CN107886507A - A kind of salient region detecting method based on image background and locus - Google Patents

Abstract

The present invention provides a salient area detection method based on the image background and spatial position, the method mainly includes the following steps: performing M scale superpixel layered segmentation on the target image to obtain M layer target sub-images; extracting superpixel block feature vectors , extract the color features and texture features of superpixel blocks; cluster background superpixel blocks; use the spatial position and the difference with the background superpixel to calculate the saliency of each superpixel block; multi-scale superpixel block saliency fusion. The advantages are: the method of the present invention can accurately judge the significant area of the image, the performance effect is good, the accuracy and calculation efficiency of the detection can be effectively improved, and the technology can be provided for the screening and analysis of massive images or video data applied to the Internet and cloud computing support.

Description

A Salient Region Detection Method Based on Image Background and Spatial Position

technical field

The invention belongs to the technical field of image salient area detection, and in particular relates to a salient area detection method based on image background and spatial position.

Background technique

From the perspective of promoting the development of advanced intelligent robots, salient area detection can enable intelligent robots to select the most relevant part of the current task from a large amount of video data received at the same time for processing. This can effectively simulate the directivity and concentration of human visual perception, laying the foundation for the completion of intelligent tasks. From the perspective of promoting intelligent applications in the visual field, applying the salient area detection method to the screening and analysis of massive images or video data from the Internet and cloud computing can effectively improve the accuracy and computational efficiency of detection; In the field of video surveillance, it can provide key area marks in the early stage for algorithms such as target recognition and hot spot tracking, and improve the calculation efficiency of related algorithms; when applied to the field of image or video transmission, it can compress key areas on images or videos in a targeted manner. Improve the efficiency of image or video transmission. In addition, the salient area detection method can also be widely used in path navigation, unmanned aerial vehicle and other fields.

In recent years, many scholars have proposed many methods for detecting salient regions or objects in images. In order to improve computational efficiency and ignore some unnecessary details in the image, most of these methods first extract the perceptually homogeneous elements of the image, such as superpixels, regions (of course there are methods that directly use pixels), and then calculate their local contrast, global Contrastive or sparse noise is used to obtain saliency values for each perceptually homogenous element, which is finally integrated to segment the entire salient object. From the research trend in recent years, compared with local contrast, global cues have attracted more attention because they can assign contrastive saliency values on similar image regions.

The patent titled "A Video Salient Area Detection Method and System" with the publication number CN 104424642 A discloses a video salient area detection method and system, by obtaining pixel-level static salient features and local area-level Static saliency features, local region-level dynamic saliency features, global-level static saliency features and global-level dynamic saliency features, using the correlation between video frames to modulate the video saliency features, based on the modulated 3D-MRF is used to set the video saliency region of the video frame, and then the Graph-cuts is used to select the optimal video saliency region to segment the video saliency region. This method uses the complementary prior of salient region detection to improve the performance of the algorithm, but when the border region of the image cannot describe the background well, such as when the characteristics of the border region are quite different, the whole border is put together to calculate the background feature. The method does not compute the background features accurately.

The patent titled "A Detection Method of a Significant Region" discloses a detection method of a significant region, which defines the basic elements involved in the difference comparison calculation as a region, so that it is in the same order of magnitude as the final detection result , thus improving the efficiency of salient region detection. However, this invention only applies local contrast such as color space conversion and image segmentation, and the effect is not good when the image target is not obvious.

The patent titled "A Deep Learning Image Salient Area Detection Method" discloses a deep learning image salient area detection method. By combining the results of different network layers under deep learning, the image is obtained at different scales. features, so as to get better detection performance; at the same time, image segmentation is used for superpixel threshold learning. However, the method proposed by the invention is affected by the image category (complex background or simple background, containing a single target or multiple targets) and quantity of its training set. This method is prone to the risk of over-application, and may show not good.

It can be seen that the above-mentioned various image salient region detection methods all have certain application limitations, resulting in a low detection accuracy and an overly complex detection algorithm.

Contents of the invention

Aiming at the defects in the prior art, the present invention provides a salient area detection method based on image background and spatial position, which can effectively solve the above problems.

The technical scheme that the present invention adopts is as follows:

The present invention provides a salient area detection method based on image background and spatial position, comprising the following steps:

Step 1: Carry out hierarchical segmentation of superpixels of M scales on the target image, where M is the total number of layers of scales to obtain M-layer target sub-images; each layer of target sub-images is composed of multiple super-pixel blocks;

Step 2, for each layer of target sub-images, perform the following steps 2.1-step 2.3:

Step 2.1, extracting the feature vector of each superpixel block in the target sub-image to obtain the feature vector of the superpixel block;

In step 2.2, the frame area of the target sub-image is regarded as the image background, and the superpixel blocks belonging to the image background are called background superpixel blocks;

Cluster the background superpixel blocks to obtain n clusters, which are: the first cluster, the second cluster...the nth cluster; the cluster center eigenvector of the first cluster is B ₁ , the eigenvector of the cluster center of the second cluster is B ₂ , and so on, the eigenvector of the cluster center of the nth cluster is B _n , therefore, the eigenvector of the cluster center B={B ₁ ,B ₂ ,...,B _n };

Step 2.3, for each superpixel block of the target sub-image, denoted as a superpixel block p, the saliency value s of the superpixel block p is calculated by the following formula:

in:

in:

D(p,B _i ) represents the distance between the superpixel block p and the cluster center feature vector B _i of the i-th cluster, i={1,2,...,n}; σ represents the scale factor;

w is the weight, which is used to measure the distance between the superpixel block p and the center point of the target sub-image of this layer, (x, y) represents the coordinates of the center point of the superpixel block p, and (x', y') represents the target of this layer The coordinates of the center point of the sub-image;

The saliency value of each superpixel block of the target sub-image of each layer is thus calculated;

Step 3, saliency fusion of multi-scale superpixel blocks to obtain the final saliency map, and detect salient regions on the saliency map, including:

Step 3.1, calculate the saliency value of any pixel point j on the fused saliency map:

The saliency value s j of pixel point _j is the average value of the saliency values of its corresponding superpixel block at all scales, namely:

Among them: s _l is the saliency value of the superpixel block of the pixel j located in the l-th layer target sub-image;

In step 3.2, the saliency values of all pixel points j form an image saliency map, and on the saliency map, the region exceeding the set threshold is the finally detected salient region.

Preferably, in step 1, the SLIC algorithm is used to perform M scale superpixel hierarchical segmentation on the target image.

Preferably, in step 2.1, extract the feature vector of each superpixel block in the target sub-image, specifically: extract the color feature and texture feature of each superpixel block, the feature vector of each superpixel block includes: RGB average value 3 components of , 256 components of RGB histogram, 3 components of HSV mean, 256 components of HSV histogram, 3 components of Lab mean, 256 components of Lab histogram and LM filter response 48 servings.

Preferably, in step 2.2, the background superpixel blocks are clustered, specifically, an improved K-Means clustering algorithm is used to cluster the background superpixel blocks.

Preferably, the background superpixel blocks are clustered using the improved K-Means clustering algorithm, specifically including:

Step 2.2.1, set the initial clustering number of the improved K-means clustering algorithm as z, that is, the final clustering obtains z clustering numbers;

Step 2.2.2, use the K-means clustering algorithm for initial clustering to obtain several initial clusters; in the initial clustering, the following method is used to calculate the distance between any two superpixel blocks:

For any two superpixel blocks in the target sub-image, they are respectively denoted as superpixel block u and superpixel block v;

Suppose the RGB average value of the superpixel block u in the extracted target sub-image is f ₁ ^u , and the RGB histogram is HSV mean is HSV histogram is Lab average is Lab histogram is The LM filter response is

Let the RGB average value of the superpixel block v in the extracted target sub-image be f ₁ ^v , and the RGB histogram be HSV mean is HSV histogram is Lab average is Lab histogram is The LM filter response is

The distance D(u,v) between superpixel block u and superpixel block v is:

Among them: N(●) means normalization;

Indicates the distance of the a-th feature between superpixel block u and superpixel block v;

Among them, a=1, 3, 5, 7, representing the RGB average feature, HSV average feature, Lab average feature and LM filter response feature; m is the total number of dimensions of each feature, and e is the number of dimensions of each feature Parameters, for the RGB average feature, its dimension is 3; for the HSV average feature, its dimension is 3; for the Lab average feature, its dimension is 3; for the LM filter response feature, its dimension is 48; is the e-th dimension component of the a-th feature of the superpixel block u; is the e-th dimension component of the a-th feature of the superpixel block v;

Indicates the distance of the cth feature between superpixel block u and superpixel block v; where c=2, 4, 6 represent RGB histogram, HSV histogram and Lab histogram respectively; b is the histogram interval number; d is the number parameter of the histogram interval; is the dth histogram value of the cth feature of the superpixel block u; is the dth histogram value of the cth feature of the superpixel block v;

Then calculate the cluster center eigenvector of each initial cluster; average the features of all superpixels in a cluster to obtain the cluster center;

Step 2.2.3, select the Euclidean distance as the similarity measure between the initial clusters, so as to calculate the difference value between the cluster centers;

Step 2.2.4, judge whether the difference between any two cluster centers is smaller than the threshold θ; set the cluster center set to A, then D(g,h) represents the Euclidean distance between the cluster center g and the cluster center h;

Step 2.2.5, if the result of step 2.2.4 is "Yes", then the number N of clusters is reduced by 1, and return to step 2.2.2 for re-clustering;

Step 2.2.6, if the result of step 2.2.4 is "no", go to step 2.2.7;

Step 2.2.7, record the number of clusters and the eigenvector of the cluster center;

In step 2.2.8, the process ends.

A salient region detection method based on image background and spatial position provided by the present invention has the following advantages:

The method of the invention can accurately judge the salient area of the image, has a good performance effect, can effectively improve the detection accuracy and calculation efficiency, and provides technical support for the screening and analysis of massive image or video data applied to the Internet and cloud computing.

Description of drawings

FIG. 1 is a schematic diagram of the overall flow of the salient region detection method based on image background and spatial position provided by the present invention;

Fig. 2 is a schematic diagram of the superpixel layered segmentation results of the salient region detection method based on image background and spatial position provided by the present invention;

Fig. 3 is a kind of method flowchart of improving K-means clustering algorithm provided by the present invention;

Figure 4 is a schematic diagram of the comparison of salient area detection results.

Detailed ways

In order to make the technical problems, technical solutions and beneficial effects solved by the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

The present invention provides a salient area detection method based on the image background and spatial position, the method mainly includes the following steps: performing M scale superpixel layered segmentation on the target image to obtain M layer target sub-images; extracting superpixel block feature vectors , extract the color features and texture features of superpixel blocks; cluster background superpixel blocks; use the spatial position and the difference with the background superpixel to calculate the saliency of each superpixel block; multi-scale superpixel block saliency fusion. The method of the invention can accurately judge the salient area of the image, has a good performance effect, can effectively improve the detection accuracy and calculation efficiency, and provides technical support for the screening and analysis of massive image or video data applied to the Internet and cloud computing.

Based on the salient region detection method of image background and spatial position, the image background superpixels are clustered on each layer of image, and the superpixel feature vector is calculated, and the superpixel is calculated according to the superpixel spatial position and its difference with the background superpixel significance value. Finally, the saliency values of the superpixels on the images of each layer are fused to obtain the final saliency map. Referring to Figure 1, the following steps are included:

Step 1: Carry out hierarchical segmentation of superpixels of M scales on the target image, where M is the total number of layers of scales to obtain M-layer target sub-images; each layer of target sub-images is composed of multiple super-pixel blocks;

In this step, the SLIC (Simple Linear Iterative Cluster) algorithm is specifically used to perform M scale superpixel hierarchical segmentation on the target image. Image superpixel hierarchical segmentation can simulate the different visual granularity of different visual cells of the human eye to achieve better results, and judge the saliency of pixels on the image from different scales, so as to finally obtain a fair and objective saliency map. Considering the characteristics of the human eye and the performance of the algorithm, a three-layer superpixel layered segmentation method can be adopted.

Step 2, for each layer of target sub-images, perform the following steps 2.1-step 2.3:

Step 2.1, extract the feature vector of each superpixel block in the target sub-image to obtain the superpixel block feature vector; specifically, extract the color feature and texture feature of each superpixel block, and the feature vector of each superpixel block includes: 3 components of RGB mean, 256 components of RGB histogram, 3 components of HSV mean, 256 components of HSV histogram, 3 components of Lab mean, 256 components of Lab histogram and LM filtering 48 components of the sensor response.

In step 2.2, the frame area of the target sub-image is regarded as the image background, and the superpixel blocks belonging to the image background are called background superpixel blocks;

Cluster the background superpixel blocks to obtain n clusters, which are: the first cluster, the second cluster...the nth cluster; the cluster center eigenvector of the first cluster is B ₁ , the eigenvector of the cluster center of the second cluster is B ₂ , and so on, the eigenvector of the cluster center of the nth cluster is B _n , therefore, the eigenvector of the cluster center B={B ₁ ,B ₂ ,...,B _n };

In general, clustering the background superpixel blocks into 1-3 cluster sets can prevent background feature vector calculation errors caused by large differences in image border superpixels, thereby giving background nodes a more accurate evaluation method.

In this step, the background superpixel blocks are clustered, specifically, an improved K-Means clustering algorithm is used to cluster the background superpixel blocks, referring to Figure 3, including the following steps:

Step 2.2.1, set the initial clustering number of the improved K-means clustering algorithm as z, that is, the final clustering results in z clustering numbers; generally, the value of z is 3;

Step 2.2.2, use the K-means clustering algorithm for initial clustering to obtain several initial clusters; in the initial clustering, the following method is used to calculate the distance between any two superpixel blocks:

For any two superpixel blocks in the target sub-image, they are respectively denoted as superpixel block u and superpixel block v;

Suppose the RGB average value of the superpixel block u in the extracted target sub-image is f ₁ ^u , and the RGB histogram is HSV mean is HSV histogram is Lab average is Lab histogram is The LM filter response is

Let the RGB average value of the superpixel block v in the extracted target sub-image be f ₁ ^v , and the RGB histogram be HSV mean is HSV histogram is Lab average is Lab histogram is The LM filter response is

The distance D(u,v) between superpixel block u and superpixel block v is:

Among them: N(●) means normalization;

Indicates the distance of the a-th feature between superpixel block u and superpixel block v;

Among them, a=1, 3, 5, 7, representing the RGB average feature, HSV average feature, Lab average feature and LM filter response feature; m is the total number of dimensions of each feature, and e is the number of dimensions of each feature Parameters, for the RGB average feature, its dimension is 3; for the HSV average feature, its dimension is 3; for the Lab average feature, its dimension is 3; for the LM filter response feature, its dimension is 48; is the e-th dimension component of the a-th feature of the superpixel block u; is the e-th dimension component of the a-th feature of the superpixel block v;

Indicates the distance of the cth feature between superpixel block u and superpixel block v; where c=2, 4, 6 represent RGB histogram, HSV histogram and Lab histogram respectively; b is the histogram interval number; d is the number parameter of the histogram interval; is the dth histogram value of the cth feature of the superpixel block u; is the dth histogram value of the cth feature of the superpixel block v;

Then calculate the cluster center eigenvector of each initial cluster; average the features of all superpixels in a cluster to obtain the cluster center;

Step 2.2.3, select the Euclidean distance as the similarity measure between the initial clusters, so as to calculate the difference value between the cluster centers;

Step 2.2.4, judge whether the difference between any two cluster centers is smaller than the threshold θ; set the cluster center set to A, then D(g,h) represents the Euclidean distance between the cluster center g and the cluster center h;

Step 2.2.5, if the result of step 2.2.4 is "Yes", then the number N of clusters is reduced by 1, and return to step 2.2.2 for re-clustering;

Step 2.2.6, if the result of step 2.2.4 is "no", go to step 2.2.7;

Step 2.2.7, record the number of clusters and the eigenvector of the cluster center;

In step 2.2.8, the process ends.

In this step, the background prior is based on the principle of photography, and the four border areas of the image are treated as the image background. At present, most algorithms that use background priors use the entire border of the image as the background to extract the feature vector of the background region. This method cannot effectively utilize the background difference of the image border. Through investigation, it is found that the area of many image borders can be divided into 1 to 3 parts, and generally less than 3 parts. Therefore, in order to describe the background area of the image well, on the basis of superpixel segmentation of the image, the present invention uses the improved k-means clustering algorithm for the background superpixel block set of the image border to divide the background superpixel blocks on the four borders of the image Clustered into 1 to 3 sets, as image background regions. All background superpixel blocks on the four borders of the image are used to form a set of background superpixel blocks, and the color features and texture features of all superpixel blocks are extracted to describe the superpixel block information.

Step 2.3, for each superpixel block of the target sub-image, denoted as a superpixel block p, the saliency value s of the superpixel block p is calculated by the following formula:

in:

in:

D(p,B _i ) represents the distance between the superpixel block p and the cluster center feature vector B _i of the i-th cluster, i={1,2,...,n}; σ represents the scale factor, usually taken as Value is 0.5;

w is the weight, which is used to measure the distance between the superpixel block p and the center point of the target sub-image of this layer, (x, y) represents the coordinates of the center point of the superpixel block p, and (x', y') represents the target of this layer The coordinates of the center point of the sub-image;

The saliency value of each superpixel block of the target sub-image of each layer is thus calculated;

When calculating the saliency value of the superpixel block in this step, the saliency calculation of the superpixel block is performed using the spatial position and the difference with the background superpixel, specifically: clustering the background superpixel block on each layer of the target sub-image, According to the spatial position of the superpixel block and its difference with the background superpixel block, the saliency value of the superpixel block is calculated. The saliency value of the superpixel block p is the weighted average of the differences between it and all background superpixel block cluster centers, and its weight is related to its distance from the center point of the layer image. The smaller the distance, the greater the weight.

Step 3, saliency fusion of multi-scale superpixel blocks to obtain the final saliency map, and detect salient regions on the saliency map, including:

Step 3.1, calculate the saliency value of any pixel point j on the fused saliency map:

The saliency value s j of pixel point _j is the average value of the saliency values of its corresponding superpixel block at all scales, namely:

Among them: s _l is the saliency value of the superpixel block of the pixel j located in the l-th layer target sub-image;

In step 3.2, the saliency values of all pixel points j form an image saliency map, and on the saliency map, the region exceeding the set threshold is the finally detected salient region.

The salient region detection method BSP based on the image background and spatial position proposed by the present invention, the classic algorithm GR, and the classic algorithm SF are respectively used to detect the salient region in the original image in Fig. 4, and the detection results are shown in Fig. 4. 4. It can be seen that the salient area detection effect of the BSP algorithm of the present invention is good, which is obviously better than the classic algorithm GR and the classic algorithm SF. In addition, the average absolute error MAE and the area under the ROC curve AUC of the three detection algorithms are calculated. The calculation results are shown in the table below. It can be seen from the table below that the MAE value of BSP is lower than that of GR and SF; the AUC value of BSP is higher than that of GR and SF, which shows that the comprehensive performance of the BSP method is good.

Table: Comparison of Three Detection Algorithms

A salient region detection method based on image background and spatial position provided by the present invention has the following advantages:

Through the method of the present invention, it is possible to accurately judge the significant area of the image, effectively improve the detection accuracy and calculation efficiency, and provide technical support for the screening and analysis of massive image or video data applied to the Internet and cloud computing, and has good application prospect.

The above is only a preferred embodiment of the present invention, it should be pointed out that, for those of ordinary skill in the art, without departing from the principle of the present invention, some improvements and modifications can also be made, and these improvements and modifications can also be made. It should be regarded as the protection scope of the present invention.

Claims (5)

1. a kind of salient region detecting method based on image background and locus, it is characterised in that comprise the following steps：

Step 1, the super-pixel for M kind yardsticks being carried out to target image is layered segmentation, wherein, M is total number of plies of yardstick, obtains M layers Target subgraph；Every layer of target subgraph is made up of multiple super-pixel block；

Step 2, for every layer of target subgraph, following steps 2.1- steps 2.3 are performed both by：

Step 2.1, the characteristic vector of each super-pixel block in target subgraph is extracted, obtains super-pixel block characteristic vector；

Step 2.2, by the frame region of target subgraph as image background, the super-pixel block for belonging to image background is referred to as background Super-pixel block；

Background super-pixel block is clustered, n cluster is obtained, is respectively：1st cluster, n-th of cluster of the 2nd cluster ...； The cluster centre characteristic vector of 1st cluster is B₁, the cluster centre characteristic vector of the 2nd cluster is B₂, the rest may be inferred, n-th The cluster centre characteristic vector of cluster is B_n, therefore, cluster centre characteristic vector B={ B₁,B₂..., B_n}；

Step 2.3, for each super-pixel block of target subgraph, super-pixel block p is expressed as, super-pixel is calculated using following formula Block p significance value s：

<mrow> <mi>s</mi> <mo>=</mo> <mfrac> <mi>w</mi> <mi>n</mi> </mfrac> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <mi>D</mi> <mrow> <mo>(</mo> <mi>p</mi> <mo>,</mo> <msub> <mi>B</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow>

Wherein：

<mrow> <mi>w</mi> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <mi>exp</mi> <mrow> <mo>(</mo> <mfrac> <mrow> <msup> <mrow> <mo>(</mo> <mi>x</mi> <mo>-</mo> <msup> <mi>x</mi> <mo>&amp;prime;</mo> </msup> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msup> <mrow> <mo>(</mo> <mi>y</mi> <mo>-</mo> <msup> <mi>y</mi> <mo>&amp;prime;</mo> </msup> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow> <mrow> <mn>2</mn> <msup> <mi>&amp;sigma;</mi> <mn>2</mn> </msup> </mrow> </mfrac> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow>

Wherein：

D(p,B_i) represent super-pixel block p and the cluster centre characteristic vector B of ith cluster_iThe distance between, i=1,2 ..., n；σ represents scale factor；

W is weights, and for weighing the distance between super-pixel block p and this layer of target subgraph central point, (x, y) represents super-pixel block P center point coordinate, (x', y') represent the center point coordinate of this layer of target subgraph；

Thus the significance value of each super-pixel block of every layer of target subgraph is calculated；

Step 3, multiple dimensioned super-pixel block conspicuousness fusion, obtains final Saliency maps, and detects on Saliency maps aobvious Work property region, is specifically included：

Step 3.1, the significance value of any pixel j on Saliency maps after merging is calculated:

Pixel j significance value s_jIt is the average value of its significance value of super-pixel block corresponding under all yardsticks, I.e.：

<mrow> <msub> <mi>s</mi> <mi>j</mi> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <mi>M</mi> </mfrac> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>l</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </munderover> <msub> <mi>s</mi> <mi>l</mi> </msub> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>3</mn> <mo>)</mo> </mrow> </mrow>

Wherein：s_lIt is the significance value for the super-pixel block that pixel j is located at l layer target subgraphs；

Step 3.2, all pixels point j significance value forms image saliency map, on Saliency maps, more than given threshold Region is the salient region eventually detected.

2. a kind of salient region detecting method based on image background and locus according to claim 1, its feature It is, in step 1, the super-pixel for carrying out M kind yardsticks to target image using SLIC algorithms is layered segmentation.

3. a kind of salient region detecting method based on image background and locus according to claim 1, its feature It is, in step 2.1, extracts the characteristic vector of each super-pixel block in target subgraph, be specially：Extract each super-pixel block Color characteristic and textural characteristics, the characteristic vector of each super-pixel block includes：3 components, the RGB histograms of RGB average values 256 components, 3 components of HSV average values, 256 components of HSV histograms, 3 components, Lab of Lab average values it is straight 48 components of 256 components and LM the wave filters response of square figure.

4. a kind of salient region detecting method based on image background and locus according to claim 3, its feature It is, in step 2.2, background super-pixel block is clustered, specifically, is surpassed using K-Means clustering algorithms are improved to background Block of pixels is clustered.

5. a kind of salient region detecting method based on image background and locus according to claim 4, its feature It is, background super-pixel block is clustered using K-Means clustering algorithms are improved, specifically included：

Step 2.2.1, the initial clustering number of improved K-means clustering algorithms is set as z, i.e., last cluster obtains z cluster Number；

Step 2.2.2, initial clustering is carried out using K-means clustering algorithms, obtains several initial clusterings；In initial clustering When, using the distance of following methods calculating any two super-pixel block：

For any two super-pixel block in target subgraph, super-pixel block u and super-pixel block v are designated as respectively；

If the RGB average values for extracting super-pixel block u in target subgraph are f₁ ^u, RGB histograms areHSV average values are HSV histograms areLab average values areLab histograms areLM wave filters respond

If the RGB average values for extracting super-pixel block v in target subgraph are f₁ ^v, RGB histograms areHSV average values are HSV histograms areLab average values areLab histograms areLM wave filters respond

The distance between super-pixel block u and super-pixel block v D (u, v) is：

<mrow> <mi>D</mi> <mrow> <mo>(</mo> <mi>u</mi> <mo>,</mo> <mi>v</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mn>7</mn> </mfrac> <mrow> <mo>(</mo> <munder> <mi>&amp;Sigma;</mi> <mrow> <mi>a</mi> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mn>3</mn> <mo>,</mo> <mn>5</mn> <mo>,</mo> <mn>7</mn> </mrow> </munder> <mi>N</mi> <mo>(</mo> <msqrt> <mrow> <munderover> <mi>&amp;Sigma;</mi> <mrow> <mi>e</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>m</mi> </munderover> <msup> <mrow> <mo>(</mo> <msubsup> <mi>f</mi> <mrow> <mi>a</mi> <mi>e</mi> </mrow> <mi>u</mi> </msubsup> <mo>-</mo> <msubsup> <mi>f</mi> <mrow> <mi>a</mi> <mi>e</mi> </mrow> <mi>v</mi> </msubsup> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow> </msqrt> <mo>)</mo> <mo>+</mo> <munder> <mi>&amp;Sigma;</mi> <mrow> <mi>c</mi> <mo>=</mo> <mn>2</mn> <mo>,</mo> <mn>4</mn> <mo>,</mo> <mn>6</mn> </mrow> </munder> <mi>N</mi> <mo>(</mo> <msqrt> <mrow> <munderover> <mi>&amp;Sigma;</mi> <mrow> <mi>d</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>b</mi> </munderover> <mfrac> <mrow> <mn>2</mn> <msup> <mrow> <mo>(</mo> <msubsup> <mi>f</mi> <mrow> <mi>c</mi> <mi>d</mi> </mrow> <mi>u</mi> </msubsup> <mo>-</mo> <msubsup> <mi>f</mi> <mrow> <mi>c</mi> <mi>d</mi> </mrow> <mi>v</mi> </msubsup> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow> <mrow> <msubsup> <mi>f</mi> <mrow> <mi>c</mi> <mi>d</mi> </mrow> <mi>u</mi> </msubsup> <mo>+</mo> <msubsup> <mi>f</mi> <mrow> <mi>c</mi> <mi>d</mi> </mrow> <mi>v</mi> </msubsup> </mrow> </mfrac> </mrow> </msqrt> <mo>)</mo> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>4</mn> <mo>)</mo> </mrow> </mrow>

Wherein：N (●) represents normalization；

Represent the distance of a-th of feature between super-pixel block u and super-pixel block v；

Wherein, a=1,3,5,7, the average value tags of RGB, the average value tags of HSV, the average value tags of Lab and LM filtering are represented respectively Device response characteristic；M is the dimension sum of each feature, and e is the number of dimensions parameter of each feature, and for the average value tags of RGB, it is tieed up Spend for 3；For the average value tags of HSV, its dimension is 3；For the average value tags of Lab, its dimension is 3；Rung for LM wave filters Feature is answered, its dimension is 48；It is e-th of component of super-pixel block u a-th of feature；It is the of super-pixel block v E-th of component of a feature；

Represent the distance of c-th of feature between super-pixel block u and super-pixel block v；Wherein, c=2, 4,6, RGB histograms, HSV histograms and Lab histograms are represented respectively；B is histogram number；D is histogram number Measure parameter；It is d-th of histogram value of super-pixel block u c-th of feature；It is the of super-pixel block v c-th of feature D histogram value；

Then the cluster centre characteristic vector of each initial clustering is calculated；The feature of all super-pixel in one cluster is distinguished Do and be averagely worth to cluster centre；

Step 2.2.3, Euclidean distance is selected as the similarity measurement between initial clustering, so as to calculate between cluster centre Difference value；

Step 2.2.4, judges whether the difference of any two cluster centre is less than threshold θ；If cluster centre collection is combined into A, thenD (g, h) represents the Euclidean distance between cluster centre g and cluster centre h；

Step 2.2.5, if step 2.2.4 result is "Yes", the number N clustered subtracts 1, and return to step 2.2.2 gathers again Class；

Step 2.2.6, if step 2.2.4 result is "No", into step 2.2.7；

Step 2.2.7, record the number of cluster and the characteristic vector of cluster centre；

Step 2.2.8, flow terminate.