CN105844625A - Movable profile image segmentation method fusing edge and area - Google Patents

Abstract

The invention discloses an active contour model image segmentation method, which is used for solving the problems of low efficiency and unsatisfactory image segmentation results in the existing image segmentation problems. The present invention comprises the following steps: input an image, obtain the edge information of the image by using the phase consistency edge detection algorithm of the LP-Garbor filter; set the threshold, remove the glitches obtained due to the influence of noise; use the centroid of each edge segment as the seed point The region grows to get the target region; finally, the target region is used as the initial contour position, and the edge information is used to rewrite the energy functional function of the active contour model to complete the image segmentation. The invention not only retains the advantages of the active contour model for fully automatic and rapid global image segmentation, but also combines the information of the edge area of the image to improve the shortcomings of the active contour model algorithm in the case of similar foreground and background colors or shadows in the foreground object. In addition, since the initialization position of the method of the present invention is relatively close to the object to be segmented, the operating efficiency is also greatly improved. At the same time, fully automatic image segmentation is realized.

Description

A Method of Active Contour Image Segmentation by Fusion of Edge and Region

technical field

The invention relates to the technical field of digital image processing, in particular to an active contour image segmentation method for fusing edges and regions.

Background technique

Image segmentation is a key step in digital image processing, which can be divided into threshold-based, region-based, edge-based segmentation methods, and segmentation methods based on specific theories. Usually a combination of methods are used.

Image edge detection is the identification of points with significant changes in image attributes, which reflect changes in different regions of the image. The most commonly used edge detection methods are first-order and second-order edge detection operators, such as Laplacian operator, Laplacian-Gauss operator, Roberts operator, Sobel operator, etc., and their core ideas are mostly based on the brightness gradient information of the image. The phase consistency assumes that the point with the largest sum of Fourier series components in the image is the feature point. There are a large number of step edges, line edges, roofs and edge information between step edges and line edges in the image. Many scholars at home and abroad have found that various types of features can appear at points with high phase consistency, including steps, lines, roofs, and Mach bands. Therefore, the phase consistency method can detect the edge of the image more effectively. Scholars Kovesi et al. proposed a phase consistency method using the Log Gabor filter, and achieved good experimental results. The present invention uses the phase consistency of the Log Gabor filter to detect the edge.

With the application of level set methods, active contour models have also achieved good development in image segmentation problems. In recent years, especially the Chan-Vese (CV) model has been well developed. Many scholars have done related research on the basis of the CV model and proposed many effective improvement schemes. The typical improvements in the early years mainly include: Kimmel et al. considered the boundary gradient information of the image, proposed the idea of combining the CV model and the GAC model, and realized a good segmentation of weak boundary images; Li Chunming et al. Reinitialization problems, the concept of a distance regularizer that avoids level set reinitialization is proposed. In addition, many scholars have also made many optimizations and improvements to the model on this basis, such as the recently proposed level set evolution model based on distance regularization, and the level set evolution model based on distance regularization considering image region information. The invention simultaneously considers the edge information and area information of the image in the CV model, and rewrites the energy functional function.

Contents of the invention

In order to solve the defects in the prior art, the present invention provides an active contour image segmentation method that fuses edge and region information. It not only retains the advantages of the global segmentation of the active contour model image segmentation algorithm, but also integrates the phase consistency edge detection algorithm to obtain the image edge information, and obtains the target area of the image with the help of the region growing method, and further segments to obtain accurate segmentation results. The problem of excessive noise in the segmentation of the active contour model has been improved, and the operating efficiency has also been improved to a certain extent.

In order to solve the problems of the technologies described above, the technical solution adopted in the present invention is:

A kind of active contour image segmentation method of fusing edge and region, is characterized in that, comprises the following steps:

Step S1: Input an image, use the phase consistency edge detection algorithm of the LP-Garbor filter to locate and obtain the edge information of the image;

Step S2: Set the threshold t, remove the edge segments whose length is less than t, and remove the glitches caused by noise;

Step S3: using the centroid of each edge segment as the seed point area to grow to obtain the target area;

Step S4: take the target area as the initial contour position, and use the active contour model to perform detailed segmentation of the image;

Step S5: Output the target image to be segmented.

Further, the above step S3 specifically includes the following steps:

Step S3.1, after obtaining the edge information of the image, find the centroid of each edge segment as the seed point, and push the marked edge point into the stack edge as the seed point;

Step S3.2. Take a seed point from the edge stack, and calculate the difference between the seed point and its four neighbor pixels (x, y). If the difference is less than the threshold E, the growth has not reached the boundary of the region, and the seed point is given the same The logo, push into the stack, seed;

Step S3.3, take out a new seed point from the stack seed, continue to grow in the four-connected direction, merge adjacent points that meet the growth conditions, and generate a new region;

Step S3.4, repeat step S3.3 until there are no adjacent points satisfying the growth condition, and the region generation process ends;

Step S3.5, take out other seed points one by one from the stack edge, and grow according to steps S3.2-S3.4.

Further, the above step S4 specifically includes the following steps:

Step S4.1, binarize the target region obtained by region growing to obtain a template T;

Step S4.2, according to the template T, set the level set function, and initialize the active contour curve;

Step S4.3, calculating the edge stop function g according to the edge information of the image;

Step S4.4, determining the energy functional of the improved CV active contour model;

Step S4.5, numerically solving the energy functional, iteratively minimizing the energy, and obtaining the final level set function;

Step S4.6, obtaining the final image segmentation result.

The method of the present invention combines the image edge area information and the active contour model image segmentation algorithm, first uses the phase consistency edge detection algorithm of the Log Gabor filter to obtain the image edge information, and uses the four-neighborhood area growing algorithm to obtain the initial area of the target, Then use the edge information and area information of the image to rewrite the CV model, and iteratively solve the numerical solution to obtain accurate image segmentation results. It not only retains the advantages of CV global segmentation, but also uses edge information to improve the insufficiency of the CV model when the background color is similar or the foreground target has noise. In addition, since the initial contour line position of this method is close to the target position to be segmented, the operating efficiency is also improved to a certain extent.

Description of drawings

Fig. 1 is a flow chart of the present invention.

detailed description

The present invention will be further described below in conjunction with the embodiments, and the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, other used embodiments obtained by persons of ordinary skill in the art without creative efforts all belong to the protection scope of the present invention.

Experimental platform of the present invention is carried out on Matlab R2011b platform, and computer system configuration: CPU is Intel (R) Core (TM) i5-3470; Main frequency is @3.20GHz; Internal memory is 4GB; Operating system is MicrosoftWindows7Professional.

The active contour image segmentation method of fusion edge and region proposed by the present invention mainly includes two modules of image edge detection by Log Gabor filter phase consistency method, further rewriting active contour model functional, and iterative numerical solution.

The first is the main steps of the Log Gabor filter phase consistency method to calculate image edge information:

The image signal I undergoes Fourier series expansion:

A _n is the amplitude of the Nth cosine component, Indicates the local phase at the x position. Solving the PC function with energy, the relationship is:

E(x)＝PC(x)∑ _n A _n (2)

The Log Garbor filter has very good performance in terms of spatial positioning and frequency separation, and it is more effective for the natural encoding of images. Log Garbor is selected as a filter with consistent phase positioning. The response equations of its sine and cosine are:

a _n and b _n are coefficients of cosine and sine of Log Gabor filter, R _n is the amplitude of local frequency, θ _n is the phase of local frequency. The PC function is expressed as:

is the weighted average of the phases. After obtaining the edge information of the image, the fine burr edge is removed by the threshold method.

Secondly, based on the edge information of the image, the region is grown to obtain the target region. The main steps are:

(1) Know the edge segment of the image, determine the centroid of each edge segment, locate it as the seed point of each edge segment, and push the marked seed point into the stack E;

(2) Take a seed point v from the stack E, and calculate the difference between the seed point v and its four neighbor pixels v(x, y). If the difference is less than the threshold Q, the region growth has not yet reached the boundary of the region, and v( x, y) are given the same identity as the seed point and pushed into the stack S;

(3) Take out a new seed point from the stack S, continue to grow according to the four-connected direction, merge adjacent points that meet the growth conditions, and generate a new area;

(4) Repeat step (3) until there are no adjacent points that meet the growth conditions, and the region generation process ends;

(5) Take out other seed points one by one from stack E, and continue to grow according to steps (2)-(4).

Finally, after determining the edge information and area information of the image, rewrite the CV energy functional and iteratively perform numerical solution:

The edge stopping function g(x,y)=1/(1+Kλ ² ) in the geodesic active contour model is rewritten by the edge information λ of the image, where K is a coefficient. The target region Obj obtained by region growing, defines the template According to the definition of the level set function, according to the template α, calculate the initial zero level set curve:

definition where μ is a constant coefficient and μ>0, C is the active contour curve, and G _σ is a Gaussian kernel function with width σ.

Rewrite the CV model energy functional:

$\begin{array}{c}\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&mu;</span>}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; L</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; g</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; \&Integral;</span>}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; x</span>}\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; the\; y</span>}\\ <span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}{<span\; class="notranslate">\; \&Integral;</span>}_{\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; x</span>}\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; the\; y</span>}\end{array}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$

Wherein, I ₀ (x, y) is the image to be segmented, Length(C) indicates the length of the boundary curve C, and Area(inside(C)) is the area of the inner area of the curve C. μ, γ≥0, λ ₁ , λ ₂ >0 are weight coefficients.

Finally, the partial derivative of the level set function is calculated by the variational method, and then the numerical solution of the model is realized by the method of finite difference. Use the method of central difference to approximate the numerical solution, right Perform half-point central difference, the partial derivative format is:

Claims (3)

1. a combination of edge and the active contour image partition method in region, it is characterised in that comprise the following steps:

Step S1: input picture, uses the phase equalization edge detection algorithm location of LP-Garbor wave filter to obtain image Marginal information；

Step S2: set threshold value t, removes the length edge section less than t, removes the burr obtained because of influence of noise；

Step S3: with the barycenter of each edge section for seed points region growing, obtain target area；

Step S4: with target area for initial profile position, uses movable contour model to carry out the detailed segmentation of image；

Step S5: export target image to be split.

Movable contour model image partition method the most according to claim 1, it is characterised in that above-mentioned step S3 is concrete Comprise the following steps:

Step S3.1, after the marginal information obtaining image, the barycenter finding each edge section is seed points, by marked limit Edge point is pressed into storehouse edge as seed points；

Step S3.2, from edge storehouse, take out a seed points, calculate seed points and its four neighborhood territory pixel (x, y) character is poor, as Really this difference is less than threshold value E, then grow and also do not arrive this zone boundary, gives the mark identical with seed points, is pressed into storehouse, seed；

Step S3.3, from storehouse seed, take out a new seed points, continue four communication direction growths, neighbouring meet grow The point of condition is incorporated to, and generates new region；

Until no longer there is the neighbouring point meeting growth conditions, this Area generation process in step S3.4, repetition step S3.3 Terminate；

Step S3.5, from storehouse edge, take out other seed points one by one, grow by S3.2-S3.4 step.

Movable contour model image partition method the most according to claim 1, it is characterised in that above-mentioned step S4 is concrete Comprise the following steps:

Target area obtained by step S4.1, region growing, binaryzation obtains template T；

Step S4.2, according to template T, define level set function, initialize active contour curve；

Step S4.3, according to image edge information, calculate corresponding Edge-stopping function g；

Step S4.4, rewriting Chan-Vese movable contour model energy functional；

Step S4.5, use finite difference method carry out numerical solution to energy functional, and iteration minimizes energy, obtains final water Flat set function；

Step S4.6, output final image segmentation result.