CN106709927A - Method for extracting target from acoustic image under complex background - Google Patents

Abstract

The invention discloses a method for extracting a target from an acoustic image under complex background. The method includes the following steps that: (1) a Sobel operator is utilized to perform edge detection on the acoustic image, so that a gradient image can be obtained; (2) the threshold value of the gradient image is selected through the 3sigma criterion, and image binarization processing is performed; (3) primary expansion processing is performed on an obtained binarized image; (4) a depth-first search algorithm is utilized to perform target search on the image which has been subjected to the expansion processing; and (5) center coordinate calculation is performed on a searched target, the final position of the target in the image is determined. A target in an acoustic image is difficult to be extracted due to causes such as the movement of background or the low signal-to-noise ratio of the target, while, with the method of the invention adopted, the above target extraction problem can be solved.

Description

A kind of target extraction method under complex background in acoustic picture

Technical field

The invention belongs to the Objective extraction side in acoustic picture under image processing field, more particularly to a kind of complex background Method.

Background technology

Than larger, feature based Point matching carries out the calculation of Objective extraction to the noise carried due to acoustic picture in optical imagery Method is not suitable for acoustic picture, therefore the conventional Objective extraction gimmick for acoustic picture mainly has two major classes, and the first is Classical background removal gimmick, that is, select suitable image threshold, is judged as effective target to the pixel more than threshold value, less than threshold The pixel of value is judged as background；Second is to utilize image difference method, will before and after two width acoustic pictures do calculus of differences, it is poor Point result is exactly effective target.The acoustic picture that above-mentioned two classes method is fixed for background or signal noise ratio (snr) of image is higher has very Strong applicability, but when acoustic picture background is that background changes in real time when movement, or signal noise ratio (snr) of image is low, for example, scheme When having a large amount of interference as in, conventional object extraction algorithm cannot solve the above problems very well.

The content of the invention

The purpose of the present invention is directed to above-mentioned deficiency, there is provided the Objective extraction side under a kind of complex background in acoustic picture Method.

In order to achieve the above object, the technical solution adopted in the present invention is as follows：Under a kind of complex background in acoustic picture Target extraction method, specifically include following steps：

Step one, the rim detection of acoustic picture f (x, y) is carried out using Sobel operators, obtains gradient image G (x, y), Computing formula is as follows：

Wherein, G_x、G_yRespectively acoustic picture is calculated as follows in x-axis, the gradient magnitude of y-axis：

G_x=f (x+1, y-1)+2f (x+1, y)+f (x+1, y+1) }-{ f (x-1, y-1)+2f (x-1, y)+f (x-1, y+ 1)}

G_y={ f (x-1, y+1)+2f (x, y+1)+f (x+1, y+1) }-{ f (x-1, y-1)+2f (x, y-1)+f (x+1, y- 1)}

Wherein, x represents the abscissa of pixel in acoustic picture, and y represents the ordinate of pixel in acoustic picture；

Step 2, the threshold value T of gradient image G (x, y) is chosen using 3 σ criterions, carries out binary conversion treatment, obtains binaryzation Image B (x, y), is calculated as follows：

Step 3, an expansion process is carried out to binary image B (x, y), obtains image b (x, y), specific formula for calculation For：

B (x, y)=b (x-1, y)=b (x+1, y)=b (x, y-1)=b (x, y+1)=1, if B (x, y)=1

Step 4, target search is carried out using Depth Priority Algorithm to image b (x, y) after expansion process, to figure The search order of picture according to the position (x, y) for from top to bottom, from left to right, recording target effective pixel points in the picture, and The effective coverage S={ (x, y) | (x, y) ∈ S } of its composition；

Step 5, target's center coordinate (x is carried out according to the target effective region S for searching_c,y_c) calculate, establish target and exist Final position in image, is calculated as follows：

Wherein m_u,v(u, v ∈ { 0,1 }) is calculated as follows：

(x, y) therein is the pixel coordinate position in the S of target effective region, and corresponding pixel value is f (x, y).

Further, the acoustic picture is the underwater picture gathered by double frequency identification sonar.

Further, the pixel number that the target effective region S in the acoustic picture is included is less than 100, is more than 10。

Further, the detailed process of selected threshold T is as follows in the step 2：

It is μ that pixel value G (x, y) of each pixel in gradient image obeys average, and variance is σ²Gaussian Profile, That is G (x, y)~N (μ, σ²), according to 3 σ criterions, the probability that G (x, y) falls outside interval [- 3 σ, 3 σ] is less than 0.3%, therefore threshold Value T=μ+β × 3 σ, wherein β are threshold value adjustment factor, μ and σ²Be calculated as follows：

Wherein N is the total number of all pixels point in gradient image.

Beneficial effects of the present invention are as follows：The present invention is proposed for the deficiency of Objective extraction gimmick in conventional acoustic image A kind of extraction algorithm based on rim detection, by asking for the gradient magnitude of acoustic picture, suitable threshold is chosen using 3 σ criterions Value carries out image binaryzation, for the expansion process in preventing Target Splitting from carrying out a morphology, finally using depth-first Searching algorithm carries out Objective extraction, and the centre coordinate position of target is established using weighted average.The method that the present invention is provided can Effectively to solve the problems, such as Objective extraction in the acoustic picture under complex background, such as background is mobile or the low environment of signal to noise ratio Under.

Brief description of the drawings

Fig. 1 is the flow chart of the inventive method；

Fig. 2 is Depth Priority Algorithm flow chart in the present invention；

Specific embodiment

The present invention is described in further details with reference to specific embodiment and accompanying drawing, but the present invention is not only limited to It is secondary.

Pass through to walk the acoustic picture that boat investigation method obtains the shoal of fish under water using double frequency identification sonar, then using the present invention Method Objective extraction is carried out to fish body therein.Fig. 1 is the algorithm flow chart of acoustic picture Objective extraction under complex background, main Implementation process is wanted to be described below：

Step one, the rim detection of acoustic picture f (x, y) is carried out using Sobel operators, obtains gradient image G (x, y), Computing formula is as follows：

Wherein, G_x、G_yRespectively acoustic picture is calculated as follows in x-axis, the gradient magnitude of y-axis：

G_x=f (x+1, y-1)+2f (x+1, y)+f (x+1, y+1) }-{ f (x-1, y-1)+2f (x-1, y)+f (x-1, y+ 1)}

G_y={ f (x-1, y+1)+2f (x, y+1)+f (x+1, y+1) }-{ f (x-1, y-1)+2f (x, y-1)+f (x+1, y- 1)}

Wherein, x represents the abscissa of pixel in acoustic picture, and y represents the ordinate of pixel in acoustic picture；

Step 2, the threshold value T of gradient image G (x, y) is chosen using 3 σ criterions, carries out binary conversion treatment, obtains binaryzation Image B (x, y), is calculated as follows：

Calculating process for threshold value T is as follows：

It is μ that pixel value G (x, y) of each pixel in gradient image obeys average, and variance is σ²Gaussian Profile, That is G (x, y)~N (μ, σ²), according to 3 σ criterions, the probability that G (x, y) falls outside interval [- 3 σ, 3 σ] is less than 0.3%, therefore threshold Value T=μ+β × 3 σ, wherein β are threshold value adjustment factor, and β values are 1.1, μ and σ in this example²Be calculated as follows：

Wherein N is the total number of all pixels point in gradient image.

Step 3, " ten types " expansion process is carried out once to binary image B (x, y), obtains image b (x, y), specific meter Calculating formula is：

B (x, y)=b (x-1, y)=b (x+1, y)=b (x, y-1)=b (x, y+1)=1, if B (x, y)=1

Step 4, carries out target and searches to image b (x, y) after expansion process according to order from top to bottom, from left to right Rope, when detecting position (x₀,y₀) corresponding to value be 1 when, the position extended target is searched using Depth Priority Algorithm Rope, as shown in Fig. 2 being described in detail below：

(1) by coordinate (x₀,y₀) pop down；

(2) judge whether current stack is empty, if empty represent that this depth-first search terminates, perform (8th) step；

(3) stack top element (x, y) is taken out；

(4) judge (whether x-1, be effectively y) that element value is 1, and is not accessed, then should for the top position of the element Coordinate pop down；

(5) whether the right positions (x, y+1) for judging the element are effectively that element value is 1, and be not accessed, then should Coordinate pop down；

(6) judge (whether x+1, be effectively y) that element value is 1, and is not accessed, then should for the lower position of the element Coordinate pop down；

(7) whether the leftward position (x, y-1) for judging the element is effectively that element value is 1, and be not accessed, then should Coordinate pop down；

(8) top-of-stack pointer subtracts one；

(9) effective coverage of this search is recorded, i.e., the coordinate position of all once stackings, statistics effective coverage is wrapped The pixel number for containing, if vegetarian refreshments number judges that this target is effective less than 100 and more than 10, and the effective district Domain is designated as S={ (x, y) | (x, y) ∈ S }

Step 5, target's center coordinate (x is carried out according to the target effective region S for searching_c,y_c) calculate, establish target and exist Final position in image, is calculated as follows：

Wherein m_u,v(u, v ∈ { 0,1 }) is calculated as follows：

(x, y) therein is the pixel coordinate position in the S of target effective region, and corresponding pixel value is in step one f(x,y)。

Based on above-mentioned flow, it is possible to obtain specific position of the effective target i.e. fish body of acoustic picture in acoustic picture Put, for follow-up data processing provides facility.

Claims (4)

1. the target extraction method under a kind of complex background in acoustic picture, it is characterised in that comprise the following steps：

Step one, the rim detection of acoustic picture f (x, y) is carried out using Sobel operators, obtains gradient image G (x, y), is calculated Formula is as follows：

$G(x,y)=\sqrt{G_x^2+G_y^2}$

Wherein, G_x、G_yRespectively acoustic picture is calculated as follows in x-axis, the gradient magnitude of y-axis：

G_x=f (x+1, y-1)+2f (x+1, y)+f (x+1, y+1) }-f (x-1, y-1)+2f (x-1, y)+f (x-1, y+1) }

G_y={ f (x-1, y+1)+2f (x, y+1)+f (x+1, y+1) }-{ f (x-1, y-1)+2f (x, y-1)+f (x+1, y-1) }

Wherein, x represents the abscissa of pixel in acoustic picture, and y represents the ordinate of pixel in acoustic picture.

Step 2, the threshold value T of gradient image G (x, y) is chosen using 3 σ criterions, carries out binary conversion treatment, obtains binary image B (x, y), is calculated as follows：

$B(x,y)=\left\{\begin{array}{c}1,G(x,y)>T\\ 0,G(x,y)\≤T\end{array}\right..$

Step 3, an expansion process is carried out to binary image B (x, y), obtains image b (x, y), and specific formula for calculation is：

B (x, y)=b (x-1, y)=b (x+1, y)=b (x, y-1)=b (x, y+1)=1, ifB (x, y)=1.

Step 4, carries out target search, to image using Depth Priority Algorithm to image b (x, y) after expansion process Search order is according to the position (x, y) for from top to bottom, from left to right, recording target effective pixel points in the picture, and its group Into effective coverage S={ (x, y) | (x, y) ∈ S }.

Step 5, target's center coordinate (x is carried out according to the target effective region S for searching_c,y_c) calculate, target is established in image In final position, be calculated as follows：

$\left\{\begin{array}{c}{x}_c=m_{1,0}/m_{0,0}\\ y_c=m_{0,1}/m_{0,0}\end{array}\right.$

Wherein m_u,v(u, v ∈ { 0,1 }) is calculated as follows：

$m_{u,v}=\underset{x}{\Σ}\underset{y}{\Σ}{x}^u{y}^{v}f(x,y)$

(x, y) therein is the pixel coordinate position in the S of target effective region, and corresponding pixel value is f (x, y).

2. the target extraction method under a kind of complex background according to claim 1 in acoustic picture, it is characterised in that institute It is the underwater picture gathered by double frequency identification sonar to state acoustic picture.

3. the target extraction method under a kind of complex background according to claim 1 in acoustic picture, it is characterised in that institute Pixel number that the target effective region S in acoustic picture included is stated less than 100, more than 10.

4. the target extraction method under a kind of complex background according to claim 1 in acoustic picture, it is characterised in that institute The detailed process for stating selected threshold T in step 2 is as follows：

It is μ that pixel value G (x, y) of each pixel in gradient image obeys average, and variance is σ²Gaussian Profile, i.e. G (x, y)~N (μ, σ²), according to 3 σ criterions, the probability that G (x, y) falls outside interval [- 3 σ, 3 σ] is less than 0.3%, therefore threshold value T =μ+β × 3 σ, wherein β are threshold value adjustment factor, μ and σ²Be calculated as follows：

$\mathrm{\μ}=\frac{1}{N}\underset{x}{\Σ}\underset{y}{\Σ}G(x,y)$

${\mathrm{\σ}}^2=\frac{1}{N}\underset{x}{\Σ}\underset{y}{\Σ}{\left(G\right(x,y)-\mathrm{\μ})}^2$

Wherein N is the total number of all pixels point in gradient image.