detection method, and program
CN110998611A (en) * 2017-08-17 2020-04-10 国际商业机器公司 Neuromorphic processing device
CN111260653A (en) * 2020-04-27 2020-06-09 腾讯科技（深圳）有限公司 Image segmentation method and device, storage medium and electronic equipment
CN111311532A (en) * 2020-03-26 2020-06-19 深圳市商汤科技有限公司 Image processing method and device, electronic device and storage medium
CN111583173A (en) * 2020-03-20 2020-08-25 北京交通大学 RGB-D image saliency target detection method
CN111598841A (en) * 2020-04-23 2020-08-28 南开大学 Example significance detection method based on regularized dense connection feature pyramid
CN111696021A (en) * 2020-06-10 2020-09-22 中国人民武装警察部队工程大学 Image self-adaptive steganalysis system and method based on significance detection
CN111915613A (en) * 2020-08-11 2020-11-10 华侨大学 Image instance segmentation method, device, equipment and storage medium
CN112149459A (en) * 2019-06-27 2020-12-29 哈尔滨工业大学（深圳） Video salient object detection model and system based on cross attention mechanism
CN112465700A (en) * 2020-11-26 2021-03-09 北京航空航天大学 Image splicing positioning device and method based on depth clustering
CN112541912A (en) * 2020-12-23 2021-03-23 中国矿业大学 Method and device for rapidly detecting saliency target in mine sudden disaster scene
CN113034365A (en) * 2021-03-19 2021-06-25 西安电子科技大学 Multi-image splicing method and system based on super pixels
US11164035B2 (en) 2018-10-31 2021-11-02 Abbyy Production Llc Neural-network-based optical character recognition using specialized confidence functions
CN114004775A (en) * 2021-11-30 2022-02-01 四川大学 Infrared and visible light image fusion method combining potential low-rank representation and convolutional neural network
CN114255351A (en) * 2022-02-28 2022-03-29 魔门塔(苏州)科技有限公司 Image processing method, device, medium, equipment and driving system
CN115965844A (en) * 2023-01-04 2023-04-14 哈尔滨工业大学 Multi-focus image fusion method based on visual saliency priori knowledge

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102222231A (en) * 2011-05-26 2011-10-19 厦门大学 Visual attention computational model based on guidance of dorsal pathway and processing method thereof
CN102567731A (en) * 2011-12-06 2012-07-11 北京航空航天大学 Extraction method for region of interest

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102222231A (en) * 2011-05-26 2011-10-19 厦门大学 Visual attention computational model based on guidance of dorsal pathway and processing method thereof
CN102567731A (en) * 2011-12-06 2012-07-11 北京航空航天大学 Extraction method for region of interest

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
XIANGQIAN WU ET.AL: "Offline Text-Independent Writer Identification Based on Scale Invariant Feature Transform", 《IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY》 *
YOUBAO TANG ET.AL: "Offline Signature Verification based on ASIFT", 《INTERNATIONAL CONFER-ENCE ON BIOMETRICS (ICB)》 *
YOUBAO TANG ET.AL: "Saliency Detection based on Graph-Structural Agglomerative Clustering", 《ACM INTERNATIONAL CONFERENCE ON MUL-TIMEDIA (ACMMM)》 *
李岳云 等: "深度卷积神经网络的显著性检测", 《中国图像图形学报》 *

Cited By (90)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106709532A (en) * 2017-01-25 2017-05-24 京东方科技集团股份有限公司 Image processing method and device
US10395167B2 (en) 2017-01-25 2019-08-27 Boe Technology Group Co., Ltd. Image processing method and device
CN106709532B (en) * 2017-01-25 2020-03-10 京东方科技集团股份有限公司 Image processing method and device
CN106911930A (en) * 2017-03-03 2017-06-30 深圳市唯特视科技有限公司 It is a kind of that the method for perceiving video reconstruction is compressed based on recursive convolution neutral net
CN106934397B (en) * 2017-03-13 2020-09-01 北京市商汤科技开发有限公司 Image processing method and device and electronic equipment
CN106934397A (en) * 2017-03-13 2017-07-07 北京市商汤科技开发有限公司 Image processing method, device and electronic equipment
US10943145B2 (en) 2017-03-13 2021-03-09 Beijing Sensetime Technology Development Co., Ltd. Image processing methods and apparatus, and electronic devices
WO2018166438A1 (en) * 2017-03-13 2018-09-20 北京市商汤科技开发有限公司 Image processing method and device and electronic device
CN107016409A (en) * 2017-03-20 2017-08-04 华中科技大学 A kind of image classification method and system based on salient region of image
CN107169954A (en) * 2017-04-18 2017-09-15 华南理工大学 A kind of image significance detection method based on parallel-convolution neutral net
CN107194933A (en) * 2017-04-24 2017-09-22 天津大学 With reference to convolutional neural networks and the brain tumor dividing method and device of fuzzy reasoning
WO2018233708A1 (en) * 2017-06-23 2018-12-27 华为技术有限公司 Method and device for detecting salient object in image
CN109118459A (en) * 2017-06-23 2019-01-01 南开大学 Image significance object detection method and device
US11430205B2 (en) 2017-06-23 2022-08-30 Huawei Technologies Co., Ltd. Method and apparatus for detecting salient object in image
CN107369160A (en) * 2017-06-28 2017-11-21 苏州比格威医疗科技有限公司 A kind of OCT image median nexus film new vessels partitioning algorithm
CN107369160B (en) * 2017-06-28 2020-04-03 苏州比格威医疗科技有限公司 Choroid neogenesis blood vessel segmentation algorithm in OCT image
CN107437246B (en) * 2017-07-05 2020-08-18 浙江大学 Common significance detection method based on end-to-end full-convolution neural network
CN107437246A (en) * 2017-07-05 2017-12-05 浙江大学 A kind of common conspicuousness detection method based on end-to-end full convolutional neural networks
CN110651300B (en) * 2017-07-14 2023-09-12 欧姆龙株式会社 Object detection device, object detection method, and program
CN110651300A (en) * 2017-07-14 2020-01-03 欧姆龙株式会社 Object detection device, object detection method, and program
CN107368831A (en) * 2017-07-19 2017-11-21 中国人民解放军国防科学技术大学 English words and digit recognition method in a kind of natural scene image
CN107368831B (en) * 2017-07-19 2019-08-02 中国人民解放军国防科学技术大学 English words and digit recognition method in a kind of natural scene image
CN107506792B (en) * 2017-08-16 2020-09-29 广西荷福智能科技有限公司 Semi-supervised salient object detection method
CN107506792A (en) * 2017-08-16 2017-12-22 上海荷福人工智能科技（集团）有限公司 A kind of semi-supervised notable method for checking object
CN110998611B (en) * 2017-08-17 2023-12-19 三星电子株式会社 Nerve morphology processing device
CN110998611A (en) * 2017-08-17 2020-04-10 国际商业机器公司 Neuromorphic processing device
CN107730546A (en) * 2017-08-25 2018-02-23 华北电力大学（保定） A kind of picture depth feature determines method and system
CN107730546B (en) * 2017-08-25 2020-11-03 华北电力大学（保定） Image depth feature determination method and system
CN107784308B (en) * 2017-10-09 2020-04-03 哈尔滨工业大学 Saliency target detection method based on chain type multi-scale full-convolution network
CN107784308A (en) * 2017-10-09 2018-03-09 哈尔滨工业大学 Conspicuousness object detection method based on the multiple dimensioned full convolutional network of chain type
CN107766810A (en) * 2017-10-10 2018-03-06 湖南省测绘科技研究所 A kind of cloud, shadow detection method
CN107766810B (en) * 2017-10-10 2021-05-14 湖南省测绘科技研究所 Cloud and shadow detection method
CN107886533A (en) * 2017-10-26 2018-04-06 深圳大学 Vision significance detection method, device, equipment and the storage medium of stereo-picture
CN107945204B (en) * 2017-10-27 2021-06-25 西安电子科技大学 Pixel-level image matting method based on generation countermeasure network
CN107945204A (en) * 2017-10-27 2018-04-20 西安电子科技大学 A kind of Pixel-level portrait based on generation confrontation network scratches drawing method
CN107767383A (en) * 2017-11-01 2018-03-06 太原理工大学 A kind of Road image segmentation method based on super-pixel
CN107767383B (en) * 2017-11-01 2021-05-11 太原理工大学 Road image segmentation method based on superpixels
CN107945109B (en) * 2017-11-06 2020-07-28 清华大学 Image splicing method and device based on convolutional network
CN107945109A (en) * 2017-11-06 2018-04-20 清华大学 Image split-joint method and device based on convolutional network
CN107967474A (en) * 2017-11-24 2018-04-27 上海海事大学 A kind of sea-surface target conspicuousness detection method based on convolutional neural networks
CN108345850B (en) * 2018-01-23 2021-06-01 哈尔滨工业大学 Scene text detection method based on region classification of stroke feature transformation and deep learning of superpixel
CN108345850A (en) * 2018-01-23 2018-07-31 哈尔滨工业大学 The scene text detection method of the territorial classification of stroke feature transformation and deep learning based on super-pixel
CN108389182B (en) * 2018-01-24 2020-07-17 北京卓视智通科技有限责任公司 Image quality detection method and device based on deep neural network
CN108389182A (en) * 2018-01-24 2018-08-10 北京卓视智通科技有限责任公司 A kind of picture quality detection method and device based on deep neural network
CN110111295B (en) * 2018-02-01 2021-06-11 北京中科奥森数据科技有限公司 Image collaborative saliency detection method and device
CN110111295A (en) * 2018-02-01 2019-08-09 北京中科奥森数据科技有限公司 A kind of image collaboration conspicuousness detection method and device
CN108961220A (en) * 2018-06-14 2018-12-07 上海大学 A kind of image collaboration conspicuousness detection method based on multilayer convolution Fusion Features
CN108961220B (en) * 2018-06-14 2022-07-12 上海大学 Image collaborative saliency detection method based on multilayer convolution feature fusion
CN109086777A (en) * 2018-07-09 2018-12-25 南京师范大学 A kind of notable figure fining method based on global pixel characteristic
CN109086777B (en) * 2018-07-09 2021-09-28 南京师范大学 Saliency map refining method based on global pixel characteristics
CN108960261B (en) * 2018-07-25 2021-09-24 扬州万方电子技术有限责任公司 Salient object detection method based on attention mechanism
CN108960261A (en) * 2018-07-25 2018-12-07 扬州万方电子技术有限责任公司 A kind of obvious object detection method based on attention mechanism
CN109409222A (en) * 2018-09-20 2019-03-01 中国地质大学（武汉） A kind of multi-angle of view facial expression recognizing method based on mobile terminal
CN109389056A (en) * 2018-09-21 2019-02-26 北京航空航天大学 A kind of track surrounding enviroment detection method of space base multi-angle of view collaboration
CN109389056B (en) * 2018-09-21 2020-05-26 北京航空航天大学 Space-based multi-view-angle collaborative track surrounding environment detection method
US11164035B2 (en) 2018-10-31 2021-11-02 Abbyy Production Llc Neural-network-based optical character recognition using specialized confidence functions
US11715288B2 (en) 2018-10-31 2023-08-01 Abbyy Development Inc. Optical character recognition using specialized confidence functions
CN109409435A (en) * 2018-11-01 2019-03-01 上海大学 A kind of depth perception conspicuousness detection method based on convolutional neural networks
CN109934241B (en) * 2019-03-28 2022-12-09 南开大学 Image multi-scale information extraction method capable of being integrated into neural network architecture
CN109934241A (en) * 2019-03-28 2019-06-25 南开大学 It can be integrated into Image Multiscale information extracting method and the application in neural network framework
CN110084221A (en) * 2019-05-08 2019-08-02 南京云智控产业技术研究院有限公司 A kind of serializing face critical point detection method of the tape relay supervision based on deep learning
CN110166850B (en) * 2019-05-30 2020-11-06 上海交通大学 Method and system for predicting panoramic video watching position by multiple CNN networks
CN110166850A (en) * 2019-05-30 2019-08-23 上海交通大学 The method and system of multiple CNN neural network forecast panoramic video viewing location
CN110222704A (en) * 2019-06-12 2019-09-10 北京邮电大学 A kind of Weakly supervised object detection method and device
CN110390327A (en) * 2019-06-25 2019-10-29 北京百度网讯科技有限公司 Foreground extracting method, device, computer equipment and storage medium
CN110390327B (en) * 2019-06-25 2022-06-28 北京百度网讯科技有限公司 Foreground extraction method and device, computer equipment and storage medium
CN112149459B (en) * 2019-06-27 2023-07-25 哈尔滨工业大学（深圳） Video saliency object detection model and system based on cross attention mechanism
CN112149459A (en) * 2019-06-27 2020-12-29 哈尔滨工业大学（深圳） Video salient object detection model and system based on cross attention mechanism
CN110472639A (en) * 2019-08-05 2019-11-19 山东工商学院 A kind of target extraction method based on conspicuousness prior information
CN110472639B (en) * 2019-08-05 2023-04-18 山东工商学院 Target extraction method based on significance prior information
CN111583173B (en) * 2020-03-20 2023-12-01 北京交通大学 RGB-D image saliency target detection method
CN111583173A (en) * 2020-03-20 2020-08-25 北京交通大学 RGB-D image saliency target detection method
CN111311532A (en) * 2020-03-26 2020-06-19 深圳市商汤科技有限公司 Image processing method and device, electronic device and storage medium
CN111598841A (en) * 2020-04-23 2020-08-28 南开大学 Example significance detection method based on regularized dense connection feature pyramid
CN111598841B (en) * 2020-04-23 2022-04-15 南开大学 Example significance detection method based on regularized dense connection feature pyramid
CN111260653A (en) * 2020-04-27 2020-06-09 腾讯科技（深圳）有限公司 Image segmentation method and device, storage medium and electronic equipment
CN111696021B (en) * 2020-06-10 2023-03-28 中国人民武装警察部队工程大学 Image self-adaptive steganalysis system and method based on significance detection
CN111696021A (en) * 2020-06-10 2020-09-22 中国人民武装警察部队工程大学 Image self-adaptive steganalysis system and method based on significance detection
CN111915613A (en) * 2020-08-11 2020-11-10 华侨大学 Image instance segmentation method, device, equipment and storage medium
CN111915613B (en) * 2020-08-11 2023-06-13 华侨大学 Image instance segmentation method, device, equipment and storage medium
CN112465700A (en) * 2020-11-26 2021-03-09 北京航空航天大学 Image splicing positioning device and method based on depth clustering
CN112465700B (en) * 2020-11-26 2022-04-26 北京航空航天大学 Image splicing positioning device and method based on depth clustering
CN112541912B (en) * 2020-12-23 2024-03-12 中国矿业大学 Rapid detection method and device for salient targets in mine sudden disaster scene
CN112541912A (en) * 2020-12-23 2021-03-23 中国矿业大学 Method and device for rapidly detecting saliency target in mine sudden disaster scene
CN113034365B (en) * 2021-03-19 2023-09-22 西安电子科技大学 Multi-picture splicing method and system based on super pixels
CN113034365A (en) * 2021-03-19 2021-06-25 西安电子科技大学 Multi-image splicing method and system based on super pixels
CN114004775A (en) * 2021-11-30 2022-02-01 四川大学 Infrared and visible light image fusion method combining potential low-rank representation and convolutional neural network
CN114255351A (en) * 2022-02-28 2022-03-29 魔门塔(苏州)科技有限公司 Image processing method, device, medium, equipment and driving system
CN115965844B (en) * 2023-01-04 2023-08-18 哈尔滨工业大学 Multi-focus image fusion method based on visual saliency priori knowledge
CN115965844A (en) * 2023-01-04 2023-04-14 哈尔滨工业大学 Multi-focus image fusion method based on visual saliency priori knowledge

Also Published As

Publication number Publication date
CN106157319B (en) 2018-11-02

Similar Documents

Publication Publication Date Title
CN106157319A (en) The significance detection method that region based on convolutional neural networks and Pixel-level merge
CN110321813B (en) Cross-domain pedestrian re-identification method based on pedestrian segmentation
CN107563422B (en) A kind of polarization SAR classification method based on semi-supervised convolutional neural networks
CN108090443B (en) Scene text detection method and system based on deep reinforcement learning
CN105551036B (en) A kind of training method and device of deep learning network
CN110210539B (en) RGB-T image saliency target detection method based on multi-level depth feature fusion
CN105205448B (en) Text region model training method and recognition methods based on deep learning
CN105139395B (en) SAR image segmentation method based on small echo pond convolutional neural networks
CN109902806A (en) Method is determined based on the noise image object boundary frame of convolutional neural networks
CN107480726A (en) A kind of Scene Semantics dividing method based on full convolution and shot and long term mnemon
CN107403430A (en) A kind of RGBD image, semantics dividing method
CN106803071A (en) Object detecting method and device in a kind of image
CN106920243A (en) The ceramic material part method for sequence image segmentation of improved full convolutional neural networks
CN106250931A (en) A kind of high-definition picture scene classification method based on random convolutional neural networks
CN107203781A (en) A kind of object detection method Weakly supervised end to end instructed based on conspicuousness
CN108447080A (en) Method for tracking target, system and storage medium based on individual-layer data association and convolutional neural networks
CN108921879A (en) The motion target tracking method and system of CNN and Kalman filter based on regional choice
CN104182772A (en) Gesture recognition method based on deep learning
CN107247952B (en) Deep supervision-based visual saliency detection method for cyclic convolution neural network
CN105787557A (en) Design method of deep nerve network structure for computer intelligent identification
CN108596240B (en) Image semantic segmentation method based on discriminant feature network
CN107633226A (en) A kind of human action Tracking Recognition method and system
CN111968127B (en) Cancer focus area identification method and system based on full-section pathological image
CN110334589A (en) A kind of action identification method of the high timing 3D neural network based on empty convolution
CN109858487A (en) Weakly supervised semantic segmentation method based on watershed algorithm and image category label

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant