CN104504681A - Threshold image segmentation method with minimal clustering distortion - Google Patents

Abstract

The invention discloses a threshold segmentation method with minimal clustering distortion. Firstly, according to the set threshold, the image is segmented into a target part and a background part; then, the sum of the value clustering segmentation distortion in the target part and the background part is calculated when segmentation is carried out according to the set threshold; and the above process is repeated on all gray levels of the image, and the gray level corresponding to the minimal sum of the segmentation distortion is found out to be the evaluated threshold. When the method of the invention is adopted, segmentation is more accurate to carry out, and the segmentation distortion is minimal.

Description

The threshold image segmentation method that a kind of cluster degree of distortion is minimum

Technical field

The invention belongs to technical field of image processing, be specifically related to the threshold image segmentation method that a kind of cluster degree of distortion is minimum;

Background technology

Iamge Segmentation is an important basic technology in image procossing, and range of application is very extensive; The target of Iamge Segmentation is target area and the background area of segmentation image, to make the subsequent treatment such as further target identification, tracking; Therefore researchers propose various image partition method, and the image partition method wherein based on threshold value is intuitively a kind of and effective image partition method; Ostu algorithm is the classical threshold method of one proposed in 1979 by Japanese scholars Otsu the earliest, the effect stable due to it and being used widely; This algorithm is according to the one dimension grey level histogram of image, and exhaustive search makes pixel be divided into target and background two class, is one dimension threshold value when inter-class variance is maximum; But the image in the face of varying, Ostu algorithm can't all obtain desirable result in all cases, as when the variance within clusters difference of two classes is larger, a sector of breakdown pixel large for variance within clusters by a class large for deflection variance within clusters, thus is divided in the little class of variance within clusters by the segmentation threshold of Otsu method;

Summary of the invention

To be solved by this invention is the undesirable problem of conventional images dividing method effect, provides the threshold image segmentation method that a kind of cluster degree of distortion is minimum;

For solving the problem, the present invention is achieved by the following technical solutions:

The threshold image segmentation method that cluster degree of distortion is minimum, comprises the steps:

Step 1, reads in image, obtains the histogram of image; If in image, the gray scale value of pixel is [0,1 ..., L-1], statistics gray-scale value is the pixel frequency n of i _i, i=0,1 ..., L-1;

Step 2, from gray scale span [0,1 ..., L-1] in get each gray-scale value successively as segmentation threshold point T, and following steps are repeated to each segmentation threshold point T:

Step 2.1, is divided into two classes according to the above-mentioned segmentation threshold point T chosen by image, that is: be that the pixel of [0, T] forms target class C by gray scale ₀, be that the pixel of [T+1, L-1] forms background classes C by gray scale ₁;

Step 2.2, calculates target class C respectively ₀interior total pixel number N ₀with background classes C ₁interior total pixel number N ₁, wherein

$N_0=\underset{i=0}{\overset{T}{\mathrm{\Σ}}}{n}_i$ ①

$N_1=\underset{i=T+1}{\overset{L-1}{\mathrm{\Σ}}}{n}_i$ ②

Step 2.3, calculates target class C respectively ₀cluster centre point M ₀with background classes C ₁cluster centre point M ₁, wherein

M ₀∈ (0, T) and meet $\underset{i=0}{\overset{M_0-1}{\mathrm{\&Sigma;}}}{n}_i<\frac{N_0}{2}$ With $\underset{i=0}{\overset{M_0}{\mathrm{\&Sigma;}}}{n}_i\&GreaterEqual;\frac{N_0}{2}$ ③

M ₁∈ (T+1, L-1) and meet $\underset{i=T+1}{\overset{M_1-1}{\mathrm{\&Sigma;}}}{n}_i<\frac{N_1}{2}$ With $\underset{i=T+1}{\overset{M_1}{\mathrm{\&Sigma;}}}{n}_i\&GreaterEqual;\frac{N_1}{2}$ ④

Step 2.4, calculates target class C respectively ₀each pixel is to class center M ₀euclidean distance and S ₀with background classes C ₁each pixel is to class center M ₁euclidean distance and S ₁, wherein

$S_0=\underset{i=0}{\overset{T}{\mathrm{\&Sigma;}}}{n}_i\&times;|M_0-i|$ ⑤

$S_1=\underset{i=T+1}{\overset{L-1}{\mathrm{\&Sigma;}}}{n}_i\&times;|M_1-i|$ ⑥

Step 2.5, calculates when carrying out cluster with segmentation threshold point T, the cluster degree of distortion D that target and background is total _t, wherein

D _T＝S ₀+S ₁⑦

Step 3, finds out cluster degree of distortion D _tfor segmentation threshold corresponding during minimum value point T, then this segmentation threshold point T is required final segmentation threshold point.

Compared with prior art, first image is divided into target and background two class by cluster by the present invention, is required threshold value when all pixels are minimum to the Euclidean distance summation at respective class center; Because the present invention adopts the method for exhaustion to obtain global optimum's cluster degree of distortion, thus make the probability of wrong point greatly reduce, image segmentation is also better.

Accompanying drawing explanation

Fig. 1-1 ~ Fig. 1-3 is followed successively by the original graph of Ms's image, classical Ostu segmentation effect figure and segmentation effect figure of the present invention.

Fig. 2-1 ~ Fig. 2-3 is followed successively by the original graph of tire image, classical Ostu segmentation effect figure and segmentation effect figure of the present invention.

Embodiment

The threshold image segmentation method that cluster degree of distortion is minimum, comprises the steps:

Step 1, reads in image, obtains the histogram of image.If the gray shade scale of piece image f is L, then in image the gray scale value of pixel be [0,1 ..., L-1], statistics gray-scale value is the pixel frequency n of i _i, i=0,1 ..., L-1.In above-mentioned gray scale span [0,1 ..., L-1] all gray-scale values are positive integer.

Step 2, from gray scale span [0,1 ..., L-1] in get each gray-scale value successively as segmentation threshold point T, and following steps are repeated to each segmentation threshold point T:

Step 2.1, is divided into two classes according to the above-mentioned segmentation threshold point T chosen by image, that is: be that the pixel of [0, T] forms target class by gray scale, be designated as C ₀, be that the pixel of [T+1, L-1] forms background classes by gray scale, be designated as C ₁;

Step 2.2, calculates target class C respectively ₀interior total pixel number N ₀with background classes C ₁interior total pixel number N ₁, wherein

$N_0=\underset{i=0}{\overset{T}{\mathrm{\&Sigma;}}}{n}_i$ ①

$N_1=\underset{i=T+1}{\overset{L-1}{\mathrm{\&Sigma;}}}{n}_i$ ②

Step 2.3, calculates target class C respectively ₀cluster centre point M ₀with background classes C ₁cluster centre point M ₁, wherein

M ₀∈ (0, T) and meet $\underset{i=0}{\overset{M_0-1}{\mathrm{\&Sigma;}}}{n}_i<\frac{N_0}{2}$ With $\underset{i=0}{\overset{M_0}{\mathrm{\&Sigma;}}}{n}_i\&GreaterEqual;\frac{N_0}{2}$ ③

M ₁∈ (T+1, L-1) and meet $\underset{i=T+1}{\overset{M_1-1}{\mathrm{\&Sigma;}}}{n}_i<\frac{N_1}{2}$ With $\underset{i=T+1}{\overset{M_1}{\mathrm{\&Sigma;}}}{n}_i\&GreaterEqual;\frac{N_1}{2}$ ④

Step 2.4, calculates target class C respectively ₀each pixel is to class center M ₀euclidean distance and S ₀with background classes C ₁each pixel is to class center M ₁euclidean distance and S ₁, wherein

$S_0=\underset{i=0}{\overset{T}{\mathrm{\&Sigma;}}}{n}_i\&times;|M_0-i|$ ⑤

$S_1=\underset{i=T+1}{\overset{L-1}{\mathrm{\&Sigma;}}}{n}_i\&times;|M_1-i|$ ⑥

Step 2.5, calculates when carrying out cluster with segmentation threshold point T, the cluster degree of distortion D that target and background is total _t, wherein

D _T＝S ₀+S ₁⑦

Step 3, finds out cluster degree of distortion D _t=D _minfor segmentation threshold corresponding during minimum value point T=T ₀, then this segmentation threshold point is required final segmentation threshold point.

Adopt Ms's image and tire image to carry out segmentation effect to the present invention with classical Ostu method below to compare; Fig. 1-1 and Fig. 2-1 is original image, Fig. 1-2 and Fig. 2-2 is segmentation result of the present invention, and Fig. 1-3 and Fig. 2-3 is classical Ostu method segmentation result; As can be seen from the contrast of Fig. 1-2 and Fig. 1-3, the present invention is more accurate with finger segmentation for the right hand wrist of Ms, the complete profile remaining wrist and finger; Left hand segmentation for Ms is also more excellent than classical Ostu method, and left index finger and the wrist of the present invention's segmentation are more complete; The inventive method can be clearly seen that the necklace on this Ms's neck in addition, and classical Ostu method loses the integrity profile of necklace; Also similar conclusion can be drawn: the bead seat that the present invention is split is round, and edge thickness maintains unanimously from Fig. 2-2 and the contrast of Fig. 2-3; And it is clearer and accurate for the air strike segmentation on profile.

Claims (1)

1. the threshold image segmentation method that cluster degree of distortion is minimum, is characterized in that comprising the steps:

Step 1, reads in image, obtains the histogram of image; If in image, the gray scale value of pixel is [0,1 ..., L-1], statistics gray-scale value is the pixel frequency n of i _i, i=0,1 ..., L-1;

Step 2, from gray scale span [0,1 ..., L-1] in get each gray-scale value successively as segmentation threshold point T, and following steps are repeated to each segmentation threshold point T:

Step 2.1, is divided into two classes according to the above-mentioned segmentation threshold point T chosen by image, and the pixel being [0, T] by gray scale forms target class C ₀, be that the pixel of [T+1, L-1] forms background classes C by gray scale ₁;

Step 2.2, calculates target class C respectively ₀interior total pixel number N ₀with background classes C ₁interior total pixel number N ₁;

Step 2.3, calculates target class C respectively ₀cluster centre point M ₀with background classes C ₁cluster centre point M ₁;

Step 2.4, calculates target class C respectively ₀each pixel is to class center M ₀euclidean distance and S ₀with background classes C ₁each pixel is to class center M ₁euclidean distance and S ₁;

Step 2.5, calculates when carrying out cluster with segmentation threshold point T, the cluster degree of distortion D that target and background is total _t;

Step 3, finds out cluster degree of distortion D _tfor segmentation threshold corresponding during minimum value point T, then this segmentation threshold point T is required final segmentation threshold point.