WO2015169061A1 - 图像分割方法及装置 - Google Patents

图像分割方法及装置 Download PDF

Info

Publication number
WO2015169061A1
WO2015169061A1 PCT/CN2014/089297 CN2014089297W WO2015169061A1 WO 2015169061 A1 WO2015169061 A1 WO 2015169061A1 CN 2014089297 W CN2014089297 W CN 2014089297W WO 2015169061 A1 WO2015169061 A1 WO 2015169061A1
Authority
WO
WIPO (PCT)
Prior art keywords
pixel
background
image
foreground
model
Prior art date
Application number
PCT/CN2014/089297
Other languages
English (en)
French (fr)
Inventor
王琳
秦秋平
陈志军
Original Assignee
小米科技有限责任公司
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 小米科技有限责任公司 filed Critical 小米科技有限责任公司
Priority to RU2014153987/08A priority Critical patent/RU2596580C2/ru
Priority to BR112014032942A priority patent/BR112014032942A2/pt
Priority to JP2016517158A priority patent/JP6125099B2/ja
Priority to KR1020147036180A priority patent/KR101670004B1/ko
Priority to MX2014015363A priority patent/MX358601B/es
Priority to US14/583,816 priority patent/US9633444B2/en
Publication of WO2015169061A1 publication Critical patent/WO2015169061A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/162—Segmentation; Edge detection involving graph-based methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N5/00—Details of television systems
    • H04N5/222—Studio circuitry; Studio devices; Studio equipment
    • H04N5/262—Studio circuits, e.g. for mixing, switching-over, change of character of image, other special effects ; Cameras specially adapted for the electronic generation of special effects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • 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
    • G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N5/00—Details of television systems
    • H04N5/222—Studio circuitry; Studio devices; Studio equipment
    • H04N5/262—Studio circuits, e.g. for mixing, switching-over, change of character of image, other special effects ; Cameras specially adapted for the electronic generation of special effects
    • H04N5/2624—Studio circuits, e.g. for mixing, switching-over, change of character of image, other special effects ; Cameras specially adapted for the electronic generation of special effects for obtaining an image which is composed of whole input images, e.g. splitscreen
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N5/00—Details of television systems
    • H04N5/222—Studio circuitry; Studio devices; Studio equipment
    • H04N5/262—Studio circuits, e.g. for mixing, switching-over, change of character of image, other special effects ; Cameras specially adapted for the electronic generation of special effects
    • H04N5/2625—Studio circuits, e.g. for mixing, switching-over, change of character of image, other special effects ; Cameras specially adapted for the electronic generation of special effects for obtaining an image which is composed of images from a temporal image sequence, e.g. for a stroboscopic effect
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N5/00—Details of television systems
    • H04N5/222—Studio circuitry; Studio devices; Studio equipment
    • H04N5/262—Studio circuits, e.g. for mixing, switching-over, change of character of image, other special effects ; Cameras specially adapted for the electronic generation of special effects
    • H04N5/2625—Studio circuits, e.g. for mixing, switching-over, change of character of image, other special effects ; Cameras specially adapted for the electronic generation of special effects for obtaining an image which is composed of images from a temporal image sequence, e.g. for a stroboscopic effect
    • H04N5/2627—Studio circuits, e.g. for mixing, switching-over, change of character of image, other special effects ; Cameras specially adapted for the electronic generation of special effects for obtaining an image which is composed of images from a temporal image sequence, e.g. for a stroboscopic effect for providing spin image effect, 3D stop motion effect or temporal freeze effect
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N5/00—Details of television systems
    • H04N5/222—Studio circuitry; Studio devices; Studio equipment
    • H04N5/262—Studio circuits, e.g. for mixing, switching-over, change of character of image, other special effects ; Cameras specially adapted for the electronic generation of special effects
    • H04N5/272—Means for inserting a foreground image in a background image, i.e. inlay, outlay
    • 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
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • 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

Definitions

  • the present disclosure relates to the field of image processing, and in particular, to an image segmentation method and apparatus.
  • Image segmentation technology is the basis of image analysis, image editing and image synthesis. Image segmentation technology can segment foreground and background from image. How to quickly and automatically segment foreground and background from image is an important research. Question.
  • the foreground sample point and the background sample point in the image manually selected by the user are received; then, the front background color likelihood model is established according to the foreground sample point manually selected by the user; finally, according to the former
  • the background color likelihood model segmentes the image to obtain the segmented foreground and background.
  • the inventors have found that the related art has at least the following drawbacks: in the conventional image segmentation method, the user must manually select the foreground sample point and the background sample point roughly, when segmenting a large number of images, The segmentation efficiency is relatively low.
  • an image segmentation method including:
  • the image is segmented according to a predetermined graph cut algorithm that segments the image using the front background classification model and edge information between pixels.
  • the acquiring the foreground sample point and the background sample point in the image according to the significance model includes:
  • the predetermined foreground threshold is greater than the predetermined background threshold, and each of the normalized significant values is located in (0, 1).
  • the front background classification model includes a foreground classification model and a background classification model
  • the pre-background classification model is established according to the significance model and the foreground sample point and the background sample point, including:
  • the background classification model is used to characterize the probability that a pixel is a background.
  • the segmenting the image according to a predetermined graph cutting algorithm comprises:
  • the undirected graph is segmented by the predetermined segmentation algorithm to complete segmentation of the image.
  • the utilizing the foreground similarity of each pixel, the background similarity of each pixel, and the similarity between adjacent pixels construct an undirected graph required by the predetermined graph cutting algorithm, including:
  • the undirected graph including a foreground vertex, a background vertex, at least one pixel vertex, a first class edge between two adjacent pixel vertices, and the pixel vertex a second type of edge between the background vertex, a third type of edge between the pixel vertex and the background vertex, a pixel vertex in the undirected graph and each pixel in the image correspond;
  • a foreground similarity of a pixel point corresponding to a pixel vertex connected to the second class edge is determined as a weight of the second class edge
  • the background similarity of the pixel points corresponding to the pixel vertices connected to the third class edge is determined as the weight of the third class edge;
  • the similarity between two pixel points corresponding to two pixel vertices connected to the first class edge is determined as the weight of the first class edge.
  • the establishing a saliency model of the image includes:
  • the image is segmented by using a predetermined over-segmentation algorithm to obtain at least one region, and the color values of the respective pixels in the same region are the same;
  • the saliency model is established based on the color values corresponding to the respective regions and the centroids of the respective regions.
  • the significance model is:
  • S i1 is the significance value of any pixel in the region R i
  • w(R j ) is the number of pixel points in the region R j
  • D S (R i , R j ) is used to characterize the region A measure of the spatial position difference between R i and the region R j
  • D C (R i , R j ) is used to characterize a measure of the color difference between the region R i and the region R j
  • N is The total number of regions obtained by dividing the image
  • D S (R i , R j ) is: Center (R i ) is the centroid of the region R i
  • Center (R j ) is the centroid of the region R j .
  • the establishing a saliency model of the image includes:
  • each pixel in the image is classified, and pixels of the same color value are classified into the same color type;
  • the significance model is established based on the color values of each color type.
  • the significance model is:
  • w(P j ) is the number of pixel points in the color type P j
  • D C (P i , P j ) is used to characterize the metric value of the color difference between the color type P i and the color type P j .
  • an image segmentation apparatus comprising:
  • a first building module for establishing a saliency model of the image
  • a sample obtaining module configured to acquire a foreground sample point and a background sample point in the image according to the significance model
  • a second establishing module configured to establish a pre-background classification model according to the saliency model established by the first establishing module and the foreground sample point and the background sample point acquired by the sample acquiring module;
  • An image segmentation module configured to segment the image according to a predetermined graph cutting algorithm, wherein the predetermined graph cut algorithm performs the image by using a front background classification model established by the second establishing module and edge information between pixels segmentation.
  • the sample obtaining module includes:
  • a first calculating unit configured to calculate a saliency value of each pixel in the image according to the saliency model
  • a normalization unit configured to normalize the significance value of each pixel calculated by the calculation unit
  • a first determining unit configured to determine, by the normalized unit, a pixel value whose saliency value is greater than a predetermined foreground threshold as the foreground sample point;
  • a second determining unit configured to determine, by the normalized unit, a pixel value whose saliency value is less than a predetermined background threshold as the background sample point;
  • the predetermined foreground threshold is greater than the predetermined background threshold, and each of the normalized significant values is located in (0, 1).
  • the front background classification model includes a foreground classification model and a background classification model
  • the second establishing module includes:
  • a first establishing unit configured to establish a foreground color likelihood model according to the foreground sample point
  • a second establishing unit configured to establish a background color likelihood model according to the background sample points
  • a first multiplication unit configured to multiply a saliency model established by the first establishing module and a foreground color likelihood model established by the first establishing unit to obtain the foreground classification model, where the foreground classification model is used The probability of characterizing a pixel as a foreground;
  • a second multiplying unit configured to multiply a saliency model established by the first establishing module and a background color likelihood model established by the second establishing unit to obtain the background classification model, where the background classification model is used The probability of characterizing a pixel as a background.
  • the image segmentation module includes:
  • a second calculating unit configured to calculate a foreground similarity of each pixel in the image by using the foreground classification model
  • a third calculating unit configured to calculate, by using the background classification model, a background similarity of each pixel in the image
  • An acquiring unit configured to acquire a similarity between adjacent pixel points in the image
  • a constructing unit configured to construct an undirected graph required by the predetermined graph cutting algorithm by using foreground similarity of each pixel, background similarity of each pixel, and similarity between adjacent pixels;
  • a first dividing unit configured to divide the undirected graph by using the predetermined segmentation algorithm, and complete segmentation of the image.
  • the constructing unit includes:
  • the undirected graph comprising a foreground vertex, a background vertex, at least one pixel vertex, and a first class between two adjacent pixel vertices a second type of edge between the pixel vertex and the background vertex, a third class edge between the pixel vertex and the background vertex, a pixel vertex in the undirected graph and the image
  • the undirected graph comprising a foreground vertex, a background vertex, at least one pixel vertex, and a first class between two adjacent pixel vertices a second type of edge between the pixel vertex and the background vertex, a third class edge between the pixel vertex and the background vertex, a pixel vertex in the undirected graph and the image
  • the undirected graph comprising a foreground vertex, a background vertex, at least one pixel vertex, and a first class between two adjacent pixel vertices a second type of edge between the pixel vertex and the background vertex, a third class edge between the
  • a first determining subunit configured, for each second class edge, a foreground similarity of a pixel point corresponding to a pixel vertex connected to the second class edge, as a weight of the second class edge;
  • a second determining subunit configured, for each third class edge, a background similarity of a pixel point corresponding to a pixel vertex connected to the third class edge, as a weight of the third class edge;
  • a third determining subunit configured, for each first class edge, a similarity between two pixel points corresponding to two pixel vertices connected to the first class edge, as the first class The weight of the edge.
  • the first establishing module includes:
  • a second dividing unit configured to perform segmentation on the image by using a predetermined over-segmentation algorithm to obtain at least one region, where color values of respective pixels in the same region are the same;
  • a fourth determining unit configured to determine a color value and a centroid of each of the regions
  • a third establishing unit configured to establish the saliency model according to color values corresponding to the respective regions and a centroid of each region.
  • the significance model is:
  • S i1 is the significance value of any pixel in the region R i
  • w(R j ) is the number of pixel points in the region R j
  • D S (R i , R j ) is used to characterize the region A measure of the spatial position difference between R i and the region R j
  • D C (R i , R j ) is used to characterize a measure of the color difference between the region R i and the region R j
  • N is The total number of regions obtained by dividing the image
  • D S (R i , R j ) is: Center (R i ) is the centroid of the region R i
  • Center (R j ) is the centroid of the region R j .
  • the first establishing module includes:
  • a categorizing unit configured to classify each pixel in the image according to a color value of each pixel, and classify the pixel points of the same color value into the same color type
  • a fourth establishing unit configured to establish the saliency model according to color values of each color type.
  • the significance model is:
  • w(P j ) is the number of pixel points in the color type P j
  • D C (P i , P j ) is used to characterize the metric value of the color difference between the color type P i and the color type P j .
  • an image segmentation apparatus comprising:
  • a memory for storing the processor executable instructions
  • processor is configured to:
  • the image is segmented according to a predetermined graph cut algorithm that segments the image using the front background classification model and edge information between pixels.
  • the pre-background classification model is used to implement image segmentation; and the related technology must manually require the user to manually select the foreground.
  • Sample points and background sample points when segmenting a large number of images, the segmentation efficiency is relatively low; since the foreground sample points and automatic sample points can be automatically acquired, and the prior background categorization model is also combined with a priori saliency
  • the model achieves the effect of automating the selection of samples and improving the accuracy of classification.
  • FIG. 1 is a flowchart of an image segmentation method according to an exemplary embodiment
  • FIG. 2A is a flowchart of an image segmentation method according to another exemplary embodiment
  • 2B is a flow chart showing a saliency model for establishing an image, according to an exemplary embodiment
  • 2E is a schematic diagram of an undirected graph, according to an exemplary embodiment
  • FIG. 3 is a block diagram of an image segmentation apparatus according to an exemplary embodiment
  • FIG. 4 is a block diagram of an image segmentation apparatus according to another exemplary embodiment
  • FIG. 5 is a block diagram of an image segmentation apparatus according to a further exemplary embodiment.
  • the "electronic devices" mentioned in this article can be smart phones, tablets, smart TVs, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture) Experts Group Audio Layer IV, motion imaging experts compress standard audio layers 4) players, laptops and desktop computers, and more.
  • FIG. 1 is a flowchart of an image segmentation method according to an exemplary embodiment. As shown in FIG. 1 , the image segmentation method is applied to an electronic device, and includes the following steps.
  • step 101 a saliency model of the image is established.
  • step 102 foreground sample points and background sample points in the image are acquired according to the significance model.
  • a pre-background classification model is established based on the significance model and the foreground sample points and the background sample points.
  • step 104 the image is segmented according to a predetermined graph cut algorithm that uses the front background classification model and the edge information between the pixels to segment the image.
  • the image segmentation method automatically determines the foreground sample point and the background sample point, combines the saliency model and the front background sample point to establish a pre-background classification model, and uses the pre-background classification model to implement the image. Segmentation; solves the problem that the related art must manually select the foreground sample point and the background sample point manually, and the segmentation efficiency is relatively low when segmenting a large number of images; since the foreground sample point can be automatically obtained and automatically
  • the sample points, combined with the a priori saliency model when establishing the pre-background classification model, can achieve the effect of automatically selecting samples and improving the classification accuracy.
  • FIG. 2A is a flowchart of an image segmentation method according to another exemplary embodiment. As shown in FIG. 2A, the image segmentation method is applied to an electronic device, and includes the following steps.
  • step 201 a saliency model of the image is established.
  • the saliency model of the image can be established in various ways, as follows:
  • FIG. 2B is a flowchart of establishing a saliency model of an image according to an exemplary embodiment, including:
  • the image is divided by using a predetermined over-segmentation algorithm to obtain at least one region, and the color values of the respective pixels in the same region are the same;
  • Dividing an image is to divide the image into different regions, and the pixels in each region are identical in a certain characteristic. For example, the color values of the pixels in a certain region after being divided are the same. Or the color values of the individual pixels in a certain area after being divided are very close.
  • the over-segmentation algorithm used here is an over-segmentation algorithm based on Mean shift.
  • various over-segmentation algorithms can also be used, such as: watershed-based over-segmentation algorithm and super-based
  • the over-segmentation algorithm of the pixel clustering, etc. does not limit the over-segmentation algorithm in this embodiment.
  • the color value of the region can be determined, and for each region, the centroid corresponding to the region can also be calculated.
  • a saliency model is established according to the color values corresponding to the respective regions and the centroids of the respective regions.
  • the saliency model established by using steps 201a to 201c can be:
  • S i1 is the significance value of any pixel in the region R i
  • w(R j ) is the number of pixel points in the region R j
  • D S (R i , R j ) is used to represent the region R i
  • D C (R i , R j ) is used to represent a measure of the color difference between the region R i and the region R j
  • N is obtained by dividing the image.
  • the total number of regions, D S (R i , R j ) is: Center(R i ) is the centroid of the region R i , and Center(R j ) is the centroid of the region R j .
  • D C (R i , R j ) can be characterized by the Euclidean distance of the average color value of the region R i and the average color value of the region R j .
  • the average color value of the region is the color value of each pixel in the region divided by the number of pixels in the region. In an ideal case, the color values of the pixels in the region are the same, and the color of the region is The value is the color value of one of the pixels. In practical applications, the color values of the pixels in the same area are not completely the same. Generally, the color values of the respective pixels are relatively close. In this case, the color value of each pixel in the area may be divided by the area. The number of pixels in the middle, the average color value of the region is obtained.
  • the saliency model can be used to characterize the significance values of pixel points in each region affected by the remaining regions in the image.
  • FIG. 2C is a flowchart of establishing a saliency model of an image according to another exemplary embodiment, including:
  • a storage space (such as a storage queue or a storage stack) corresponding to a color value may be set for storing a pixel, and the number of the storage space may be 256*256*256, and the image is sequentially read.
  • the pixel points are placed in the storage space corresponding to the color value of the pixel point, so that the color values of the respective pixel points saved in each storage space are the same.
  • a saliency model is established based on the color values of each color type.
  • w(P j ) is the number of pixel points in the color type P j
  • D C (P i , P j ) is used to characterize the metric value of the color difference between the color type P i and the color type P j .
  • the number of pixels corresponding to the same color type may be very small, and the color of the pixels is opposite to other pixels.
  • the significance of the saliency value of the color is not significant. Therefore, in a possible implementation manner, in order to reduce the amount of calculation, a color type with more pixels can be selected to establish a saliency model.
  • step 202 a significance value for each pixel in the image is calculated based on the significance model.
  • step 203 the significance values of the individual pixel points are normalized.
  • the significance value of each pixel can usually be normalized to (0, 1).
  • step 204 a pixel point whose normalized significance value is greater than a predetermined foreground threshold is determined as a foreground sample point.
  • the predetermined foreground threshold may be set according to actual conditions, for example, the predetermined foreground threshold may be set to 0.8.
  • step 205 a pixel point whose normalized significance value is less than a predetermined background threshold is determined as a background sample point.
  • the predetermined foreground threshold may be set according to actual conditions, for example, the predetermined foreground threshold may be set to 0.25.
  • the predetermined foreground threshold is greater than a predetermined background threshold.
  • step 206 a foreground color likelihood model is established based on the foreground sample points.
  • a color likelihood model can be established by mathematical modeling based on histogram statistics, or a color likelihood model can be established based on a Gaussian Mixture Model. Color model. When the sample point used in establishing the color likelihood model is the foreground sample point, the obtained color likelihood model is determined as the foreground color likelihood model.
  • step 207 a background color likelihood model is established based on the background sample points.
  • the color likelihood model can be established by mathematical modeling based on histogram statistics, or the color likelihood model color model can be established based on the mixed Gaussian model.
  • the sample point used in establishing the color likelihood model is the background sample point
  • the obtained color likelihood model is determined as the background color likelihood model.
  • step 208 the saliency model is multiplied by the foreground color likelihood model to obtain a foreground classification model, and the foreground classification model is used to characterize the probability that the pixel is foreground.
  • a priori saliency model and an enhanced foreground likelihood model can be combined to obtain a foreground classification model.
  • the saliency model can be multiplied by the foreground color likelihood model to Get a foreground classification model.
  • step 209 the saliency model is multiplied by the background color likelihood model to obtain a background classification model, and the background classification model is used to characterize the probability that the pixel is the background.
  • a priori saliency model and an enhanced background likelihood model can be combined to obtain a background classification model.
  • the saliency model can be correlated with the background color likelihood model. Multiply to get the background classification model.
  • step 210 the foreground classification model is used to calculate the foreground similarity of each pixel in the image.
  • the foreground classification model is used to characterize the probability that the pixel is foreground, that is, the similarity between the pixel and the foreground, the foreground classification model can be directly used to calculate the foreground similarity of each pixel in the image.
  • the background classification model is used to calculate the background similarity of each pixel in the image.
  • the background classification model is used to represent the probability that the pixel is the background, that is, the similarity between the pixel and the background, the background classification model can be directly used to calculate the background similarity of each pixel in the image.
  • step 212 the similarity between adjacent pixel points in the image is acquired.
  • the undirected map required for the predetermined graph cut algorithm is constructed using the foreground similarity of each pixel point, the background similarity of each pixel point, and the similarity between adjacent pixel points.
  • FIG. 2D it is a flowchart for constructing an undirected graph, which utilizes foreground similarity of each pixel, background similarity of each pixel, and adjacent pixel points, according to an exemplary embodiment.
  • the degree of similarity, when constructing the undirected graph required by the predetermined graph cutting algorithm may include:
  • the undirected graph including the foreground vertex, the background vertex, the at least one pixel vertex, the first type of edge between the adjacent two pixel vertices, the pixel vertex and the background vertex
  • the second type of edge, the third type of edge between the pixel vertex and the background vertex, the pixel vertex in the undirected graph corresponds to each pixel point in the image;
  • the pixel vertices in the undirected graph are obtained by mapping each pixel in the image, that is, the pixels included in the image.
  • the number of pixels is the same as the number of pixel vertices in the constructed undirected graph, and each pixel point corresponds to one pixel vertex, and each pixel vertex corresponds to one pixel point.
  • FIG. 2E is a schematic diagram of an undirected graph including pixel vertices corresponding to pixel points in an image, which are shown here for simplicity, according to an exemplary embodiment. Only 9 pixel vertices are shown.
  • the undirected graph further includes a foreground vertex S and a background vertex T, wherein the pixel vertices are connected to form a first class edge s1, and the foreground vertex S is connected with any one pixel vertex to form a second class edge. S2, the background vertex T is connected with any one of the pixel vertices to form a third type of edge s3.
  • For each second type of edge determine a foreground similarity of a pixel corresponding to a pixel vertex connected to the second type of edge as a weight of the second type of edge;
  • a pixel corresponding to the pixel vertex may be determined, and the foreground similarity of the pixel is used as the weight of the second type of edge between the pixel vertex and the foreground vertex.
  • For each third class edge determine a background similarity of a pixel point corresponding to a pixel vertex connected to the third class edge as a weight of the third class edge;
  • a pixel corresponding to the pixel vertex may be determined, and the background similarity of the pixel is used as the weight of the third type of edge between the pixel vertex and the background vertex.
  • the similarity between two pixel points corresponding to two pixel vertices connected to the first class edge is determined as the weight of the first class edge.
  • step 214 the undirected graph is segmented using a predetermined segmentation algorithm to complete the segmentation of the image.
  • the predetermined segmentation algorithm may be a Graph cut algorithm, which may perform segmentation of the image by using the undirected graph constructed in step 213 above.
  • the method of segmenting an undirected graph by using the Graph cut algorithm can be implemented by those skilled in the art, and will not be described in detail herein.
  • the image segmentation method automatically determines the foreground sample point and the background sample point, combines the saliency model and the front background sample point to establish a pre-background classification model, and uses the pre-background classification model to implement the image. Segmentation; solves the problem that the related art must manually select the foreground sample point and the background sample point manually, and the segmentation efficiency is relatively low when segmenting a large number of images; since the foreground sample point and the automatic sample point can be automatically obtained, Moreover, a priori saliency model is also combined in the establishment of the pre-background classification model, which achieves the effect of automatically selecting samples and improving the classification accuracy.
  • FIG. 3 is a block diagram of an image segmentation apparatus, as shown in FIG. 3, which is applied to an electronic device.
  • the image segmentation device includes, but is not limited to, a first setup module 302, The sample acquisition module 304, the second establishment module 306, and the image segmentation module 308.
  • the first building module 302 is configured to establish a saliency model of the image.
  • the sample acquisition module 304 is configured to acquire foreground sample points and background sample points in the image based on the significance model.
  • the second establishing module 306 is configured to establish a pre-background classification model according to the saliency model established by the first establishing module and the foreground sample point and the background sample point acquired by the sample acquiring module.
  • the image segmentation module 308 is configured to segment the image according to a predetermined graph cut algorithm that uses the front background classification model established by the second building module and the edge information between the pixels to segment the image.
  • the image segmentation apparatus automatically determines the foreground sample point and the background sample point, combines the saliency model and the front background sample point to establish a pre-background classification model, and uses the pre-background classification model to implement the image. Segmentation; solves the problem that the related art must manually select the foreground sample point and the background sample point manually, and the segmentation efficiency is relatively low when segmenting a large number of images; since the foreground sample point and the automatic sample point can be automatically obtained, Moreover, a priori saliency model is also combined in the establishment of the pre-background classification model, which achieves the effect of automatically selecting samples and improving the classification accuracy.
  • FIG. 4 is a block diagram of an image segmentation apparatus, as shown in FIG. 4, which is applied to an electronic device, the image segmentation device including but not limited to: a first setup module 402, according to another exemplary embodiment. a sample acquisition module 404, a second setup module 406, and an image segmentation module 408.
  • the first building module 402 is configured to establish a saliency model of the image.
  • the sample acquisition module 404 is configured to acquire foreground sample points and background sample points in the image based on the significance model.
  • the second establishing module 406 is configured to establish a pre-background classification model according to the saliency model established by the first establishing module 402 and the foreground sample points and background sample points acquired by the sample obtaining module 404.
  • the image segmentation module 408 is configured to segment the image according to a predetermined graph cut algorithm that uses the front background classification model established by the second building module 406 and the edge information between the pixels to segment the image.
  • the sample obtaining module 404 may include: a first calculating unit 404a, a normalizing unit 404b, a first determining unit 404c, and a second determining unit 404d.
  • the first computing unit 404a is configured to calculate a significance value for each pixel in the image based on the significance model.
  • the normalization unit 404b is configured to normalize the significance values of the respective pixel points calculated by the first calculation unit 404a.
  • the first determining unit 404c is configured to determine a pixel point whose normalization value after the normalization unit 404b is greater than a predetermined foreground threshold as a foreground sample point.
  • the second determining unit 404d is configured to determine a pixel point whose normalization value after the normalization unit 404b is less than a predetermined background threshold as a background sample point.
  • the predetermined foreground threshold is greater than a predetermined background threshold, and each of the normalized significant values is located in (0, 1).
  • the front background classification model includes a foreground classification model and a background classification model
  • the second establishing module 406 may include: a first establishing unit 406a and a second establishing unit. 406b, a first multiplying unit 406c and a second multiplying unit 406d.
  • the first establishing unit 406a is configured to establish a foreground color likelihood model according to the foreground sample points
  • the second establishing unit 406b is configured to establish a background color likelihood model according to the background sample points;
  • the first multiplying unit 406c is configured to multiply the saliency model established by the first establishing module 402 and the foreground color likelihood model established by the first establishing unit 406a to obtain a foreground classification model, and the foreground classification model is used to represent the pixel points. Probability for the foreground;
  • the second multiplying unit 406d is configured to multiply the saliency model established by the first establishing module 402 and the background color likelihood model established by the second establishing unit 406b to obtain a background classification model, and the background classification model is used to represent the pixel points. The probability of being the background.
  • the image segmentation module 408 may include: a second calculation unit 408a, a third calculation unit 408b, an acquisition unit 408c, a construction unit 408d, and a first segmentation unit 408e. .
  • the second computing unit 408a is configured to calculate a foreground similarity of each pixel in the image using the foreground classification model
  • the third computing unit 408b is configured to calculate a background similarity of each pixel in the image using a background classification model
  • the obtaining unit 408c is configured to acquire a similarity between adjacent pixel points in the image
  • the constructing unit 408d is configured to construct an undirected graph required by the predetermined graph cutting algorithm by using the foreground similarity of each pixel point, the background similarity of each pixel point, and the similarity between adjacent pixel points;
  • the first segmentation unit 408e is configured to segment the undirected image by using a predetermined segmentation algorithm to complete segmentation of the image.
  • the constructing unit 408d includes: a constructing subunit 408d1, a first determining subunit 408d2, a second determining subunit 408d3, and a third determining subunit 408d4.
  • the construction sub-unit 408d1 is configured to construct an undirected map required by a predetermined graph cut algorithm, the undirected graph including a foreground vertex, a background vertex, at least one pixel vertex, a first type of edge between two adjacent pixel vertices, The second type of edge between the pixel vertex and the background vertex, the third type of edge between the pixel vertex and the background vertex, and the pixel vertex in the undirected graph corresponds to each pixel point in the image.
  • the first determining sub-unit 408d2 is configured to determine, for each second class edge, a foreground similarity of a pixel point corresponding to a pixel vertex connected to the second class edge as a weight of the second class edge;
  • the second determining sub-unit 408d3 is configured to determine, for each third class edge, a background similarity of a pixel point corresponding to a pixel vertex connected to the third class edge as a weight of the third class edge;
  • the third determining sub-unit 408d4 is configured to determine, for each first class edge, a similarity between two pixel points corresponding to two pixel vertices connected to the first class edge as the first class side Weight.
  • the first establishing module 402 may include: a second dividing unit 402a, a fourth determining unit 402b, and a third establishing unit 402c.
  • the second dividing unit 402a is configured to divide the image by using a predetermined over-segmentation algorithm to obtain at least one region, and the color values of the respective pixels in the same region are the same;
  • the fourth determining unit 402b is configured to determine a color value and a centroid of each region
  • the third establishing unit 402c is configured to be based on the color values corresponding to the respective regions and the centroids of the respective regions. Establish a significant model.
  • S i1 is the significance value of any pixel in the region R i
  • w(R j ) is the number of pixel points in the region R j
  • D S (R i , R j ) is used to represent the region R i
  • D C (R i , R j ) is used to represent a measure of the color difference between the region R i and the region R j
  • N is obtained by dividing the image.
  • the total number of regions, D S (R i , R j ) is: Center(R i ) is the centroid of the region R i , and Center(R j ) is the centroid of the region R j .
  • the first establishing module 402 may include: a categorizing unit 402d and a fourth establishing unit 402e.
  • the categorizing unit 402d is configured to classify each pixel in the image according to the color value of each pixel, and classify the pixel points of the same color value into the same color type;
  • the fourth establishing unit 402e is configured to establish a significance model based on color values of each color type.
  • the significance model is:
  • w(P j ) is the number of pixel points in the color type P j
  • D C (P i , P j ) is used to characterize the metric value of the color difference between the color type P i and the color type P j .
  • the image segmentation apparatus automatically determines the foreground sample point and the background sample point, combines the saliency model and the front background sample point to establish a pre-background classification model, and uses the pre-background classification model to implement the image. Segmentation; solves the problem that the related art must manually select the foreground sample point and the background sample point manually, and the segmentation efficiency is relatively low when segmenting a large number of images; since the foreground sample point and the automatic sample point can be automatically obtained, Moreover, a priori saliency model is also combined in the establishment of the pre-background classification model, which achieves the effect of automatically selecting samples and improving the classification accuracy.
  • FIG. 5 is a block diagram of an apparatus 500 for image segmentation, according to an exemplary embodiment.
  • device 500 can be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a gaming console, a tablet device, a medical device, a fitness device, a personal digital assistant, and the like.
  • apparatus 500 can include one or more of the following components: processing component 502, memory 504, power component 506, multimedia component 508, audio component 510, input/output (I/O) interface 512, sensor component 514, And a communication component 516.
  • Processing component 502 typically controls the overall operation of device 500, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations.
  • Processing component 502 can include one or more processors 518 to execute the fingers Order to complete all or part of the steps of the above method.
  • processing component 502 can include one or more modules to facilitate interaction between component 502 and other components.
  • processing component 502 can include a multimedia module to facilitate interaction between multimedia component 508 and processing component 502.
  • Memory 504 is configured to store various types of data to support operation at device 500. Examples of such data include instructions for any application or method operating on device 500, contact data, phone book data, messages, pictures, videos, and the like.
  • the memory 504 can be implemented by any type of volatile or non-volatile storage device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read only memory (EEPROM), erasable.
  • SRAM static random access memory
  • EEPROM electrically erasable programmable read only memory
  • EPROM Electrically erasable programmable read only memory
  • PROM Programmable Read Only Memory
  • ROM Read Only Memory
  • Magnetic Memory Flash Memory
  • Disk Disk or Optical Disk.
  • Power component 506 provides power to various components of device 500.
  • Power component 506 can include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power for device 500.
  • the multimedia component 508 includes a screen between the device 500 and the user that provides an output interface.
  • the screen can include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from the user.
  • the touch panel includes one or more touch sensors to sense touches, slides, and gestures on the touch panel. The touch sensor may sense not only the boundary of the touch or sliding action, but also the duration and pressure associated with the touch or slide operation.
  • the multimedia component 508 includes a front camera and/or a rear camera. When the device 500 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 and rear camera can be a fixed optical lens system or have focal length and optical zoom capabilities.
  • the audio component 510 is configured to output and/or input an audio signal.
  • audio component 510 includes a microphone (MIC) that is configured to receive an external audio signal when device 500 is in an operational mode, such as a call mode, a recording mode, and a voice recognition mode.
  • the received audio signal may be further stored in memory 504 or transmitted via communication component 516.
  • audio component 510 also includes a speaker for outputting an audio signal.
  • the I/O interface 512 provides an interface between the processing component 502 and the peripheral interface module, which may be a keyboard, a click wheel, a button, or the like. These buttons may include, but are not limited to, a home button, a volume button, a start button, and a lock button.
  • Sensor assembly 514 includes one or more sensors for providing device 500 with various aspects of status assessment.
  • sensor assembly 514 can detect an open/closed state of device 500, a relative positioning of components, such as the display and keypad of device 500, and sensor component 514 can also detect a change in position of one component of device 500 or device 500. The presence or absence of user contact with device 500, device 500 orientation or acceleration/deceleration, and temperature variation of device 500.
  • Sensor assembly 514 can include a proximity sensor configured to detect the presence of nearby objects without any physical contact.
  • Sensor assembly 514 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications.
  • the sensor component 514 can also include an acceleration sensor, a gyro sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.
  • Communication component 516 is configured to facilitate wired or wireless communication between device 500 and other devices.
  • the device 500 can access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof.
  • communication component 516 receives broadcast signals or broadcast associated information from an external broadcast management system via a broadcast channel.
  • the communication component 516 also includes a near field communication (NFC) module to facilitate short range communication.
  • NFC near field 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
  • apparatus 500 may be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable A gate array (FPGA), controller, microcontroller, microprocessor, or other electronic component implementation for performing the above methods.
  • ASICs application specific integrated circuits
  • DSPs digital signal processors
  • DSPDs digital signal processing devices
  • PLDs programmable logic devices
  • FPGA field programmable A gate array
  • controller microcontroller, microprocessor, or other electronic component implementation for performing the above methods.
  • non-transitory computer readable storage medium comprising instructions, such as a memory 504 comprising instructions executable by processor 518 of apparatus 500 to perform the above method.
  • the non-transitory computer readable storage medium may be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device.

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)
  • Character Input (AREA)
  • Image Processing (AREA)

Abstract

本公开揭示了一种图像分割方法及装置，属于图像处理领域。所述图像分割方法包括：建立图像的显著性模型；根据显著性模型获取图像中的前景样本点和背景样本点；根据显著性模型以及前景样本点和背景样本点，建立前背景分类模型；根据预定图割算法对图像进行分割，预定图割算法利用前背景分类模型以及像素点之间的边缘信息对图像进行分割。通过自动确定前背景样本点，并结合显著性模型以建立前背景分类模型，利用该前背景分类模型实现图像分割；解决了相关技术中必须需要用户手动粗略地选定前景样本点和背景样本点，在对大量图像进行分割时，分割效率比较低的问题；达到了可以实现自动化选取样本，提高了分类精确度的效果。

Description

图像分割方法及装置
本申请基于申请号为201410187226.7、申请日为2014年5月5日的中国专利申请提出，并要求该中国专利申请的优先权，该中国专利申请的全部内容在此引入本申请作为参考。
技术领域
本公开涉及图像处理领域，特别涉及一种图像分割方法及装置。
背景技术
图像分割技术是图像分析、图像编辑和图像合成等领域的基础，图像分割技术可以从图像中分割出前景和背景，如何快速的、自动的从图像中分割出前景和背景是目前研究的一个重要课题。
在相关的图像分割方法中，首先，接收用户手动选定的图像中的前景样本点和背景样本点；然后，根据用户手动选定的前景样本点建立前背景颜色似然模型；最后，根据前背景颜色似然模型对图像进行分割，得到分割后的前景和背景。
发明人在实现本公开的过程中，发现相关技术至少存在如下缺陷：在传统的图像分割方法中，必须需要用户手动粗略地选定前景样本点和背景样本点，在对大量图像进行分割时，分割效率比较低。
发明内容
为了解决相关技术中必须需要用户手动粗略地选定前景样本点和背景样本点，在对大量图像进行分割时，分割效率比较低的问题，本公开提供一种图像分割方法及装置。所述技术方案如下：
根据本公开实施例的第一方面，提供一种图像分割方法，包括：
建立图像的显著性模型；
根据所述显著性模型获取所述图像中的前景样本点和背景样本点；
根据所述显著性模型以及所述前景样本点和所述背景样本点，建立前背景分类模型；
根据预定图割算法对所述图像进行分割，所述预定图割算法利用所述前背景分类模型以及像素点之间的边缘信息对所述图像进行分割。
可选的，所述根据所述显著性模型获取所述图像中的前景样本点和背景样本点，包括：
根据所述显著性模型，计算所述图像中各个像素点的显著性值；
将各个像素点的显著性值进行归一化；
将归一化后的显著性值大于预定前景阈值的像素点确定为所述前景样本点；
将归一化后的显著性值小于预定背景阈值的像素点确定为所述背景样本点；
其中，所述预定前景阈值大于所述预定背景阈值，归一化后的各个显著值均位于(0，1)中。
可选的，所述前背景分类模型包括前景分类模型和背景分类模型，所述根据所述显著性模型以及所述前景样本点和所述背景样本点，建立前背景分类模型，包括：
根据所述前景样本点建立前景颜色似然模型；
根据所述背景样本点建立背景颜色似然模型；
将所述显著性模型与所述前景颜色似然模型相乘，得到所述前景分类模型，所述前景分类模型用于表征像素点为前景的概率；
将所述显著性模型与所述背景颜色似然模型相乘，得到所述背景分类模型，所述背景分类模型用于表征像素点为背景的概率。
可选的，所述根据预定图割算法对所述图像进行分割，包括：
利用所述前景分类模型计算所述图像中每个像素点的前景相似度；
利用所述背景分类模型计算所述图像中每个像素点的背景相似度；
获取所述图像中相邻像素点之间的相似度；
利用各个像素点的前景相似度、各个像素点的背景相似度以及相邻像素点之间的相似度，构造所述预定图割算法所需的无向图；
利用所述预定分割算法对所述无向图进行分割，完成对所述图像的分割。
可选的，所述利用各个像素点的前景相似度、各个像素点的背景相似度以及相邻像素点之间的相似度，构造所述预定图割算法所需的无向图，包括：
构建所述预定图割算法所需的无向图，所述无向图包括前景顶点、背景顶点、至少一个像素顶点、相邻的两个像素顶点之间的第一类边、所述像素顶点与所述背景顶点之间的第二类边，所述像素顶点与所述背景顶点之间的第三类边，所述无向图中的像素顶点与所述图像中的各个像素点一一对应；
对于每条第二类边，将与所述第二类边相连的像素顶点所对应的像素点的前景相似度，确定为所述第二类边的权值；
对于每条第三类边，将与所述第三类边相连的像素顶点所对应的像素点的背景相似度，确定为所述第三类边的权值；
对于每条第一类边，将与所述第一类边相连的两个像素顶点所对应的两个像素点之间的相似度，确定为所述第一类边的权值。
可选的，所述建立图像的显著性模型，包括：
利用预定过分割算法对所述图像进行过分割，得到至少一个区域，同一个所述区域中各个像素点的颜色值相同；
确定每个所述区域的颜色值和质心；
根据各个区域所对应的颜色值以及各个区域的质心，建立所述显著性模型。
可选的，所述显著性模型为：
Figure PCTCN2014089297-appb-000001
其中，Si1为区域Ri中任一像素点的显著性值，w(Rj)为区域Rj中的像素点的个数，DS(Ri,Rj)用于表征所述区域Ri和所述区域Rj之间空间位置差异的度量值，DC(Ri,Rj)用于表征所述区域Ri和所述区域Rj之间颜色差异的度量值，N为对所述图像进行过分割后得到的区域的总个数，DS(Ri,Rj)为：
Figure PCTCN2014089297-appb-000002
Center(Ri)为所述区域Ri的质心，Center(Rj)为所述区域Rj的质心，当所述图像中各个像素点的坐标均归一化到[0,1]时，
Figure PCTCN2014089297-appb-000003
可选的，所述建立图像的显著性模型，包括：
按照各个像素点的颜色值，对所述图像中各个像素点进行归类，将相同颜色值的像素点归类为同一种颜色类型；
根据每种颜色类型的颜色值，建立所述显著性模型。
可选的，所述显著性模型为：
Figure PCTCN2014089297-appb-000004
其中，w(Pj)为颜色类型Pj中像素点的个数，DC(Pi,Pj)用于表征颜色类型Pi和颜色类型Pj之间颜色差异的度量值。
根据本公开实施例的第二方面，提供一种图像分割装置，包括：
第一建立模块，用于建立图像的显著性模型；
样本获取模块，用于根据所述显著性模型获取所述图像中的前景样本点和背景样本点；
第二建立模块，用于根据所述第一建立模块建立的显著性模型以及所述样本获取模块获取的前景样本点和所述背景样本点，建立前背景分类模型；
图像分割模块，用于根据预定图割算法对所述图像进行分割，所述预定图割算法利用所述第二建立模块建立的前背景分类模型以及像素点之间的边缘信息对所述图像进行分割。
可选的，所述样本获取模块，包括：
第一计算单元，用于根据所述显著性模型，计算所述图像中各个像素点的显著性值；
归一化单元，用于将所述计算单元计算出的各个像素点的显著性值进行归一化；
第一确定单元，用于将所述归一化单元归一化后的显著性值大于预定前景阈值的像素点确定为所述前景样本点；
第二确定单元，用于将所述归一化单元归一化后的显著性值小于预定背景阈值的像素点确定为所述背景样本点；
其中，所述预定前景阈值大于所述预定背景阈值，归一化后的各个显著值均位于(0，1)中。
可选的，所述前背景分类模型包括前景分类模型和背景分类模型，所述第二建立模块，包括：
第一建立单元，用于根据所述前景样本点建立前景颜色似然模型；
第二建立单元，用于根据所述背景样本点建立背景颜色似然模型；
第一相乘单元，用于将所述第一建立模块建立的显著性模型与所述第一建立单元建立的前景颜色似然模型相乘，得到所述前景分类模型，所述前景分类模型用于表征像素点为前景的概率；
第二相乘单元，用于将所述第一建立模块建立的显著性模型与所述第二建立单元建立的背景颜色似然模型相乘，得到所述背景分类模型，所述背景分类模型用于表征像素点为背景的概率。
可选的，所述图像分割模块，包括：
第二计算单元，用于利用所述前景分类模型计算所述图像中每个像素点的前景相似度；
第三计算单元，用于利用所述背景分类模型计算所述图像中每个像素点的背景相似度；
获取单元，用于获取所述图像中相邻像素点之间的相似度；
构造单元，用于利用各个像素点的前景相似度、各个像素点的背景相似度以及相邻像素点之间的相似度，构造所述预定图割算法所需的无向图；
第一分割单元，用于利用所述预定分割算法对所述无向图进行分割，完成对所述图像的分割。
可选的，所述构造单元，包括：
构建子单元，用于构建所述预定图割算法所需的无向图，所述无向图包括前景顶点、背景顶点、至少一个像素顶点、相邻的两个像素顶点之间的第一类边、所述像素顶点与所述背景顶点之间的第二类边，所述像素顶点与所述背景顶点之间的第三类边，所述无向图中的像素顶点与所述图像中的各个像素点一一对应
第一确定子单元，用于对于每条第二类边，将与所述第二类边相连的像素顶点所对应的像素点的前景相似度，确定为所述第二类边的权值；
第二确定子单元，用于对于每条第三类边，将与所述第三类边相连的像素顶点所对应的像素点的背景相似度，确定为所述第三类边的权值；
第三确定子单元，用于对于每条第一类边，将与所述第一类边相连的两个像素顶点所对应的两个像素点之间的相似度，确定为所述第一类边的权值。
可选的，所述第一建立模块，包括：
第二分割单元，用于利用预定过分割算法对所述图像进行过分割，得到至少一个区域，同一个所述区域中各个像素点的颜色值相同；
第四确定单元，用于确定每个所述区域的颜色值和质心；
第三建立单元，用于根据各个区域所对应的颜色值以及各个区域的质心，建立所述显著性模型。
可选的，所述显著性模型为：
Figure PCTCN2014089297-appb-000005
其中，Si1为区域Ri中任一像素点的显著性值，w(Rj)为区域Rj中的像素点的个数，DS(Ri,Rj)用于表征所述区域Ri和所述区域Rj之间空间位置差异的度量值，DC(Ri,Rj)用于表征所述区域Ri和所述区域Rj之间颜色差异的度量值，N为对所述图像进行过分割后得到的区域的总个数，DS(Ri,Rj)为：
Figure PCTCN2014089297-appb-000006
Center(Ri)为所述区域Ri的质心，Center(Rj)为所述区域Rj的质心，当所述图像中各个像素点的坐标均归一化到[0,1]时，
Figure PCTCN2014089297-appb-000007
可选的，所述第一建立模块，包括：
归类单元，用于按照各个像素点的颜色值，对所述图像中各个像素点进行归类，将相同颜色值的像素点归类为同一种颜色类型；
第四建立单元，用于根据每种颜色类型的颜色值，建立所述显著性模型。
可选的，所述显著性模型为：
Figure PCTCN2014089297-appb-000008
其中，w(Pj)为颜色类型Pj中像素点的个数，DC(Pi,Pj)用于表征颜色类型Pi和颜色类型Pj之间颜色差异的度量值。
根据本公开实施例的第三方面，提供一种图像分割装置，包括：
处理器；
用于存储所述处理器可执行指令的存储器；
其中，所述处理器被配置为：
建立图像的显著性模型；
根据所述显著性模型获取所述图像中的前景样本点和背景样本点；
根据所述显著性模型以及所述前景样本点和所述背景样本点，建立前背景分类模型；
根据预定图割算法对所述图像进行分割，所述预定图割算法利用所述前背景分类模型以及像素点之间的边缘信息对所述图像进行分割。
本公开的实施例提供的技术方案可以包括以下有益效果：
通过自动确定前景样本点和背景样本点，结合显著性模型以及前背景样本点建立前背景分类模型，利用该前背景分类模型实现图像分割；解决了相关技术中必须需要用户手动粗略地选定前景样本点和背景样本点，在对大量图像进行分割时，分割效率比较低的问题；由于可以自动获取前景样本点和自动样本点，且在建立前背景分类模型时还结合了先验的显著性模型，达到了可以实现自动化选取样本，提高了分类精确度的效果。
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的，并不能限制本公开。
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，示出了符合本发明的实施例，并于说明书一起用于解释本发明的原理。
图1是根据一示例性实施例示出的一种图像分割方法的流程图；
图2A是根据另一示例性实施例示出的一种图像分割方法的流程图；
图2B是根据一示例性实施例示出的一种建立图像的显著性模型的流程图；
图2C是根据另一示例性实施例示出的一种建立图像的显著性模型的流程图；
图2D是根据一示例性实施例示出的一种构造无向图的流程图；
图2E是根据一示例性实施例示出的一种无向图的示意图；
图3是根据一示例性实施例示出的一种图像分割装置的框图；
图4是根据另一示例性实施例示出的一种图像分割装置的框图；
图5是根据再示例性实施例示出的一种图像分割装置的框图。
具体实施方式
这里将详细地对示例性实施例进行说明，其示例表示在附图中。下面的描述涉及附图时，除非另有表示，不同附图中的相同数字表示相同或相似的要素。以下示例性实施例中所描述的实施方式并不代表与本发明相一致的所有实施方式。相反，它们仅是与如所附权利要求书中所详述的、本发明的一些方面相一致的装置和方法的例子。
文中所讲的“电子设备”可以是智能手机、平板电脑、智能电视、电子书阅读器、MP3播放器(Moving Picture Experts Group Audio Layer III，动态影像专家压缩标准音频层面3)、MP4(Moving Picture Experts Group Audio Layer IV，动态影像专家压缩标准音频层面4)播放器、膝上型便携计算机和台式计算机等等。
图1是根据一示例性实施例示出的一种图像分割方法的流程图，如图1所示，该图像分割方法应用于电子设备中，包括以下步骤。
在步骤101中，建立图像的显著性模型。
在步骤102中，根据显著性模型获取图像中的前景样本点和背景样本点。
在步骤103中，根据显著性模型以及前景样本点和背景样本点，建立前背景分类模型。
在步骤104中，根据预定图割算法对图像进行分割，预定图割算法利用前背景分类模型以及像素点之间的边缘信息对图像进行分割。
综上所述，本公开实施例中提供的图像分割方法，通过自动确定前景样本点和背景样本点，结合显著性模型以及前背景样本点建立前背景分类模型，利用该前背景分类模型实现图像分割；解决了相关技术中必须需要用户手动粗略地选定前景样本点和背景样本点，在对大量图像进行分割时，分割效率比较低的问题；由于可以自动获取前景样本点和自动 样本点，且在建立前背景分类模型时还结合了先验的显著性模型，达到了可以实现自动化选取样本，提高了分类精确度的效果。
图2A是根据另一示例性实施例示出的一种图像分割方法的流程图，如图2A所示，该图像分割方法应用于电子设备中，包括以下步骤。
在步骤201中，建立图像的显著性模型。
在实际应用中，可以通过多种方式建立图像的显著性模型，具体如下：
在第一种方式下，请参见图2B所示，其是根据一示例性实施例示出的一种建立图像的显著性模型的流程图，包括：
201a，利用预定过分割算法对图像进行过分割，得到至少一个区域，同一个区域中各个像素点的颜色值相同；
对图像进行过分割即是将图像分割成不同的区域，每个区域中的像素点在某一个特性上是相同的，比如被过分割后的某一个区域中的各个像素点的颜色值相同，或者被过分割后的某一个区域中的各个像素点的颜色值非常接近。
这里采用的过分割算法是基于均值飘移(Mean shift)的过分割算法，在实际应用中，还可以采用其他各种过分割算法，比如可以包括：基于分水岭(watershed)的过分割算法和基于超像素聚类的过分割算法等，本实施例并不对过分割算法进行限定。
201b，确定每个区域的颜色值和质心；
由于过分割之后的区域中各个像素点具有相同的颜色值，因此可以确定出该区域的颜色值，且针对每个区域，也可以计算出区域所对应的质心。
201c，根据各个区域所对应的颜色值以及各个区域的质心，建立显著性模型。
利用步骤201a至201c所建立的显著性模型可以为：
Figure PCTCN2014089297-appb-000009
其中，Si1为区域Ri中任一像素点的显著性值，w(Rj)为区域Rj中的像素点的个数，DS(Ri,Rj)用于表征区域Ri和区域Rj之间空间位置差异的度量值，DC(Ri,Rj)用于表征区域Ri和区域Rj之间颜色差异的度量值，N为对图像进行过分割后得到的区域的总个数，DS(Ri,Rj)为：
Figure PCTCN2014089297-appb-000010
Center(Ri)为区域Ri的质心，Center(Rj)为区域Rj的质心，当图像中各个像素点的坐标均归一化到[0,1]时，
Figure PCTCN2014089297-appb-000011
DC(Ri,Rj)可以用区域Ri的平均颜色值和区域Rj的平均颜色值的欧氏距离来表征。区域的平均颜色值即为该区域中各个像素点的颜色值之后除以该区域中像素点的个数，在理想情况下，区域中各个像素点的颜色值均相同，此时该区域的颜色值即为其中一个像素点的颜色值。而在实际应用中，同一个区域中的各个像素点的颜色值并不是完全相同，通常各个像素点的颜色值比较接近，此时则可以该区域中各个像素点的颜色值之后除以该区域中像素点的个数，得到该区域的平均颜色值。
由该显著性模型的构成可知，该显著性模型可以用于表征每个区域中的像素点的显著性值受到图像中其余各个区域的影响。
在第二种方式下，请参见图2C所示，其是根据另一示例性实施例示出的一种建立图像的显著性模型的流程图，包括：
201d，按照各个像素点的颜色值，对图像中各个像素点进行归类，将相同颜色值的像素点归类为同一种颜色类型；
在实际应用中，可以设置用于存储像素点的与颜色值对应的存储空间(比如存储队列或存储栈等)，存储空间的个数通常可以为256*256*256个，依次读取图像中的像素点，将该像素点放入与该像素点的颜色值对应的存储空间中，这样每个存储空间中所保存的各个像素点的颜色值均相同。
当读取完该图像中的各个像素点之后，统计每个存储空间中包存储的像素点的个数。
201e，根据每种颜色类型的颜色值，建立显著性模型。
根据每种颜色类型的颜色值，建立得到的显著性模型为：
Figure PCTCN2014089297-appb-000012
其中，w(Pj)为颜色类型Pj中像素点的个数，DC(Pi,Pj)用于表征颜色类型Pi和颜色类型Pj之间颜色差异的度量值。
需要说明的是，在实际应用中，通过步骤201d对图像中的像素点进行分类之后，同一种颜色类型所对应的像素点的个数可能会非常少，这些像素点的颜色对其他像素点的颜色的显著性值影响并不大，因此在一种可能的实现方式中，为了减少计算量，则可以选择像素点较多的颜色类型，建立显著性模型。
在步骤202中，根据显著性模型，计算图像中各个像素点的显著性值。
在步骤203中，将各个像素点的显著性值进行归一化。
通常可以将各个像素点的显著性值归一化到(0，1)中。
在步骤204中，将归一化后的显著性值大于预定前景阈值的像素点确定为前景样本点。
当将各个像素点的显著性值归一化到(0，1)中时，预定前景阈值可以根据实际情况设定，比如该预定前景阈值可以设置为0.8。
在步骤205中，将归一化后的显著性值小于预定背景阈值的像素点确定为背景样本点。
当将各个像素点的显著性值归一化到(0，1)中时，预定前景阈值可以根据实际情况设定，比如该预定前景阈值可以设置为0.25。
通常来讲，预定前景阈值大于预定背景阈值。
这样以来，就可以根据建立的显著性模型自动确定出前景样本点和背景样本点。
在步骤206中，根据前景样本点建立前景颜色似然模型。
在实际应用中，建立颜色似然模型的方式比较多，比如可以通过基于直方图统计的数学建模方式建立颜色似然模型，也可以通过基于混合高斯模型(Gaussian Mixture Model)建立颜色似然模型颜色模型。当建立颜色似然模型时所使用的样本点为前景样本点时，得到的颜色似然模型则确定为前景颜色似然模型。
在步骤207中，根据背景样本点建立背景颜色似然模型。
同理，可以通过基于直方图统计的数学建模方式建立颜色似然模型，也可以通过基于混合高斯模型建立颜色似然模型颜色模型。当建立颜色似然模型时所使用的样本点为背景样本点时，得到的颜色似然模型则确定为背景颜色似然模型。
在步骤208中，将显著性模型与前景颜色似然模型相乘，得到前景分类模型，前景分类模型用于表征像素点为前景的概率。
为了加强对图像的前景分割时的精确度，可以结合先验的显著性模型以及增强型的前景似然模型，获取前景分类模型，比如可以将显著性模型与前景颜色似然模型相乘，以得到前景分类模型。
在步骤209中，将显著性模型与背景颜色似然模型相乘，得到背景分类模型，背景分类模型用于表征像素点为背景的概率。
同理，为了加强对图像的背景分割时的精确度，可以结合先验的显著性模型以及增强型的背景似然模型，获取背景分类模型，比如可以将显著性模型与背景颜色似然模型相乘，以得到背景分类模型。
在步骤210中，利用前景分类模型计算图像中每个像素点的前景相似度。
由于前景分类模型时用于表征像素点为前景的概率，也即该像素点与前景的相似度，因此可以直接利用前景分类模型计算图像中每个像素点的前景相似度。
在步骤211中，利用背景分类模型计算图像中每个像素点的背景相似度。
同理，由于背景分类模型时用于表征像素点为背景的概率，也即该像素点与背景的相似度，因此可以直接利用背景分类模型计算图像中每个像素点的背景相似度。
在步骤212中，获取图像中相邻像素点之间的相似度。
在步骤213中，利用各个像素点的前景相似度、各个像素点的背景相似度以及相邻像素点之间的相似度，构造预定图割算法所需的无向图。
请参见图2D所示，其是根据一示例性实施例示出的一种构造无向图的流程图，在利用各个像素点的前景相似度、各个像素点的背景相似度以及相邻像素点之间的相似度，构造预定图割算法所需的无向图时，可以包括：
213a，构建预定图割算法所需的无向图，无向图包括前景顶点、背景顶点、至少一个像素顶点、相邻的两个像素顶点之间的第一类边、像素顶点与背景顶点之间的第二类边，像素顶点与背景顶点之间的第三类边，无向图中的像素顶点与图像中的各个像素点一一对应；
无向图中的像素顶点是图像中各个像素点进行映射得到的，也即图像中包含的像素点 的个数与构建的无向图中的像素顶点的个数相同，且每一个像素点对应一个像素顶点，每一个像素顶点对应一个像素点。
请参见图2E所示，其是根据一示例性实施例示出的一种无向图的示意图，该无向图包括像素顶点，这些像素顶点与图像中的像素点一一对应，为了简化，这里仅示出了9个像素顶点，该无向图还包括前景顶点S和背景顶点T，其中像素顶点之间连接形成第一类边s1，前景顶点S与任意一个像素顶点连接形成第二类边s2，背景顶点T与任意一个像素顶点连接形成第三类边s3。
213b，对于每条第二类边，将与第二类边相连的像素顶点所对应的像素点的前景相似度，确定为第二类边的权值；
比如，对于一个选定的像素顶点，可以确定与该像素顶点对应的像素点，将该像素点的前景相似度作为该像素顶点与前景顶点之间第二类边的权值。
213c，对于每条第三类边，将与第三类边相连的像素顶点所对应的像素点的背景相似度，确定为第三类边的权值；
比如，对于一个选定的像素顶点，可以确定与该像素顶点对应的像素点，将该像素点的背景相似度作为该像素顶点与背景顶点之间第三类边的权值。
213d，对于每条第一类边，将与第一类边相连的两个像素顶点所对应的两个像素点之间的相似度，确定为第一类边的权值。
在步骤214中，利用预定分割算法对无向图进行分割，完成对图像的分割。
预定分割算法可以为Graph cut(图割)算法，该算法可以利用上述经过步骤213构造的无向图完成对图像的分割。利用Graph cut算法对无向图进行分割的方法是本领域所属技术人员都能够实现的，这里就不再详述。
综上所述，本公开实施例中提供的图像分割方法，通过自动确定前景样本点和背景样本点，结合显著性模型以及前背景样本点建立前背景分类模型，利用该前背景分类模型实现图像分割；解决了相关技术中必须需要用户手动粗略地选定前景样本点和背景样本点，在对大量图像进行分割时，分割效率比较低的问题；由于可以自动获取前景样本点和自动样本点，且在建立前背景分类模型时还结合了先验的显著性模型，达到了可以实现自动化选取样本，提高了分类精确度的效果。
下述为本公开装置实施例，可以用于执行本公开方法实施例。对于本公开装置实施例中未披露的细节，请参照本公开方法实施例。
图3是根据一示例性实施例示出的一种图像分割装置的框图，如图3所示，该图像分割装置应用于电子设备中，该图像分割装置包括但不限于：第一建立模块302、样本获取模块304、第二建立模块306和图像分割模块308。
该第一建立模块302被配置为建立图像的显著性模型。
该样本获取模块304被配置为根据显著性模型获取图像中的前景样本点和背景样本点。
该第二建立模块306被配置为根据第一建立模块建立的显著性模型以及样本获取模块获取的前景样本点和背景样本点，建立前背景分类模型。
该图像分割模块308被配置为根据预定图割算法对图像进行分割，预定图割算法利用第二建立模块建立的前背景分类模型以及像素点之间的边缘信息对图像进行分割。
综上所述，本公开实施例中提供的图像分割装置，通过自动确定前景样本点和背景样本点，结合显著性模型以及前背景样本点建立前背景分类模型，利用该前背景分类模型实现图像分割；解决了相关技术中必须需要用户手动粗略地选定前景样本点和背景样本点，在对大量图像进行分割时，分割效率比较低的问题；由于可以自动获取前景样本点和自动样本点，且在建立前背景分类模型时还结合了先验的显著性模型，达到了可以实现自动化选取样本，提高了分类精确度的效果。
图4是根据另一示例性实施例示出的一种图像分割装置的框图，如图4所示，该图像分割装置应用于电子设备中，该图像分割装置包括但不限于：第一建立模块402、样本获取模块404、第二建立模块406和图像分割模块408。
该第一建立模块402被配置为建立图像的显著性模型。
该样本获取模块404被配置为根据显著性模型获取图像中的前景样本点和背景样本点。
该第二建立模块406被配置为根据第一建立模块402建立的显著性模型以及样本获取模块404获取的前景样本点和背景样本点，建立前背景分类模型。
该图像分割模块408被配置为根据预定图割算法对图像进行分割，预定图割算法利用第二建立模块406建立的前背景分类模型以及像素点之间的边缘信息对图像进行分割。
在图4所示实施例中的第一种可能的实现方式中，样本获取模块404可以包括：第一计算单元404a、归一化单元404b、第一确定单元404c和第二确定单元404d。
该第一计算单元404a被配置为根据显著性模型，计算图像中各个像素点的显著性值。
该归一化单元404b被配置为将第一计算单元404a计算出的各个像素点的显著性值进行归一化。
该第一确定单元404c被配置为将归一化单元404b归一化后的显著性值大于预定前景阈值的像素点确定为前景样本点。
该第二确定单元404d被配置为将归一化单元404b归一化后的显著性值小于预定背景阈值的像素点确定为背景样本点。
其中，预定前景阈值大于预定背景阈值，归一化后的各个显著值均位于(0，1)中。
在图4所示实施例中的第二种可能的实现方式中，前背景分类模型包括前景分类模型和背景分类模型，该第二建立模块406可以包括：第一建立单元406a、第二建立单元406b、第一相乘单元406c和第二相乘单元406d。
该第一建立单元406a被配置为根据前景样本点建立前景颜色似然模型；
该第二建立单元406b被配置为根据背景样本点建立背景颜色似然模型；
该第一相乘单元406c被配置为将第一建立模块402建立的显著性模型与第一建立单元406a建立的前景颜色似然模型相乘，得到前景分类模型，前景分类模型用于表征像素点为前景的概率；
该第二相乘单元406d被配置为将第一建立模块402建立的显著性模型与第二建立单元406b建立的背景颜色似然模型相乘，得到背景分类模型，背景分类模型用于表征像素点为背景的概率。
在图4所示实施例中的第三种可能的实现方式中，图像分割模块408可以包括：第二计算单元408a、第三计算单元408b、获取单元408c、构造单元408d和第一分割单元408e。
该第二计算单元408a被配置为利用前景分类模型计算图像中每个像素点的前景相似度；
该第三计算单元408b被配置为利用背景分类模型计算图像中每个像素点的背景相似度；
该获取单元408c被配置为获取图像中相邻像素点之间的相似度；
该构造单元408d被配置为利用各个像素点的前景相似度、各个像素点的背景相似度以及相邻像素点之间的相似度，构造预定图割算法所需的无向图；
该第一分割单元408e被配置为利用预定分割算法对无向图进行分割，完成对图像的分割。
在图4所示实施例中的第四种可能的实现方式中，构造单元408d，包括：构建子单元408d1、第一确定子单元408d2、第二确定子单元408d3和第三确定子单元408d4。
该构建子单元408d1被配置为构建预定图割算法所需的无向图，无向图包括前景顶点、背景顶点、至少一个像素顶点、相邻的两个像素顶点之间的第一类边、像素顶点与背景顶点之间的第二类边，像素顶点与背景顶点之间的第三类边，无向图中的像素顶点与图像中的各个像素点一一对应
该第一确定子单元408d2被配置为对于每条第二类边，将与第二类边相连的像素顶点所对应的像素点的前景相似度，确定为第二类边的权值；
该第二确定子单元408d3被配置为对于每条第三类边，将与第三类边相连的像素顶点所对应的像素点的背景相似度，确定为第三类边的权值；
该第三确定子单元408d4被配置为对于每条第一类边，将与第一类边相连的两个像素顶点所对应的两个像素点之间的相似度，确定为第一类边的权值。
在图4所示实施例中的第五种可能的实现方式中，第一建立模块402可以包括：第二分割单元402a、第四确定单元402b和第三建立单元402c。
该第二分割单元402a被配置为利用预定过分割算法对图像进行过分割，得到至少一个区域，同一个区域中各个像素点的颜色值相同；
该第四确定单元402b被配置为确定每个区域的颜色值和质心；
该第三建立单元402c被配置为根据各个区域所对应的颜色值以及各个区域的质心， 建立显著性模型。
在图4所示实施例中的第六种可能的实现方式中，显著性模型为：
Figure PCTCN2014089297-appb-000013
其中，Si1为区域Ri中任一像素点的显著性值，w(Rj)为区域Rj中的像素点的个数，DS(Ri,Rj)用于表征区域Ri和区域Rj之间空间位置差异的度量值，DC(Ri,Rj)用于表征区域Ri和区域Rj之间颜色差异的度量值，N为对图像进行过分割后得到的区域的总个数，DS(Ri,Rj)为：
Figure PCTCN2014089297-appb-000014
Center(Ri)为区域Ri的质心，Center(Rj)为区域Rj的质心，当图像中各个像素点的坐标均归一化到[0,1]时，
Figure PCTCN2014089297-appb-000015
在图4所示实施例中的第七种可能的实现方式中，第一建立模块402可以包括：归类单元402d和第四建立单元402e。
该归类单元402d被配置为按照各个像素点的颜色值，对图像中各个像素点进行归类，将相同颜色值的像素点归类为同一种颜色类型；
该第四建立单元402e被配置为根据每种颜色类型的颜色值，建立显著性模型。
在图4所示实施例中的第八种可能的实现方式中，显著性模型为：
Figure PCTCN2014089297-appb-000016
其中，w(Pj)为颜色类型Pj中像素点的个数，DC(Pi,Pj)用于表征颜色类型Pi和颜色类型Pj之间颜色差异的度量值。
综上所述，本公开实施例中提供的图像分割装置，通过自动确定前景样本点和背景样本点，结合显著性模型以及前背景样本点建立前背景分类模型，利用该前背景分类模型实现图像分割；解决了相关技术中必须需要用户手动粗略地选定前景样本点和背景样本点，在对大量图像进行分割时，分割效率比较低的问题；由于可以自动获取前景样本点和自动样本点，且在建立前背景分类模型时还结合了先验的显著性模型，达到了可以实现自动化选取样本，提高了分类精确度的效果。
关于上述实施例中的装置，其中各个模块执行操作的具体方式已经在有关该方法的实施例中进行了详细描述，此处将不做详细阐述说明。
图5是根据一示例性实施例示出的一种用于图像分割的装置500的框图。例如，装置500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。
参照图5，装置500可以包括以下一个或多个组件：处理组件502，存储器504，电源组件506，多媒体组件508，音频组件510，输入/输出(I/O)的接口512，传感器组件514，以及通信组件516。
处理组件502通常控制装置500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502可以包括一个或多个处理器518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502可以包括一个或多个模块，便于处理组件502和其他组件之间的交互。例如，处理组件502可以包括多媒体模块，以方便多媒体组件508和处理组件502之间的交互。
存储器504被配置为存储各种类型的数据以支持在装置500的操作。这些数据的示例包括用于在装置500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504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506为装置500的各种组件提供电力。电源组件506可以包括电源管理系统，一个或多个电源，及其他与为装置500生成、管理和分配电力相关联的组件。
多媒体组件508包括在所述装置500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508包括一个前置摄像头和/或后置摄像头。当装置500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510被配置为输出和/或输入音频信号。例如，音频组件510包括一个麦克风(MIC)，当装置500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504或经由通信组件516发送。在一些实施例中，音频组件510还包括一个扬声器，用于输出音频信号。
I/O接口512为处理组件502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514包括一个或多个传感器，用于为装置500提供各个方面的状态评估。例如，传感器组件514可以检测到装置500的打开/关闭状态，组件的相对定位，例如所述组件为装置500的显示器和小键盘，传感器组件514还可以检测装置500或装置500一个组件的位置改变，用户与装置500接触的存在或不存在，装置500方位或加速/减速和装置500的温度变化。传感器组件514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514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514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516被配置为便于装置500和其他设备之间有线或无线方式的通信。装置500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516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516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，装置500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，还提供了一种包括指令的非临时性计算机可读存储介质，例如包括指令的存储器504，上述指令可由装置500的处理器518执行以完成上述方法。例如，所述非临时性计算机可读存储介质可以是ROM、随机存取存储器(RAM)、CD-ROM、磁带、软盘和光数据存储设备等。
本领域技术人员在考虑说明书及实践这里公开的发明后，将容易想到本发明的其它实施方案。本申请旨在涵盖本发明的任何变型、用途或者适应性变化，这些变型、用途或者适应性变化遵循本发明的一般性原理并包括本公开未公开的本技术领域中的公知常识或惯用技术手段。说明书和实施例仅被视为示例性的，本发明的真正范围和精神由下面的权利要求指出。
应当理解的是，本发明并不局限于上面已经描述并在附图中示出的精确结构，并且可以在不脱离其范围进行各种修改和改变。本发明的范围仅由所附的权利要求来限制。

Claims (19)

  1. 一种图像分割方法，其特征在于，包括：
    建立图像的显著性模型；
    根据所述显著性模型获取所述图像中的前景样本点和背景样本点；
    根据所述显著性模型以及所述前景样本点和所述背景样本点，建立前背景分类模型；
    根据预定图割算法对所述图像进行分割，所述预定图割算法利用所述前背景分类模型以及像素点之间的边缘信息对所述图像进行分割。
  2. 根据权利要求1所述的方法，其特征在于，所述根据所述显著性模型获取所述图像中的前景样本点和背景样本点，包括：
    根据所述显著性模型，计算所述图像中各个像素点的显著性值；
    将各个像素点的显著性值进行归一化；
    将归一化后的显著性值大于预定前景阈值的像素点确定为所述前景样本点；
    将归一化后的显著性值小于预定背景阈值的像素点确定为所述背景样本点；
    其中，所述预定前景阈值大于所述预定背景阈值，归一化后的各个显著值均位于(0，1)中。
  3. 根据权利要求1所述的方法，其特征在于，所述前背景分类模型包括前景分类模型和背景分类模型，所述根据所述显著性模型以及所述前景样本点和所述背景样本点，建立前背景分类模型，包括：
    根据所述前景样本点建立前景颜色似然模型；
    根据所述背景样本点建立背景颜色似然模型；
    将所述显著性模型与所述前景颜色似然模型相乘，得到所述前景分类模型，所述前景分类模型用于表征像素点为前景的概率；
    将所述显著性模型与所述背景颜色似然模型相乘，得到所述背景分类模型，所述背景分类模型用于表征像素点为背景的概率。
  4. 根据权利要求3所述的方法，其特征在于，所述根据预定图割算法对所述图像进行分割，包括：
    利用所述前景分类模型计算所述图像中每个像素点的前景相似度；
    利用所述背景分类模型计算所述图像中每个像素点的背景相似度；
    获取所述图像中相邻像素点之间的相似度；
    利用各个像素点的前景相似度、各个像素点的背景相似度以及相邻像素点之间的相似度，构造所述预定图割算法所需的无向图；
    利用所述预定分割算法对所述无向图进行分割，完成对所述图像的分割。
  5. 根据权利要求4所述的方法，其特征在于，所述利用各个像素点的前景相似度、各个像素点的背景相似度以及相邻像素点之间的相似度，构造所述预定图割算法所需的无向图，包括：
    构建所述预定图割算法所需的无向图，所述无向图包括前景顶点、背景顶点、至少一个像素顶点、相邻的两个像素顶点之间的第一类边、所述像素顶点与所述背景顶点之间的第二类边，所述像素顶点与所述背景顶点之间的第三类边，所述无向图中的像素顶点与所述图像中的各个像素点一一对应；
    对于每条第二类边，将与所述第二类边相连的像素顶点所对应的像素点的前景相似度，确定为所述第二类边的权值；
    对于每条第三类边，将与所述第三类边相连的像素顶点所对应的像素点的背景相似度，确定为所述第三类边的权值；
    对于每条第一类边，将与所述第一类边相连的两个像素顶点所对应的两个像素点之间的相似度，确定为所述第一类边的权值。
  6. 根据权利要求1至5中任一所述的方法，其特征在于，所述建立图像的显著性模型，包括：
    利用预定过分割算法对所述图像进行过分割，得到至少一个区域，同一个所述区域中各个像素点的颜色值相同；
    确定每个所述区域的颜色值和质心；
    根据各个区域所对应的颜色值以及各个区域的质心，建立所述显著性模型。
  7. 根据权利要求6所述的方法，其特征在于，所述显著性模型为：
    Figure PCTCN2014089297-appb-100001
    其中，Si1为区域Ri中任一像素点的显著性值，w(Rj)为区域Rj中的像素点的个数，DS(Ri,Rj)用于表征所述区域Ri和所述区域Rj之间空间位置差异的度量值，DC(Ri,Rj)用于表征所述区域Ri和所述区域Rj之间颜色差异的度量值，N为对所述图像进行过分割后得到的区域的总个数，DS(Ri,Rj)为：
    Figure PCTCN2014089297-appb-100002
    Center(Ri)为所述区域Ri的质心，Center(Rj)为所述区域Rj的质心，当所述图像中各个像素点的坐标均归一化到[0,1]时，
    Figure PCTCN2014089297-appb-100003
  8. 根据权利要求1至5中任一所述的方法，其特征在于，所述建立图像的显著性模型，包括：
    按照各个像素点的颜色值，对所述图像中各个像素点进行归类，将相同颜色值的像素点归类为同一种颜色类型；
    根据每种颜色类型的颜色值，建立所述显著性模型。
  9. 根据权利要求8所述的方法，其特征在于，所述显著性模型为：
    Figure PCTCN2014089297-appb-100004
    其中，w(Pj)为颜色类型Pj中像素点的个数，DC(Pi,Pj)用于表征颜色类型Pi和颜色类型Pj之间颜色差异的度量值。
  10. 一种图像分割装置，其特征在于，包括：
    第一建立模块，用于建立图像的显著性模型；
    样本获取模块，用于根据所述显著性模型获取所述图像中的前景样本点和背景样本点；
    第二建立模块，用于根据所述第一建立模块建立的显著性模型以及所述样本获取模块获取的前景样本点和所述背景样本点，建立前背景分类模型；
    图像分割模块，用于根据预定图割算法对所述图像进行分割，所述预定图割算法利用所述第二建立模块建立的前背景分类模型以及像素点之间的边缘信息对所述图像进行分割。
  11. 根据权利要求10所述的装置，其特征在于，所述样本获取模块，包括：
    第一计算单元，用于根据所述显著性模型，计算所述图像中各个像素点的显著性值；
    归一化单元，用于将所述第一计算单元计算出的各个像素点的显著性值进行归一化；
    第一确定单元，用于将所述归一化单元归一化后的显著性值大于预定前景阈值的像素点确定为所述前景样本点；
    第二确定单元，用于将所述归一化单元归一化后的显著性值小于预定背景阈值的像素点确定为所述背景样本点；
    其中，所述预定前景阈值大于所述预定背景阈值，归一化后的各个显著值均位于(0，1)中。
  12. 根据权利要求10所述的装置，其特征在于，所述前背景分类模型包括前景分类模型和背景分类模型，所述第二建立模块，包括：
    第一建立单元，用于根据所述前景样本点建立前景颜色似然模型；
    第二建立单元，用于根据所述背景样本点建立背景颜色似然模型；
    第一相乘单元，用于将所述第一建立模块建立的显著性模型与所述第一建立单元建立的前景颜色似然模型相乘，得到所述前景分类模型，所述前景分类模型用于表征像素点为前景的概率；
    第二相乘单元，用于将所述第一建立模块建立的显著性模型与所述第二建立单元建立的背景颜色似然模型相乘，得到所述背景分类模型，所述背景分类模型用于表征像素点为背景的概率。
  13. 根据权利要求12所述的装置，其特征在于，所述图像分割模块，包括：
    第二计算单元，用于利用所述前景分类模型计算所述图像中每个像素点的前景相似度；
    第三计算单元，用于利用所述背景分类模型计算所述图像中每个像素点的背景相似度；
    获取单元，用于获取所述图像中相邻像素点之间的相似度；
    构造单元，用于利用各个像素点的前景相似度、各个像素点的背景相似度以及相邻像 素点之间的相似度，构造所述预定图割算法所需的无向图；
    第一分割单元，用于利用所述预定分割算法对所述无向图进行分割，完成对所述图像的分割。
  14. 根据权利要求13所述的装置，其特征在于，所述构造单元，包括：
    构建子单元，用于构建所述预定图割算法所需的无向图，所述无向图包括前景顶点背景顶点、至少一个像素顶点、相邻的两个像素顶点之间的第一类边、所述像素顶点与所述背景顶点之间的第二类边，所述像素顶点与所述背景顶点之间的第三类边，所述无向图中的像素顶点与所述图像中的各个像素点一一对应
    第一确定子单元，用于对于每条第二类边，将与所述第二类边相连的像素顶点所对应的像素点的前景相似度，确定为所述第二类边的权值；
    第二确定子单元，用于对于每条第三类边，将与所述第三类边相连的像素顶点所对应的像素点的背景相似度，确定为所述第三类边的权值；
    第三确定子单元，用于对于每条第一类边，将与所述第一类边相连的两个像素顶点所对应的两个像素点之间的相似度，确定为所述第一类边的权值。
  15. 根据权利要求10至14中任一所述的装置，其特征在于，所述第一建立模块，包括：
    第二分割单元，用于利用预定过分割算法对所述图像进行过分割，得到至少一个区域，同一个所述区域中各个像素点的颜色值相同；
    第三确定单元，用于确定每个所述区域的颜色值和质心；
    第三建立单元，用于根据各个区域所对应的颜色值以及各个区域的质心，建立所述显著性模型。
  16. 根据权利要求15所述的装置，其特征在于，所述显著性模型为：
    Figure PCTCN2014089297-appb-100005
    其中，Si1为区域Ri中任一像素点的显著性值，w(Rj)为区域Rj中的像素点的个数，DS(Ri,Rj)用于表征所述区域Ri和所述区域Rj之间空间位置差异的度量值，DC(Ri,Rj)用于表征所述区域Ri和所述区域Rj之间颜色差异的度量值，N为对所述图像进行过分割后得到的区域的总个数，DS(Ri,Rj)为：
    Figure PCTCN2014089297-appb-100006
    Center(Ri)为所述区域Ri的质心，Center(Rj)为所述区域Rj的质心，当所述图像中各个像素点的坐标均归一化到[0,1]时，
    Figure PCTCN2014089297-appb-100007
  17. 根据权利要求10至14中任一所述的装置，其特征在于，所述第一建立模块，包括：
    归类单元，用于按照各个像素点的颜色值，对所述图像中各个像素点进行归类，将相同颜色值的像素点归类为同一种颜色类型；
    第四建立单元，用于根据每种颜色类型的颜色值，建立所述显著性模型。
  18. 根据权利要求17所述的方法，其特征在于，所述显著性模型为：
    Figure PCTCN2014089297-appb-100008
    其中，w(Pj)为颜色类型Pj中像素点的个数，DC(Pi,Pj)用于表征颜色类型Pi和颜色类型Pj之间颜色差异的度量值。
  19. 一种图像分割装置，其特征在于，包括：
    处理器；
    用于存储所述处理器可执行指令的存储器；
    其中，所述处理器被配置为：
    建立图像的显著性模型；
    根据所述显著性模型获取所述图像中的前景样本点和背景样本点；
    根据所述显著性模型以及所述前景样本点和所述背景样本点，建立前背景分类模型；
    根据预定图割算法对所述图像进行分割，所述预定图割算法利用所述前背景分类模型以及像素点之间的边缘信息对所述图像进行分割。
PCT/CN2014/089297 2014-05-05 2014-10-23 图像分割方法及装置 WO2015169061A1 (zh)

Priority Applications (6)

Application Number Priority Date Filing Date Title
RU2014153987/08A RU2596580C2 (ru) 2014-05-05 2014-10-23 Способ и устройство для сегментации изображения
BR112014032942A BR112014032942A2 (pt) 2014-05-05 2014-10-23 método e dispositivo para segmentação de imagens
JP2016517158A JP6125099B2 (ja) 2014-05-05 2014-10-23 画像分割方法、画像分割装置、プログラム、及び記録媒体
KR1020147036180A KR101670004B1 (ko) 2014-05-05 2014-10-23 화상 분할 방법, 장치, 프로그램 및 기록매체
MX2014015363A MX358601B (es) 2014-05-05 2014-10-23 Método y dispositivo para segmentación de imagen.
US14/583,816 US9633444B2 (en) 2014-05-05 2014-12-29 Method and device for image segmentation

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201410187226.7 2014-05-05
CN201410187226.7A CN103996189B (zh) 2014-05-05 2014-05-05 图像分割方法及装置

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US14/583,816 Continuation US9633444B2 (en) 2014-05-05 2014-12-29 Method and device for image segmentation

Publications (1)

Publication Number Publication Date
WO2015169061A1 true WO2015169061A1 (zh) 2015-11-12

Family

ID=51310345

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2014/089297 WO2015169061A1 (zh) 2014-05-05 2014-10-23 图像分割方法及装置

Country Status (8)

Country Link
EP (1) EP2942753B1 (zh)
JP (1) JP6125099B2 (zh)
KR (1) KR101670004B1 (zh)
CN (1) CN103996189B (zh)
BR (1) BR112014032942A2 (zh)
MX (1) MX358601B (zh)
RU (1) RU2596580C2 (zh)
WO (1) WO2015169061A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110111338A (zh) * 2019-04-24 2019-08-09 广东技术师范大学 一种基于超像素时空显著性分割的视觉跟踪方法
CN112347975A (zh) * 2020-11-23 2021-02-09 浙江华络通信设备有限公司 一种可视化融合通信调度系统及其实现方法
CN112634312A (zh) * 2020-12-31 2021-04-09 上海商汤智能科技有限公司 图像背景处理方法、装置、电子设备及存储介质
CN112800915A (zh) * 2021-01-20 2021-05-14 北京百度网讯科技有限公司 建筑物变化检测方法、装置、电子设备以及存储介质

Families Citing this family (36)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104217440B (zh) * 2014-09-28 2017-03-01 民政部国家减灾中心 一种从遥感图像中提取建成区的方法
CN104809729B (zh) * 2015-04-29 2018-08-28 山东大学 一种鲁棒的图像显著性区域自动分割方法
CN105069774B (zh) * 2015-06-30 2017-11-10 长安大学 基于多示例学习与图割优化的目标分割方法
CN105353674B (zh) * 2015-11-03 2018-04-13 浙江华奕生态建设有限公司 一种多功能生活污水处理集成控制系统及方法
CN105342769A (zh) * 2015-11-20 2016-02-24 宁波大业产品造型艺术设计有限公司 智能电动轮椅
CN105608717A (zh) * 2015-12-22 2016-05-25 肖古华 一种ct系统和ct图像重建方法
CN105395210A (zh) * 2015-12-22 2016-03-16 管迪 一种ct扫描成像方法
CN105662488A (zh) * 2016-01-12 2016-06-15 刘书强 一种断指探查台
CN105748105A (zh) * 2016-01-18 2016-07-13 刘焕涛 一种乳腺检查系统
CN106650744B (zh) * 2016-09-16 2019-08-09 北京航空航天大学 局部形状迁移指导的图像对象共分割方法
CN108073871A (zh) * 2016-11-18 2018-05-25 北京体基科技有限公司 基于两种预定区域识别手部区域的方法及装置
CN108073881A (zh) * 2016-11-18 2018-05-25 北京体基科技有限公司 基于预定区域和临近像素识别手部区域的方法及装置
CN106898001B (zh) * 2017-01-04 2020-04-14 努比亚技术有限公司 一种实现交互式图像分割的方法、装置及终端
CN106934774B (zh) * 2017-03-07 2019-11-29 李强 一种利用超声波治疗肛瘘的控制系统
CN107067030A (zh) * 2017-03-29 2017-08-18 北京小米移动软件有限公司 相似图片检测的方法和装置
CN107451522A (zh) * 2017-04-28 2017-12-08 山东省农业可持续发展研究所 一种农业干旱监测与预警预报方法
CN108960247B (zh) * 2017-05-22 2022-02-25 阿里巴巴集团控股有限公司 图像显著性检测方法、装置以及电子设备
CN107273893A (zh) * 2017-06-14 2017-10-20 武汉梦之蓝科技有限公司 一种智能城市绿化遥感调查的数据校正控制系统
CN107088028A (zh) * 2017-06-29 2017-08-25 武汉洁美雅科技有限公司 一种智能新式干湿两用吸尘器机器人控制系统
CN107452003A (zh) * 2017-06-30 2017-12-08 大圣科技股份有限公司 一种含有深度信息的图像分割的方法及装置
CN107573029A (zh) * 2017-10-23 2018-01-12 福建泉州皓佳新材料有限公司 一种陶瓷废釉水制备的多彩艺术砖及其制备方法
CN107894252A (zh) * 2017-11-14 2018-04-10 江苏科沃纺织有限公司 一种实时监测地埋伸缩式喷滴灌装置运行状态的监控系统
CN108764325B (zh) * 2018-05-23 2022-07-08 腾讯科技（深圳）有限公司 图像识别方法、装置、计算机设备和存储介质
CN109035326A (zh) * 2018-06-19 2018-12-18 北京理工大学 基于亚像素图像识别的高精度定位技术
CN109173263B (zh) * 2018-08-31 2021-08-24 腾讯科技（深圳）有限公司 一种图像数据处理方法和装置
CN109168220A (zh) * 2018-09-28 2019-01-08 湖北民族学院 一种提高青柚产量品质的光控方法及系统
JP6716765B1 (ja) * 2018-12-28 2020-07-01 キヤノン株式会社 画像処理装置、画像処理システム、画像処理方法、プログラム
CN110909724B (zh) * 2019-10-08 2023-11-28 华北电力大学 一种多目标图像的缩略图生成方法
CN110853063B (zh) * 2019-10-31 2023-04-07 广州方硅信息技术有限公司 图像分割信息的处理方法、装置、设备和存储介质
CN111539969B (zh) * 2020-04-23 2023-06-09 武汉铁路职业技术学院 图像边缘检测方法、装置、计算机设备及存储介质
CN111597934A (zh) * 2020-04-30 2020-08-28 重庆科技学院 用于为统计应用处理训练数据的系统和方法
CN111640129B (zh) * 2020-05-25 2023-04-07 电子科技大学 一种应用于室内墙面施工机器人的视觉砂浆识别系统
CN111724396B (zh) * 2020-06-17 2023-07-14 泰康保险集团股份有限公司 图像分割方法及装置、计算机可读存储介质、电子设备
CN114708464B (zh) * 2022-06-01 2022-08-30 广东艺林绿化工程有限公司 一种基于道路垃圾分类的市政环卫清扫垃圾车清扫方法
CN115841490B (zh) * 2023-02-23 2023-05-09 山东泗水山岭石材厂 基于图像处理的采矿过程矿石颗粒分割方法及系统
CN116721115A (zh) * 2023-06-15 2023-09-08 小米汽车科技有限公司 金相组织获取方法、装置、存储介质及芯片

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1081648B1 (en) * 1999-09-02 2005-05-11 Xerox Corporation Method for processing a digital image
JP2008042346A (ja) * 2006-08-02 2008-02-21 Canon Inc 画像処理方法、画像処理装置
US7899248B2 (en) * 2007-08-30 2011-03-01 Seiko Epson Corporation Fast segmentation of images
US8009911B2 (en) * 2007-08-30 2011-08-30 Siemens Aktiengesellschaft Interactive image segmentation on directed graphs
CN102982545A (zh) * 2012-11-22 2013-03-20 清华大学深圳研究生院 一种图像深度估计方法
CN103208123A (zh) * 2013-04-19 2013-07-17 广东图图搜网络科技有限公司 图像分割方法与系统

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
RU2121714C1 (ru) * 1997-02-19 1998-11-10 Закрытое акционерное общество "Медицинские компьютерные системы" Способ автоматизированной сегментации изображения цитологического препарата
WO2006066325A1 (en) * 2004-12-21 2006-06-29 Canon Kabushiki Kaisha Segmenting digital image and producing compact representation
JP5235770B2 (ja) * 2009-04-27 2013-07-10 日本電信電話株式会社 顕著領域映像生成方法、顕著領域映像生成装置、プログラムおよび記録媒体
JP5464739B2 (ja) * 2010-01-22 2014-04-09 Ｋｄｄｉ株式会社 画像領域分割装置、画像領域分割方法および画像領域分割プログラム
RU2489752C2 (ru) * 2011-08-15 2013-08-10 Дмитрий Валерьевич Шмунк Способ сегментации изображений
JP2013077296A (ja) 2011-09-14 2013-04-25 Ricoh Co Ltd 画像処理装置、画像処理方法およびプログラム

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1081648B1 (en) * 1999-09-02 2005-05-11 Xerox Corporation Method for processing a digital image
JP2008042346A (ja) * 2006-08-02 2008-02-21 Canon Inc 画像処理方法、画像処理装置
US7899248B2 (en) * 2007-08-30 2011-03-01 Seiko Epson Corporation Fast segmentation of images
US8009911B2 (en) * 2007-08-30 2011-08-30 Siemens Aktiengesellschaft Interactive image segmentation on directed graphs
CN102982545A (zh) * 2012-11-22 2013-03-20 清华大学深圳研究生院 一种图像深度估计方法
CN103208123A (zh) * 2013-04-19 2013-07-17 广东图图搜网络科技有限公司 图像分割方法与系统

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110111338A (zh) * 2019-04-24 2019-08-09 广东技术师范大学 一种基于超像素时空显著性分割的视觉跟踪方法
CN112347975A (zh) * 2020-11-23 2021-02-09 浙江华络通信设备有限公司 一种可视化融合通信调度系统及其实现方法
CN112347975B (zh) * 2020-11-23 2023-04-07 浙江华络通信设备有限公司 一种可视化融合通信调度系统及其实现方法
CN112634312A (zh) * 2020-12-31 2021-04-09 上海商汤智能科技有限公司 图像背景处理方法、装置、电子设备及存储介质
CN112634312B (zh) * 2020-12-31 2023-02-24 上海商汤智能科技有限公司 图像背景处理方法、装置、电子设备及存储介质
CN112800915A (zh) * 2021-01-20 2021-05-14 北京百度网讯科技有限公司 建筑物变化检测方法、装置、电子设备以及存储介质
CN112800915B (zh) * 2021-01-20 2023-06-27 北京百度网讯科技有限公司 建筑物变化检测方法、装置、电子设备以及存储介质

Also Published As

Publication number Publication date
BR112014032942A2 (pt) 2017-06-27
CN103996189A (zh) 2014-08-20
EP2942753B1 (en) 2018-07-18
RU2014153987A (ru) 2016-07-20
EP2942753A1 (en) 2015-11-11
JP2016522951A (ja) 2016-08-04
KR20160009477A (ko) 2016-01-26
KR101670004B1 (ko) 2016-10-27
JP6125099B2 (ja) 2017-05-10
MX358601B (es) 2018-08-27
RU2596580C2 (ru) 2016-09-10
CN103996189B (zh) 2017-10-03
MX2014015363A (es) 2016-02-19

Similar Documents

Publication Publication Date Title
WO2015169061A1 (zh) 图像分割方法及装置
RU2577188C1 (ru) Способ, аппарат и устройство для сегментации изображения
RU2628494C1 (ru) Способ и устройство для генерирования фильтра изображения
US9633444B2 (en) Method and device for image segmentation
US10007841B2 (en) Human face recognition method, apparatus and terminal
WO2021031609A1 (zh) 活体检测方法及装置、电子设备和存储介质
US20170154206A1 (en) Image processing method and apparatus
WO2017088470A1 (zh) 图像分类方法及装置
WO2020134866A1 (zh) 关键点检测方法及装置、电子设备和存储介质
WO2017092127A1 (zh) 视频归类方法及装置
CN106650575A (zh) 人脸检测方法及装置
EP2998960B1 (en) Method and device for video browsing
WO2017020476A1 (zh) 关联用户的确定方法及装置
CN107563994B (zh) 图像的显著性检测方法及装置
RU2632578C2 (ru) Способ и устройство выделения характеристики
WO2017140109A1 (zh) 压力检测方法和装置
CN109034150B (zh) 图像处理方法及装置
CN106534951B (zh) 视频分割方法和装置
WO2015196715A1 (zh) 图像重定位方法、装置及终端
EP3975046B1 (en) Method and apparatus for detecting occluded image and medium
WO2017140108A1 (zh) 压力检测方法和装置
CN111259967A (zh) 图像分类及神经网络训练方法、装置、设备及存储介质
WO2023273498A1 (zh) 深度检测方法及装置、电子设备和存储介质
TWI770531B (zh) 人臉識別方法、電子設備和儲存介質
JP2022535639A (ja) 性能パラメータ決定方法および装置、電子機器、記憶媒体、並びにプログラム製品

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: MX/A/2014/015363

Country of ref document: MX

ENP Entry into the national phase

Ref document number: 2016517158

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 20147036180

Country of ref document: KR

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2014153987

Country of ref document: RU

Kind code of ref document: A

REG Reference to national code

Ref country code: BR

Ref legal event code: B01A

Ref document number: 112014032942

Country of ref document: BR

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

Ref document number: 14891404

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 14891404

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 112014032942

Country of ref document: BR

Kind code of ref document: A2

Effective date: 20141230