CN109685806B - Image significance detection method and device - Google Patents
Image significance detection method and device Download PDFInfo
- Publication number
- CN109685806B CN109685806B CN201811355398.5A CN201811355398A CN109685806B CN 109685806 B CN109685806 B CN 109685806B CN 201811355398 A CN201811355398 A CN 201811355398A CN 109685806 B CN109685806 B CN 109685806B
- Authority
- CN
- China
- Prior art keywords
- initial
- superpixel
- target image
- saliency
- path
- 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.)
- Expired - Fee Related
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
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
- G06T2207/10004—Still image; Photographic image
Abstract
The embodiment of the invention provides an image significance detection method and device, wherein the method comprises the following steps: extracting a plurality of features from a target image, generating a feature map, respectively carrying out segmentation of a plurality of scales on the target image based on a superpixel segmentation algorithm to obtain initial superpixel blocks, and obtaining a significant value of each initial superpixel block according to each feature map and each initial superpixel block; generating initial saliency maps under the scales according to the saliency values of the initial super-pixel blocks, and optimizing the initial saliency maps based on an ant colony optimization algorithm to obtain optimal saliency maps under the scales; and acquiring a final saliency map of the target image according to the optimal saliency maps under the scales. The embodiment of the invention improves the accuracy of the image significance detection under low illumination.
Description
Technical Field
The embodiment of the invention belongs to the technical field of image processing, and particularly relates to an image saliency detection method and device.
Background
The visual saliency information of the image can reflect the degree of stimulation of the human visual system by different areas in the image. The detection of the salient object of the image is a basic research in the field of computer vision, and important areas or objects which can draw the attention of human eyes can be found out from massive image data through the detection of the saliency.
Most of the existing image significance detection models are only suitable for visible light environments, however, in practical application scenes, low-light environments often occur, such as environments affected by weather interference such as rain, snow and haze or poor night light conditions, so that the measurement of features is easily affected by noise interference, and many factors such as background change and fuzzy texture, and the reliability of significance detection results is greatly reduced.
Although the existing research has achieved some achievements in image significance detection, the effect of image significance detection under low illumination is still poor due to low signal-to-noise ratio and limited effective features under low illumination. Therefore, how to detect the saliency of the low-illumination image is still a challenging problem.
Disclosure of Invention
In order to overcome the above-mentioned problem of poor image saliency detection effect under low light or at least partially solve the above-mentioned problem, embodiments of the present invention provide an image saliency detection method and apparatus.
According to a first aspect of the embodiments of the present invention, there is provided an image saliency detection method, including:
extracting a plurality of features from a target image, generating a feature map, respectively carrying out segmentation of a plurality of scales on the target image based on a superpixel segmentation algorithm to obtain initial superpixel blocks, and obtaining a significant value of each initial superpixel block according to each feature map and each initial superpixel block;
generating initial saliency maps under the scales according to the saliency values of the initial super-pixel blocks, and optimizing the initial saliency maps based on an ant colony optimization algorithm to obtain optimal saliency maps under the scales;
and acquiring a final saliency map of the target image according to the optimal saliency maps under the scales.
According to a second aspect of embodiments of the present invention, there is provided an image saliency detection apparatus including:
the segmentation module is used for extracting a plurality of features from a target image, generating a feature map, respectively segmenting the target image in a plurality of scales based on a superpixel segmentation algorithm, acquiring initial superpixel blocks, and acquiring a significant value of each initial superpixel block according to each feature map and each initial superpixel block;
the optimization module is used for generating initial saliency maps under all scales according to the saliency values of all the initial super-pixel blocks, optimizing all the initial saliency maps based on an ant colony optimization algorithm and obtaining optimal saliency maps under all the scales;
and the acquisition module is used for acquiring a final saliency map of the target image according to the optimal saliency map under each scale.
According to a third aspect of the embodiments of the present invention, there is also provided an electronic apparatus, including:
at least one processor; and
at least one memory communicatively coupled to the processor, wherein:
the memory stores program instructions executable by the processor to invoke the method for detecting image saliency provided by any of the various possible implementations of the first aspect.
According to a fourth aspect of embodiments of the present invention, there is also provided a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method for detecting image saliency provided by any of the various possible implementations of the first aspect.
The embodiment of the invention provides an image saliency detection method and device, the method comprises the steps of dividing a target image into a plurality of initial super-pixel blocks by using a super-pixel division algorithm, obtaining an initial saliency map of the target image according to multi-dimensional features of the initial super-pixel blocks, regarding each initial super-pixel block in the initial saliency map as a node, constructing a map model, optimizing the initial saliency map by using an ant colony optimization algorithm on the basis of the map model, detecting the saliency of the target image according to the optimized initial saliency map, improving the saliency detection accuracy of a low-illumination image, and providing support for night safety monitoring, complex environment target positioning and other problems.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
Fig. 1 is a schematic overall flow chart of an image saliency detection method according to an embodiment of the present invention;
fig. 2 is a schematic overall flowchart of an image saliency detection method according to another embodiment of the present invention;
fig. 3 is a schematic view of an overall structure of an image saliency detection apparatus according to an embodiment of the present invention;
fig. 4 is a schematic view of an overall structure of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
An embodiment of the present invention provides an image saliency detection method, and fig. 1 is a schematic overall flow chart of the image saliency detection method provided by the embodiment of the present invention, where the method includes: s101, extracting a plurality of features from a target image, generating a feature map, respectively carrying out segmentation of a plurality of scales on the target image based on a superpixel segmentation algorithm, obtaining initial superpixel blocks, and obtaining a significant value of each initial superpixel block according to each feature map and each initial superpixel block;
in which, the target is subjected to multi-dimensional feature extraction, and the embodiment is not limited to the kind of the extracted features. The super-pixel segmentation algorithm is an irregular pixel block which is formed by adjacent pixels with similar texture, color, brightness and other characteristics and has certain visual significance, the pixels are grouped by utilizing the similarity of the characteristics among the pixels, and a large number of pixels are replaced by a small number of super-pixels to express image characteristics, so that the complexity of image post-processing is reduced. The target image is an image that needs to be subjected to saliency detection. The initial superpixel block is a superpixel obtained by directly using a superpixel segmentation algorithm to segment the target image. Each initial superpixel block has a class label. The present embodiment is applicable to a variety of superpixel segmentation algorithms, using a superpixel segmentation algorithm to segment the target image into a plurality of initial superpixel blocks. And acquiring the multi-dimensional characteristics of the initial superpixel blocks from the multi-dimensional characteristics of the target image according to the areas of the initial superpixel blocks in the target image, and acquiring the significant value of each initial superpixel block by using a global contrast method according to the multi-dimensional characteristics of each initial superpixel block. The target image is divided into a plurality of scales respectively based on the superpixel division algorithm, namely the target image is divided into a plurality of times by using the superpixel division algorithm, and the initial number of superpixels in each division result is different.
S102, generating initial saliency maps under all scales according to the saliency values of all the initial super-pixel blocks, and optimizing all the initial saliency maps based on an ant colony optimization algorithm to obtain optimal saliency maps under all the scales;
an initial saliency map of the target image is generated using the initial super-pixel blocks at the scales. The ant colony optimization algorithm is a probabilistic algorithm for searching for an optimized path, has the characteristics of distribution calculation, information positive feedback and heuristic search, and is essentially a heuristic global optimization algorithm in the evolutionary algorithm. And updating the initial superpixel block by using an ant colony optimization algorithm according to the characteristics of the initial superpixel block to obtain a plurality of superpixel regions. The super-pixel region is obtained by updating the initial super-pixel block by using an ant colony optimization algorithm.
S103, acquiring a final saliency map of the target image according to the optimal saliency map under each scale.
And taking the multidimensional feature of the corresponding area of each super pixel area in the target image as the multidimensional feature corresponding to each super pixel area. And calculating the significant value of each super pixel region by using a global contrast method according to the multi-dimensional features corresponding to each super pixel region. And acquiring a saliency map of the target image by using the saliency values of the super pixel regions.
In the embodiment, the target image is divided into a plurality of initial superpixel blocks by using a superpixel division algorithm, the initial saliency map of the target image is obtained according to the multidimensional characteristics of each initial superpixel block, each initial superpixel block in the initial saliency map is taken as a node to construct a map model, the initial saliency map is optimized by using an ant colony optimization algorithm on the basis of the map model, the saliency of the target image is detected according to the optimized initial saliency map, the saliency detection accuracy of the low-light image is improved, and support is provided for the problems of night safety monitoring, complex environment target positioning and the like.
On the basis of the above embodiment, the step of extracting a plurality of features from the target image in the present embodiment specifically includes: extracting color features, texture features, direction features and gradient features of the target image; wherein the color features of the target image comprise color features of the target image in LAB, HSV and YCbCr spaces.
Specifically, 12 low-level visual features including color features, texture features, direction features, and gradient features are extracted from the target image. The color features are obtained by converting the target image in RGB format into images in LAB, HSV and YCbCr color spaces, and then extracting 9 features from 9 channels of three color spaces. The texture features are expressed by using the two-dimensional entropy of the target image, and have strong resistance to geometric deformation and background noise. The Gabor filters in different directions are used for extracting the direction characteristics with rotation invariance and global characteristics, so that the influence of weak illumination is reduced. By averaging the horizontal and vertical gradients, gradient signatures can be calculated to describe the amplitude variation of the local gray scale.
On the basis of the foregoing embodiment, in this embodiment, the target image is divided in multiple scales based on a superpixel division algorithm, and the step of obtaining an initial superpixel block specifically includes: determining an initial seed point in the target image according to a plurality of preset super-pixel numbers; updating the initial seed points according to the gradient value of each pixel point in the neighborhood of each initial seed point to obtain new seed points; and taking pixels in a preset range with each new seed point as a center as neighborhood pixels of each new seed point, distributing a class label to the neighborhood pixels of each new seed point according to the distance between each new seed point and the neighborhood pixels of each new seed point, and taking the pixels with the same class label as an initial super-pixel block.
Specifically, the present embodiment performs superpixel segmentation on the target image using a Simple Linear Iterative Clustering (SLIC) algorithm. First, a seed point, i.e., a cluster center, is initialized. And uniformly distributing seed points in the target image according to the preset number K of the super pixels. Assuming that the target picture has N pixel points in total and is pre-divided into K super pixels with the same size, the size of each super pixel is N/K, and the distance between adjacent seed points is approximately S ═ sqrt (N/K). In order to avoid that the seed point falls on the contour boundary with a large gradient and affects the subsequent clustering effect, in this embodiment, the seed point is reselected in the n × n neighborhood of the seed point, and generally n is equal to 3. The specific method is to calculate the gradient values of all pixel points in the neighborhood of each seed point and move the seed point to the place with the minimum gradient in the neighborhood of the seed point. Then, a category label is allocated to each pixel point in the neighborhood around each seed point, namely, which cluster center belongs to. Unlike the standard k-means search in the whole graph, the search range of SLIC is limited to 2S by 2S, which can speed up convergence.
On the basis of the foregoing embodiment, in this embodiment, the step of obtaining the saliency value of each of the initial super-pixel blocks according to the multidimensional feature of each of the initial super-pixel blocks specifically includes: calculating one-dimensional entropy of each feature map of the target image; taking the features of the preset number with the maximum one-dimensional entropy of the target image as a target feature map; and acquiring the significant value of each initial superpixel block based on a global contrast method according to the target feature map of each initial superpixel block.
Specifically, a feature map is extracted from each dimension of the feature of the target image, and the one-dimensional entropy of each feature map of the target image is calculated according to the following formula:
wherein, entropy is one-dimensional entropy of any characteristic diagram, piAnd expressing the proportion of the pixel point with the gray level i in the characteristic diagram. And selecting a preset number of feature maps with the largest dimension entropy value from all feature maps of the target image to perform significance detection. For example, 12-dimensional features of the target image are extracted, 9 feature maps with the largest dimensional entropy are selected from the 12 feature maps of the target image for saliency detection, and features corresponding to the selected 9 feature maps are used as target features, namely features used for saliency detection, so that the optimal visual features of various weak-light images are selected in an adaptive mode. Because the size of the feature map is the same as that of the target image, the corresponding area of each initial superpixel block in 9 feature maps is used as 9 types of target feature maps of each initial superpixel block according to the area of each initial superpixel block in the target image. And acquiring the significant value of each initial superpixel block based on a global contrast method according to the 9 target feature maps of each initial superpixel block. Similarly, according to the corresponding area of each super-pixel area in the target image, the corresponding area of each super-pixel area in 9 feature maps is used as 9 target features corresponding to each super-pixel area, and according to the 9 target feature maps corresponding to each super-pixel area, the significant value of each super-pixel area is obtained based on the global contrast method.
On the basis of the foregoing embodiment, in this embodiment, the step of merging the initial superpixel blocks based on an ant colony optimization algorithm to obtain a plurality of superpixel regions specifically includes: calculating the distance between any two of the initial superpixel blocks; comparing the distance between any two initial superpixel blocks with a preset threshold value to obtain the information content of a path between any two initial superpixel blocks; calculating the probability of the ant selecting the path according to the information content of the path between any two initial superpixel blocks; when the ants move according to the probability, the information quantity on the path is updated, and the probability that the ants select the path is calculated according to the updated information quantity of the path until a preset termination condition is reached; and if the probability of the ant selecting path in each iteration is greater than that in the first iteration, combining the initial superpixel blocks corresponding to the paths selected by the ants, and taking the superpixel obtained after combination as a superpixel region.
Specifically, each initial superpixel block is regarded as a node to construct a graph model, m ants expand on randomly selected nodes, and the process of searching for the optimal superpixel is used as the process of finding a food source in a search space by the ant. Any two initial superpixel blocks siAnd sjThe distance between them defines dijComprises the following steps:
where m represents the dimension of an ant or node, | · | represents the Euclidean distance, wkThe representation weighting factor is determined based on the degree of contribution of each initial superpixel block in the cluster. Denote the cluster radius by r, Iij(t) denotes the time t from siTo sjAmount of information on the inter-path, dijDenotes from siTo sjThen:
according to the heuristic function and the information quantity on each path, calculating ant selection slave nodes siTo node sjThe path probability of (2). And in the moving process of the ants, the residual information quantity on each path is changed, the information quantity on the paths is updated after each iteration is finished, and in the next iteration, the probability that the ants select each path is calculated by using the updated information quantity on each path until a preset termination condition is reached, such as the iteration times reach a preset constant. Updating the amount of information on the path as the ant moves according to the probability by:
Iij(t+1)＝(1-ρ)·Iij(t)+ρ·ΔIij；
where I and j represent the numbers of any two initial superpixel blocks, Iij(t) denotes the initial superpixel block s at time tiTo the initial super-pixel block sjAmount of information of (1), Iij(t +1) denotes the initial superpixel block s at time t +1iTo the initial super-pixel block sjP represents a rate of change in disappearance of the information amount with time,
indicating the information increment of m ants on each path at the end of the iteration. After all m ants move, the information matrix may be updated as:
wherein the content of the first and second substances,
representing an information attenuation parameter, Iij(0) An information matrix representing the first loop, and n represents the total number of iterations.
If the probability of an ant selecting any path in each iteration is greater than the probability of the ant selecting the path in the first iteration, that is to say
The initial superpixel block s corresponding to the path is usedjIs combined to siThe neighborhood of (2) takes each merged superpixel as a superpixel region. Wherein the content of the first and second substances,
selecting initial superpixel block s for kth ant in t iterationiTo the initial super-pixel block sjProbability of paths between, P (0) selects from the initial superpixel block s for ants in the first iterationiTo the initial super-pixel block sjThe probability of a path in between. The obtained optimal clustering center is as follows:
wherein s isjTo do as requiredMerged initial superpixel block, niFor the number of initial superpixel blocks to be merged, OjAs a food source, can be used as siThen calculating the optimal significant value of each super pixel region, and generating a significant map according to the significant values of all the super pixel regions.
On the basis of the foregoing embodiment, in this embodiment, the step of calculating the probability that the ant selects the path according to the information amount of the path between any two initial superpixel blocks by using the following formula specifically includes:
where k denotes the ant number, I and j denote the numbers of any two initial superpixel blocks, Iij(t) denotes the initial superpixel block s at time tiTo the initial super-pixel block sjηij＝1/dijRepresenting a heuristic function, dijIs s isiTo sjDistance of (d), omega(s)i) Denotes siα denotes an information heuristic, β denotes the influence of heuristic information
On the basis of the foregoing embodiment, in this embodiment, the step of acquiring a final saliency map of the target image according to the optimal saliency map under each of the scales specifically includes: fusing the optimal saliency maps under each scale; and filtering the fused saliency map to obtain a final saliency map of the target image.
Specifically, the target image is segmented for multiple times by using SLIC algorithms with different preset superpixel numbers, as shown in fig. 2, for example, the target image is segmented by using SLIC algorithms with preset superpixel numbers of 100, 200, and 300, respectively, to obtain three superpixel segmentation results. Simultaneously extracting 12 features of the target image, including values L, A and B of the target image on three channels in an LAB space, values H, S and V of the target image on three channels in an HSV space, values Y, Cb and Cr of the target image on three channels in a YCbCr space, a texture feature T of the target image, an orientation feature O of the target image, and a gradient feature G of the target image. And calculating the one-dimensional entropy of each feature map according to the feature maps corresponding to the 12 features of the target image. And selecting 9 feature graphs with the largest one-dimensional entropy as optimal features. And obtaining three initial saliency maps corresponding to the three segmentation results based on a global contrast method according to 9 characteristics of each initial superpixel block in each superpixel segmentation result, optimizing each initial saliency map by using an ant colony optimization algorithm, and then obtaining the optimization results of the three initial saliency maps based on the global contrast method. And fusing the three optimization results, and filtering the fusion result to obtain a final saliency map of the target image.
In another embodiment of the present invention, an image saliency detection apparatus is provided, which is used for implementing the method in the foregoing embodiments. Therefore, the description and definition in the foregoing embodiments of the image saliency detection method may be used for understanding of each execution module in the embodiments of the present invention. Fig. 3 is a schematic diagram of an overall structure of an image saliency detection apparatus provided in an embodiment of the present invention, where the apparatus includes a segmentation module 301, an optimization module 302, and an acquisition module 303; wherein:
a segmentation module 301, configured to extract a plurality of features from a target image, generate a feature map, perform multiple-scale segmentation on the target image based on a superpixel segmentation algorithm, respectively, obtain initial superpixel blocks, and obtain a significant value of each initial superpixel block according to each feature map and each initial superpixel block;
in which, the target is subjected to multi-dimensional feature extraction, and the embodiment is not limited to the kind of the extracted features. The super-pixel segmentation algorithm is an irregular pixel block which is formed by adjacent pixels with similar texture, color, brightness and other characteristics and has certain visual significance, the pixels are grouped by utilizing the similarity of the characteristics among the pixels, and a large number of pixels are replaced by a small number of super-pixels to express image characteristics, so that the complexity of image post-processing is reduced. The target image is an image that needs to be subjected to saliency detection. The initial superpixel block is a superpixel obtained by directly using a superpixel segmentation algorithm to segment the target image. Each initial superpixel block has a class label. The segmentation module 301 segments the target image into a plurality of initial superpixel blocks using a superpixel segmentation algorithm, and this embodiment is applicable to various superpixel segmentation algorithms. And acquiring the multi-dimensional characteristics of the initial superpixel blocks from the multi-dimensional characteristics of the target image according to the areas of the initial superpixel blocks in the target image, and acquiring the significant value of each initial superpixel block by using a global contrast method according to the multi-dimensional characteristics of each initial superpixel block.
The optimization module 302 is configured to generate an initial saliency map at each scale according to the saliency value of each initial super-pixel block, optimize each initial saliency map based on an ant colony optimization algorithm, and obtain an optimal saliency map at each scale;
an initial saliency map of the target image is generated using the saliency values of each initial super-pixel block. The ant colony optimization algorithm is a probabilistic algorithm for searching for an optimized path, has the characteristics of distribution calculation, information positive feedback and heuristic search, and is essentially a heuristic global optimization algorithm in the evolutionary algorithm. The optimization module 302 updates the initial superpixel block using an ant colony optimization algorithm according to the characteristics of the initial superpixel block to obtain a superpixel region. The super-pixel region is obtained by updating the initial super-pixel block by using an ant colony optimization algorithm.
The obtaining module 303 is configured to obtain a final saliency map of the target image according to the optimal saliency map at each of the scales.
And taking the multidimensional feature of the corresponding area of each super pixel area in the target image as the multidimensional feature corresponding to each super pixel area. The obtaining module 303 calculates the saliency value of each super pixel region by using a global contrast method according to the multi-dimensional features corresponding to each super pixel region. And acquiring a final saliency map of the target image by using the saliency values of the super-pixel region at each scale.
In this embodiment, a target image is divided into a plurality of initial superpixel blocks by using a superpixel division algorithm, an initial saliency map of the target image is obtained according to multidimensional features of each initial superpixel block, each initial superpixel block in the initial saliency map is regarded as a node to construct a map model, the initial saliency map is optimized by using an ant colony optimization algorithm on the basis of the map model, the saliency of the target image is detected according to the optimized initial saliency map, the saliency detection accuracy of a low-light image is improved, and support is provided for problems of night safety monitoring, complex environment target positioning and the like.
On the basis of the foregoing embodiment, the segmentation module in this embodiment is further configured to: extracting color features, texture features, direction features and gradient features of the target image; wherein the color features of the target image comprise color features of the target image in LAB, HSV and YCbCr spaces.
On the basis of the foregoing embodiment, the segmentation module in this embodiment is specifically configured to: respectively determining initial seed points in the target image according to the number of the plurality of preset super pixels; updating the initial seed points according to the gradient value of each pixel point in the neighborhood of each initial seed point to obtain new seed points; and taking pixels in a preset range with each new seed point as a center as neighborhood pixels of each new seed point, distributing a class label to the neighborhood pixels of each new seed point according to the distance between each new seed point and the neighborhood pixels of each new seed point, and taking the pixels with the same class label as an initial super-pixel block.
On the basis of the foregoing embodiment, the segmentation module in this embodiment is further configured to: calculating one-dimensional entropy of each feature map of the target image; taking the feature maps with the maximum one-dimensional entropy of the target image as target feature maps in preset numbers; and acquiring the significant value of each initial superpixel block based on a global contrast method according to the target feature map of each initial superpixel block.
On the basis of the foregoing embodiment, the optimization module in this embodiment is specifically configured to: calculating the distance between any two of the initial superpixel blocks; comparing the distance between any two initial superpixel blocks with a preset threshold value to obtain the information content of a path between any two initial superpixel blocks; calculating the probability of the ant selecting the path according to the information content of the path between any two initial superpixel blocks; when the ants move according to the probability, the information quantity on the path is updated, and the probability that the ants select the path is calculated according to the updated information quantity of the path until a preset termination condition is reached; and if the probability of the ant selecting path in each iteration is greater than that in the first iteration, updating the initial superpixel block corresponding to the path selected by the ant, and taking the superpixel obtained after updating as a superpixel area.
On the basis of the foregoing embodiment, in this embodiment, the step of calculating, by the optimization module, the probability that an ant selects a path between any two initial superpixel blocks according to the information amount of the path specifically includes:
where k denotes the ant number, I and j denote the numbers of any two initial superpixel blocks, Iij(t) denotes the initial superpixel block s at time tiTo the initial super-pixel block sjηij＝1/dijRepresenting a heuristic function, dijIs s isiTo sjDistance of (d), omega(s)i) Denotes siα, information heuristics, β, heuristics,
on the basis of the foregoing embodiments, the obtaining module in this embodiment is specifically configured to: fusing the optimal saliency maps under each scale; and filtering the fused saliency map to obtain a final saliency map of the target image.
The embodiment provides an electronic device, and fig. 4 is a schematic view of an overall structure of the electronic device according to the embodiment of the present invention, where the electronic device includes: at least one processor 401, at least one memory 402, and a bus 403; wherein the content of the first and second substances,
the processor 401 and the memory 402 communicate with each other via a bus 403;
the memory 402 stores program instructions executable by the processor 401, and the processor calls the program instructions to perform the methods provided by the above method embodiments, for example, the methods include: extracting a plurality of features from a target image, generating a feature map, respectively carrying out segmentation of a plurality of scales on the target image based on a superpixel segmentation algorithm to obtain initial superpixel blocks, and obtaining a significant value of each initial superpixel block according to each feature map and each initial superpixel block; generating initial saliency maps under the scales according to the saliency values of the initial super-pixel blocks, and optimizing the initial saliency maps based on an ant colony optimization algorithm to obtain optimal saliency maps under the scales; and acquiring a final saliency map of the target image according to the optimal saliency maps under the scales.
The present embodiments provide a non-transitory computer-readable storage medium storing computer instructions that cause a computer to perform the methods provided by the above method embodiments, for example, including: extracting a plurality of features from a target image, generating a feature map, respectively carrying out segmentation of a plurality of scales on the target image based on a superpixel segmentation algorithm to obtain initial superpixel blocks, and obtaining a significant value of each initial superpixel block according to each feature map and each initial superpixel block; generating initial saliency maps under the scales according to the saliency values of the initial super-pixel blocks, and optimizing the initial saliency maps based on an ant colony optimization algorithm to obtain optimal saliency maps under the scales; and acquiring a final saliency map of the target image according to the optimal saliency maps under the scales.
Those of ordinary skill in the art will understand that: all or part of the steps for implementing the method embodiments may be implemented by hardware related to program instructions, and the program may be stored in a computer readable storage medium, and when executed, the program performs the steps including the method embodiments; and the aforementioned storage medium includes: various media that can store program codes, such as ROM, RAM, magnetic or optical disks.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (9)
1. An image saliency detection method characterized by comprising:
extracting a plurality of features from a target image, generating a feature map, respectively carrying out segmentation of a plurality of scales on the target image based on a superpixel segmentation algorithm to obtain initial superpixel blocks, and obtaining a significant value of each initial superpixel block according to each feature map and each initial superpixel block;
generating initial saliency maps under the scales according to the saliency values of the initial super-pixel blocks, and optimizing the initial saliency maps based on an ant colony optimization algorithm to obtain optimal saliency maps under the scales;
acquiring a final saliency map of the target image according to the optimal saliency map under each scale;
wherein, each initial saliency map is optimized, and the step of obtaining the optimal saliency map under each scale specifically comprises:
calculating the distance between any two of the initial superpixel blocks;
comparing the distance between any two initial superpixel blocks with a preset threshold value to obtain the information content of a path between any two initial superpixel blocks;
calculating the probability of the ant selecting the path according to the information content of the path between any two initial superpixel blocks;
when the ants move according to the probability, the information quantity on the path is updated, and the probability that the ants select the path is calculated according to the updated information quantity of the path until a preset termination condition is reached;
and if the probability of the ant selecting path in each iteration is greater than that in the first iteration, updating the initial superpixel block corresponding to the path selected by the ant.
2. The method according to claim 1, wherein the step of extracting a plurality of features from the target image specifically comprises:
extracting color features, texture features, direction features and gradient features of the target image; wherein the color features of the target image comprise color features of the target image in LAB, HSV and YCbCr spaces.
3. The method according to claim 1, wherein the target image is segmented at a plurality of scales based on a superpixel segmentation algorithm, and the step of obtaining the initial superpixel block specifically comprises:
respectively determining initial seed points in the target image according to the number of the plurality of preset super pixels;
updating the initial seed points according to the gradient value of each pixel point in the neighborhood of each initial seed point to obtain new seed points;
and taking pixels in a preset range with each new seed point as a center as neighborhood pixels of each new seed point, distributing a class label to the neighborhood pixels of each new seed point according to the distance between each new seed point and the neighborhood pixels of each new seed point, and taking the pixels with the same class label as an initial super-pixel block.
4. The method according to claim 1, wherein the step of deriving from each of said feature maps and each of said initial superpixel blocks a saliency value for each of said initial superpixel blocks comprises:
calculating one-dimensional entropy of each feature map of the target image;
taking the feature maps with the maximum one-dimensional entropy of the target image as target feature maps in preset numbers;
and acquiring the significant value of each initial superpixel block based on a global contrast method according to the target feature map of each initial superpixel block.
5. The method as claimed in claim 1, wherein the step of calculating the probability of ants selecting a path between any two initial superpixel blocks according to the information amount of the path comprises:
where k denotes the ant number, I and j denote the numbers of any two initial superpixel blocks, Iij(t) denotes the initial superpixel block s at time tiTo the initial super-pixel block sjηij＝1/dijRepresenting a heuristic function, dijIs s isiTo sjDistance ofFrom, omega(s)i) Denotes siα represents an information heuristic, β represents the impact of heuristic information.
6. The method according to any one of claims 1 to 5, wherein the step of obtaining the final saliency map of the target image from the optimal saliency map at each of the scales specifically comprises:
fusing the optimal saliency maps under each scale;
and filtering the fused saliency map to obtain a final saliency map of the target image.
7. An image saliency detection apparatus characterized by comprising:
the segmentation module is used for extracting a plurality of features from a target image, generating a feature map, respectively segmenting the target image in a plurality of scales based on a superpixel segmentation algorithm, acquiring initial superpixel blocks, and acquiring a significant value of each initial superpixel block according to each feature map and each initial superpixel block;
the optimization module is used for generating initial saliency maps under all scales according to the saliency values of all the initial super-pixel blocks, optimizing all the initial saliency maps based on an ant colony optimization algorithm and obtaining optimal saliency maps under all the scales;
the acquisition module is used for acquiring a final saliency map of the target image according to the optimal saliency map under each scale;
wherein, the optimization module is specifically configured to:
calculating the distance between any two of the initial superpixel blocks;
comparing the distance between any two initial superpixel blocks with a preset threshold value to obtain the information content of a path between any two initial superpixel blocks;
calculating the probability of the ant selecting the path according to the information content of the path between any two initial superpixel blocks;
when the ants move according to the probability, the information quantity on the path is updated, and the probability that the ants select the path is calculated according to the updated information quantity of the path until a preset termination condition is reached;
and if the probability of the ant selecting path in each iteration is greater than that in the first iteration, updating the initial superpixel block corresponding to the path selected by the ant, and taking the superpixel obtained after updating as a superpixel area.
8. An electronic device, comprising:
at least one processor, at least one memory, and a bus; wherein the content of the first and second substances,
the processor and the memory complete mutual communication through the bus;
the memory stores program instructions executable by the processor, the processor invoking the program instructions to perform the method of any of claims 1 to 6.
9. A non-transitory computer-readable storage medium storing computer instructions that cause a computer to perform the method of any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811355398.5A CN109685806B (en) | 2018-11-14 | 2018-11-14 | Image significance detection method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811355398.5A CN109685806B (en) | 2018-11-14 | 2018-11-14 | Image significance detection method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109685806A CN109685806A (en) | 2019-04-26 |
| CN109685806B true CN109685806B (en) | 2020-06-30 |
Family
ID=66184682
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811355398.5A Expired - Fee Related CN109685806B (en) | 2018-11-14 | 2018-11-14 | Image significance detection method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109685806B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110070539A (en) * | 2019-04-28 | 2019-07-30 | 重庆大学 | Image quality evaluating method based on comentropy |
| CN111242941B (en) * | 2020-01-20 | 2023-05-30 | 南方科技大学 | Salient region detection method and device based on visual attention |
| CN111583279A (en) * | 2020-05-12 | 2020-08-25 | 重庆理工大学 | Super-pixel image segmentation method based on PCBA |
| CN114332633B (en) * | 2022-03-01 | 2022-06-10 | 北京化工大学 | Radar image target detection and identification method and equipment and storage medium |
| CN114581207B (en) * | 2022-05-09 | 2022-09-13 | 农夫铺子发展集团有限公司 | Commodity image big data accurate pushing method and system for E-commerce platform |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930542A (en) * | 2012-10-31 | 2013-02-13 | 电子科技大学 | Detection method for vector saliency based on global contrast |
| CN103400129A (en) * | 2013-07-22 | 2013-11-20 | 中国科学院光电技术研究所 | Target tracking method based on frequency domain saliency |
| CN107527348A (en) * | 2017-07-11 | 2017-12-29 | 湖州师范学院 | Conspicuousness detection method based on multi-scale division |
| CN108345892A (en) * | 2018-01-03 | 2018-07-31 | 深圳大学 | A kind of detection method, device, equipment and the storage medium of stereo-picture conspicuousness |
-
2018
- 2018-11-14 CN CN201811355398.5A patent/CN109685806B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930542A (en) * | 2012-10-31 | 2013-02-13 | 电子科技大学 | Detection method for vector saliency based on global contrast |
| CN103400129A (en) * | 2013-07-22 | 2013-11-20 | 中国科学院光电技术研究所 | Target tracking method based on frequency domain saliency |
| CN107527348A (en) * | 2017-07-11 | 2017-12-29 | 湖州师范学院 | Conspicuousness detection method based on multi-scale division |
| CN108345892A (en) * | 2018-01-03 | 2018-07-31 | 深圳大学 | A kind of detection method, device, equipment and the storage medium of stereo-picture conspicuousness |
Non-Patent Citations (1)
| Title |
|---|
| Particle Swarm Optimization Based Salient Object Detection for Low Contrast Images;Nan Mu et al.;《Springer International Publishing AG 2017》;20171231;第602-612页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109685806A (en) | 2019-04-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109685806B (en) | Image significance detection method and device | |
| US9042648B2 (en) | Salient object segmentation | |
| CN105184763B (en) | Image processing method and device | |
| CN109522908A (en) | Image significance detection method based on area label fusion | |
| CN109919944B (en) | Combined superpixel graph-cut optimization method for complex scene building change detection | |
| CN110866896B (en) | Image saliency target detection method based on k-means and level set super-pixel segmentation | |
| JP5939056B2 (en) | Method and apparatus for positioning a text region in an image | |
| CN109993753B (en) | Method and device for segmenting urban functional area in remote sensing image | |
| CN106157330B (en) | Visual tracking method based on target joint appearance model | |
| CN108022244B (en) | Hypergraph optimization method for significant target detection based on foreground and background seeds | |
| US20170178341A1 (en) | Single Parameter Segmentation of Images | |
| Feng et al. | A color image segmentation method based on region salient color and fuzzy c-means algorithm | |
| WO2019197021A1 (en) | Device and method for instance-level segmentation of an image | |
| CN109636809B (en) | Image segmentation level selection method based on scale perception | |
| JP6426441B2 (en) | Density measuring device, density measuring method, and program | |
| Zhang et al. | Saliency detection via local structure propagation | |
| CN109658378B (en) | Pore identification method and system based on soil CT image | |
| Li et al. | Unsupervised road extraction via a Gaussian mixture model with object-based features | |
| CN110705634A (en) | Heel model identification method and device and storage medium | |
| CN111222546B (en) | Multi-scale fusion food image classification model training and image classification method | |
| CN111274964B (en) | Detection method for analyzing water surface pollutants based on visual saliency of unmanned aerial vehicle | |
| CN108986210B (en) | Method and device for reconstructing three-dimensional scene | |
| CN115019163A (en) | City factor identification method based on multi-source big data | |
| Geetha et al. | An improved method for segmentation of point cloud using minimum spanning tree | |
| US9607398B2 (en) | Image processing apparatus and method of controlling the same |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200630 Termination date: 20201114 |