CN104504692B - The extracting method of notable object in image based on region contrast - Google Patents

Abstract

The invention discloses a method for extracting salient objects in an image based on regional contrast, and its specific steps are as follows: (1), input the original image, and the saliency mark of the original image is , the object probability map of the original image is ; (2), calculate the fusion coefficient of the saliency map and the object probability map; (3), calculate the regional contrast fusion map according to the fusion coefficient, and extract the salient objects in the image. This method uses the saliency map and the object probability map to calculate the fusion map of the regional contrast map, and extracts the salient objects in the image. significant object.

Description

Extraction Method of Salient Objects in Images Based on Region Contrast

technical field

The invention relates to the technical fields of computer information and image processing, in particular to a method for extracting salient objects in an image.

Background technique

According to the research of psychology and human vision, when people observe an image, they do not pay equal attention to each area of the image, thus producing a saliency map corresponding to the degree of attention. In most cases, when people observe an image, they will not distribute attention evenly on the entire image, but will focus on an object in the image, and such an object is called a salient object. If the salient objects can be extracted automatically, it will be of great help to image scaling, image recognition, image retrieval and other application fields. The salient object extraction method was applied in this context. It aims to accurately extract the salient objects in the image after the background is removed. For example, Rother et al. published "Grabcut : Graph Cutting Method for Interactive Foreground Extraction", this paper uses artificially animated rectangular windows to designate candidate salient object regions, and then uses graph cutting method to extract salient objects. However, the wide application of this method is limited because it needs to be manually specified and only rectangular windows can be used to define salient object candidate regions. Cheng et al. published the paper "Saliency Region Detection Based on Global Contrast" at the IEEE Computer Vision and Pattern Recognition Conference in 2011. This paper uses global color contrast and spatial region contrast to obtain a saliency map, and then according The iterative Grabcut graph cutting method is used to segment the image and extract the salient objects in the image. The specific steps of the method are as follows:

(1) The saliency value of a pixel is defined by the color contrast between the pixel and other pixels in the image, and pixels with the same color are assigned the same saliency value.

(2) Colors with less frequent occurrence are discarded based on histogram statistics, and the significance value of each color is replaced by the weighted average of similar color significance values.

(3) Use a graph-based image segmentation method to divide the image into several regions, and use the Euclidean distance between the centers of gravity of the two regions to calculate the spatial region contrast to obtain a saliency map.

(4) Take a fixed threshold for the saliency map, and use the Grabcut map cutting method for image segmentation.

(5) Use dilation and erosion to manipulate the resulting image after image segmentation to obtain a new image to be segmented, and then use the Grabcut image cutting method to perform image segmentation.

(6) Repeat step (5) until convergence. The final result map is obtained, that is, the extracted salient objects.

Liu et al. published the article "Saliency Tree: A New Saliency Detection Framework" in the Image Processing Journal of the Institute of Electrical and Electronics Engineers in 2014. This paper uses tree-shaped nodes to represent small areas in the image. By measuring the global contrast, spatial sparsity and object priority, merging the original small regions to generate a saliency map, and finally using the maximum between-class variance value to binarize the saliency map to extract salient objects in the image. This method improves the accuracy of the saliency map, however, the maximum inter-class variance value method in this method still cannot fully extract multiple salient objects in the image when extracting salient objects. Alexe et al published "Measuring Objects in Image Window" in the Journal of Pattern Analysis and Machine Intelligence of the Institute of Electrical and Electronics Engineers in 2012. This paper proposed the concept and calculation method of detecting objects with image windows, that is, rectangular windows. By calculating the probability that a large number of rectangular windows contain objects, the Bayesian formula combined with multiple clues is used to obtain the location probability of the area where the salient objects are located, and the object probability map is obtained. The concrete steps of this method are as follows:

(1) Use the frequency-domain residual method to obtain multi-scale saliency cues, and generate a large number of rectangular windows.

(2) Use the chi-square distance of the color space histogram to calculate the color contrast clues between rectangular windows.

(3) Use the Canny operator to detect the boundary and get the edge density clue.

(4) Use a graph-based image segmentation method to segment the image into several regions, and obtain regional bifurcation clues according to the minimum regional difference between the region inside the rectangular window and the region outside the rectangular window.

(5) Gaussian distribution is used to estimate the location and size clues of the rectangular window.

(6) Use the Bayesian formula to integrate the clues obtained from steps (1)-(5), calculate the location probability of the area where the salient object is located, and obtain the object probability map.

but. The disadvantage of the above method is that this method only uses a rectangular window to represent the position probability of the salient object, but does not contain accurate contour information of the salient object, and cannot accurately extract the salient object in the image.

To sum up, the existing methods for extracting salient objects in images cannot accurately and completely extract salient objects in images, which affects the wide application of salient object extraction.

Contents of the invention

The object of the present invention is to propose a method for extracting salient objects in an image based on regional contrast, which can accurately and completely extract the salient objects in the image.

In order to achieve the above object, the technical scheme adopted in the present invention is as follows:

A method for extracting salient objects in an image based on regional contrast, the specific steps are as follows:

(1), input the original image, the saliency map of the original image is marked as , the object probability map of the original image is ;

(2), calculating the fusion coefficient of the saliency map and the object probability map;

(3) According to the fusion coefficient, the regional contrast fusion map is calculated, and the salient objects in the image are extracted.

The fusion coefficient of the described calculation saliency map and object probability map of above-mentioned steps (2), its concrete steps are as follows:

(2-1), set the first threshold value map of the saliency map , the second threshold map of the saliency map , the third threshold map of the saliency map , and the threshold map of the object probability map is , which are as follows:

The saliency value of the pixels in the saliency map is normalized, and its normalized value is [0,1]. Among them, three different thresholds are set, respectively, the first threshold is recorded as , = 0.75; set the second threshold as the maximum between-class variance value of the significance map, denoted as ; Let the third threshold be the average saliency value of the saliency map, denoted as ; In the saliency map, the pixel value of the pixel whose saliency value is greater than or equal to the above threshold is recorded as 1, and the pixel value of the pixel smaller than the above threshold is recorded as 0, and the threshold map corresponding to the saliency map is obtained,

Set and the first threshold The corresponding threshold map is the first threshold map of the saliency map, denoted as , set with the second threshold The corresponding threshold map is the second threshold map of the saliency map, denoted as , set with the third threshold The corresponding threshold map is the third threshold map of the saliency map, denoted as , the object probability map Medium object probability value is greater than or equal to the first threshold The pixel value of the pixel is recorded as 1, which is less than the first threshold The pixel value of the pixel is recorded as 0, and the threshold map of the object probability map is obtained ;

(2-2), the overlap ratio value of the pixels between the first threshold map of the saliency map and the second threshold map of the saliency map and the first threshold map of the saliency map and the third threshold map of the saliency map The sum of the coincidence ratio values of the pixels between is denoted as , whose calculation formula is:

Among them, p represents the pixel in the threshold map, represents a pixel belonging to the first threshold map of the saliency map, is the pixel value corresponding to pixel p in the first threshold map of the saliency map, represents a pixel belonging to the second threshold map of the saliency map, is the pixel value corresponding to pixel p in the second threshold value map of the saliency map, represents a pixel belonging to the third threshold map of the saliency map, is the pixel value corresponding to pixel p in the third threshold map of the saliency map;

(2-3), calculate the overlapping area of the first threshold map of the saliency map and the threshold map of the object probability map, denoted as , whose calculation formula is:

(2)

in, express and the intersection of express and the union of represents the overlap area of the first threshold map of the saliency map and the threshold map of the object probability map;

(2-4), set the diagonal length of the original image, denoted as D, and calculate the normalized centroid distance between the centroid point of the first threshold map of the saliency map and the centroid point of the threshold map of the object probability map, recorded as , whose calculation formula is:

(3)

in, represents the centroid point of the first threshold map of the saliency map, A threshold map representing the object detection map the centroid point of is calculation arrive Euclidean distance between;

(2-5), according to the sum of the overlapping ratio values of pixels between the threshold maps, the overlapping area of the first threshold map of the saliency map and the threshold map of the object probability map, and the The centroid distance value between the centroid point of the first threshold value map of the above-mentioned saliency map and the centroid point of the threshold value map of the object probability map is used to calculate the fusion coefficient of the saliency map and the object probability map, which is denoted as , its calculation formula is:

(4)

According to the fusion coefficient described in the above-mentioned steps (3), calculate the regional contrast fusion map, extract the salient object in the image, and its specific steps are as follows:

(3-1), according to the fusion coefficient of the saliency map and the object probability map, the saliency map and the object probability map are fused, and the regional contrast fusion map is calculated, which is denoted as , whose calculation formula is:

(5)

in is the saliency map, express Pick the largest number between and 1, , exp represents the e index, Indicates that the pixel probability value in the object probability map is greater than or equal to The pixel value of the pixel, Indicates that the pixel probability value in the object probability map is less than or equal to The pixel value of the pixel, Indicates that the pixel probability value in the object probability map is between , The pixel values of the pixels between, Represents the region contrast fusion map;

(3-2), set the fourth threshold to 0.2, regional contrast fusion map The pixel value of the pixel greater than or equal to the fourth threshold is recorded as 1, and the pixel value of the pixel smaller than the fourth threshold is recorded as 0, and the threshold value map corresponding to the regional contrast fusion map is obtained, and then the Grabcut image cutting method is used for image segmentation, and the image is extracted notable objects.

Compared with the existing technology, the method for extracting salient objects in an image based on regional contrast has the following advantages: the method uses a saliency map and an object probability map to calculate a regional contrast fusion map to extract salient objects in an image. A single method of extracting salient objects using a saliency map or an object probability map can more accurately and completely extract salient objects in an image.

Description of drawings

Fig. 1 is the flow chart of the extraction method of salient object in the image based on regional contrast of the present invention;

Figure 2(a) is the input original image;

Figure 2(b) is the saliency map of the original image;

Figure 2(c) is the object probability map of the original image;

Figure 3(a) is a regional contrast fusion map;

Figure 3(b) is the salient objects extracted after the threshold map of the region contrast fusion map is cut with the Grabcut map.

detailed description

Embodiments of the present invention will be described in further detail below in conjunction with the accompanying drawings.

The simulation experiment carried out by the present invention is programmed on a PC test platform with a CPU of 3.5GHz and a memory of 16G.

As shown in Figure 1, the method for extracting salient objects in an image based on regional contrast of the present invention, its specific steps are as follows:

(1), input the original image, the saliency map and the object probability map of the original image, the specific steps are as follows:

Input the original image, as shown in Figure 2(a), the saliency map of the original image is denoted as , as shown in Figure 2(b), the object probability map of the original image, denoted as , as shown in Figure 2(c);

(2), calculate the fusion coefficient of the saliency map and the object probability map, the specific steps are as follows:

(2-1), set the first threshold value map of the saliency map , the second threshold map of the saliency map , the third threshold map of the saliency map , the threshold map of the object probability map , which are as follows:

The saliency value of the pixels in the saliency map is normalized, and its normalized value is [0,1]. Among them, three different thresholds are set, respectively, the first threshold is recorded as , = 0.75; set the second threshold as the maximum between-class variance value of the significance map, denoted as ; Let the third threshold be the average saliency value of the saliency map, denoted as ; In the saliency map, the pixel value of the pixel whose saliency value is greater than or equal to the above threshold is recorded as 1, and the pixel value of the pixel smaller than the above threshold is recorded as 0, and the threshold map corresponding to the saliency map is obtained,

Set and the first threshold The corresponding threshold map is the first threshold map of the saliency map, denoted as , set with the second threshold The corresponding threshold map is the second threshold map of the saliency map, denoted as , set with the third threshold The corresponding threshold map is the third threshold map of the saliency map, denoted as ,

Object Probability Map Medium object probability value is greater than or equal to the first threshold The pixel value of the pixel is recorded as 1, which is less than the first threshold The pixel value of the pixel is recorded as 0, and the threshold map of the object probability map is obtained ;

(2-2), the overlap ratio value of the pixels between the first threshold map of the saliency map and the second threshold map of the saliency map and the first threshold map of the saliency map and the third threshold map of the saliency map The sum of the coincidence ratio values of the pixels between is denoted as , whose calculation formula is:

(1)

Among them, p represents the pixel in the threshold map, represents a pixel belonging to the first threshold map of the saliency map, is the pixel value corresponding to pixel p in the first threshold map of the saliency map, represents a pixel belonging to the second threshold map of the saliency map, is the pixel value corresponding to pixel p in the second threshold value map of the saliency map, represents a pixel belonging to the third threshold map of the saliency map, is the pixel value corresponding to pixel p in the third threshold map of the saliency map;

(2-3), calculate the overlapping area of the first threshold map of the saliency map and the threshold map of the object probability map, denoted as , whose calculation formula is:

(2)

in, express and the intersection of express and the union of is the overlap area of the first threshold map of the saliency map and the threshold map of the object probability map;

(2-4), set the diagonal length of the original image, denoted as D, and calculate the normalized centroid distance between the centroid point of the first threshold map of the saliency map and the centroid point of the threshold map of the object probability map, recorded as , whose calculation formula is:

(3)

in, represents the centroid point of the first threshold map of the saliency map, A threshold map representing the object detection map the centroid point of is calculation arrive Euclidean distance between;

(2-5), according to the sum of the overlapping ratio values of pixels between the threshold maps, the overlapping area of the first threshold map of the saliency map and the threshold map of the object probability map, and the The centroid distance value between the centroid point of the first threshold value map of the above-mentioned saliency map and the centroid point of the threshold value map of the object probability map is used to calculate the fusion coefficient of the saliency map and the object probability map, which is denoted as , its calculation formula is:

(4)

(3), according to the fusion coefficient, calculate the regional contrast fusion map, and extract the salient objects in the image, the specific steps are as follows:

(3-1), according to the threshold fusion coefficient of the saliency map and the object probability map, the saliency map and the object probability map are fused to calculate the regional contrast fusion map, denoted as , whose calculation formula is:

(5)

in, is the saliency map, express Pick the largest number between and 1, , exp represents the e index, Indicates that the pixel probability value in the object probability map is greater than or equal to The pixel value of the pixel, Indicates that the pixel probability value in the object probability map is less than or equal to The pixel value of the pixel, Indicates that the pixel probability value in the object probability map is between , The pixel values of the pixels between, Represents the regional contrast fusion map, as shown in Figure 3(a), the regional contrast fusion map in Figure 3(a) highlights the torso of the entire salient object more completely than the saliency map in Figure 2(b), Figure 3(a) The region-contrast fusion map in 3(a) more accurately delineates the outline of the entire salient object compared to the object probability map in Fig. 2(c);

(3-2), set the fourth threshold to 0.2, regional contrast fusion map The pixel value of the pixel greater than or equal to the fourth threshold is recorded as 1, and the pixel value of the pixel smaller than the fourth threshold is recorded as 0, and the threshold value map corresponding to the regional contrast fusion map is obtained, and then the Grabcut image cutting method is used for image segmentation, and the image is extracted The salient objects of , as shown in Figure 3(b).

It can be seen from the above simulation experiment results that the method of the present invention utilizes the saliency map and the object probability map to obtain a regional contrast fusion map, and extract salient objects from the regional contrast fusion map, which can be more accurate and complete.

Claims (2)

1. an extraction method of salient object in the image based on regional contrast, it is characterized in that, its concrete steps are as follows: (1), input the original image, the saliency map of the original image is marked as S, and the object probability map of the original image is marked as O; (2), calculate the fusion coefficient of the saliency map and the object probability map, and its concrete steps are as follows: (2-1), let the first threshold map S _f of the saliency map, the second threshold map S _o of the saliency map, the third threshold map S _a of the saliency map, and the threshold map of the object probability map be O _f , The details are as follows: The saliency value of the pixels in the saliency map is normalized, and its normalized value is [0,1], where three different thresholds are set, respectively, the first threshold is denoted as T _f , T _f =0.75; set the second threshold as the maximum between-class variance value of the significance map, denoted as T _o ; set the third threshold as the average significance value of the significance map, denoted as T _a ; the significance value of the significance map The pixel value of the pixel greater than or equal to the above threshold is recorded as 1, and the pixel value of the pixel smaller than the above threshold is recorded as 0, and the threshold map corresponding to the saliency map is obtained, Let the threshold map corresponding to the first threshold T _f be the first threshold map of the saliency map, denoted as S _f , let the threshold map corresponding to the second threshold T _o be the second threshold map of the saliency map, denoted as S _o , let the threshold map corresponding to the third threshold T _a be the third threshold map of the saliency map, denoted as S _a , and the pixel value of the pixel whose object probability value is greater than or equal to the first threshold T _f in the object probability map O is denoted as 1, the pixel value of the pixel smaller than the first threshold T _f is recorded as 0, and the threshold map O _f of the object probability map is obtained; (2-2), the overlap ratio value of the pixels between the first threshold map of the saliency map and the second threshold map of the saliency map and the first threshold map of the saliency map and the third threshold map of the saliency map The sum of the coincidence ratio values of the pixels between is denoted as Q(S), and its calculation formula is: $\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; S</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}}{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}}{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; +</span>\frac{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}}{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; a</span>}}}{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; a</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$ Among them, p represents the pixel in the threshold map, p ∈ S _f represents the pixel in the first threshold map belonging to the saliency map, and S _f (p) is the pixel value corresponding to the pixel p in the first threshold map of the saliency map , p∈S _o represents the pixel belonging to the second threshold map of the saliency map, S _o (p) is the pixel value corresponding to the pixel p in the second threshold map of the saliency map, and p∈S _a represents the pixel belonging to the saliency map The pixel in the third threshold map of the figure, S _a (p) is the pixel value corresponding to the pixel p in the third threshold map of the saliency map; (2-3), calculate the overlapping area of the first threshold map of the saliency map and the threshold map of the object probability map, denoted as r(S _f , O _f ), and its calculation formula is: $\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; o</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; \&cap;</span>{\mathrm{<span\; class="notranslate">\; o</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}}{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; \&cup;</span>{\mathrm{<span\; class="notranslate">\; o</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$ Among them, S _f ∩ O _f represents the intersection of S _f and O _f , S _f ∪ O _f represents the union of S _f and O _f , r(S _f , O _f ) represents the first threshold map of the saliency map and the object the overlap area of the threshold map of the probability map; (2-4), set the diagonal length of the original image, denoted as D, and calculate the normalized centroid distance between the centroid point of the first threshold map of the saliency map and the centroid point of the threshold map of the object probability map, It is recorded as d(S _f ,O _f ), and its calculation formula is: $\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; o</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; S</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; D.</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$ Among them, μ _S denotes the centroid point of the first threshold map of the saliency map, μ _o denotes the centroid point of the threshold map O _f of the object probability map, and ||·|| ₂ is the Euclidean distance between μ _S and μ _o ; (2-5), according to the sum of the overlapping ratio values of the pixels between the threshold maps, the overlapping area of the first threshold map of the saliency map and the threshold map of the object probability map, the salience The centroid distance value between the centroid point of the first threshold value map of the property map and the centroid point of the threshold value map of the object probability map is used to calculate the saliency map and the fusion coefficient of the object probability map, which is denoted as IS, and its calculation formula is: $\mathrm{<span\; class="notranslate">\; I</span>}\mathrm{<span\; class="notranslate">\; S</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 4</span>}}<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; S</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; o</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; \&lsqb;</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; o</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; \}</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ;</span>$ (3) According to the fusion coefficient, the regional contrast fusion map is calculated, and the salient objects in the image are extracted. 2. the method for extracting salient objects in the image based on regional contrast according to claim 1, is characterized in that, according to fusion coefficient described in above-mentioned step (3), calculates regional contrast fusion map, extracts the salient object in the image, The specific steps are as follows: (3-1), according to the fusion coefficient of the saliency map and the object probability map, the saliency map and the object probability map are fused to calculate the regional contrast fusion map, which is denoted as S _fusion , and its calculation formula is: S _fusion ＝S+α _f O(p|O(p)≥T _f )+α _b O(p|O(p)≤T _b )+O(p|T _b <O(p)<T _f ) (5) where S is the saliency map, T _f is the first threshold in the above step (2-1), T _a is the third threshold in the above step (2-1), IS is the Fusion coefficient of saliency map and object probability map, α _f =max(α _b +0.7,1) means α _f takes the maximum number between α _b +0.7 and 1, α _b =exp(-IS), exp means e Index, O(p|O(p)≥T _f ) represents the pixel value of the pixel whose probability value is greater than or equal to T _f in the object probability map, O(p|O(p)≤T _b ) represents the pixel in the object probability map The pixel value of the pixel whose probability value is less than or equal to T _b , O(p|T _b <O(p)<T _f ) means the pixel value of the pixel whose probability value of the pixel in the object probability map is between T _f and T _b , S _fusion represents the regional contrast fusion map; (3-2), assuming that the fourth threshold is 0.2, the pixel value of the pixel in the regional contrast fusion map S _fusion greater than or equal to the fourth threshold is recorded as 1, and the pixel value of the pixel smaller than the fourth threshold is recorded as 0, and the regional contrast fusion is obtained The threshold value map corresponding to the graph, and then use the Grabcut graph cutting method to segment the image and extract the salient objects in the image.