CN105335962A

CN105335962A - Tobacco field acquisition image segmentation method

Info

Publication number: CN105335962A
Application number: CN201510551490.9A
Authority: CN
Inventors: 陈泽鹏
Original assignee: China National Tobacco Corp Guangdong Branch
Current assignee: China National Tobacco Corp Guangdong Branch
Priority date: 2015-08-31
Filing date: 2015-08-31
Publication date: 2016-02-17

Abstract

The invention proposes a method for segmenting images collected from tobacco fields. When the method detects that there is a large change in the illumination in the scene, the homomorphic filtering algorithm is used to suppress the impact of illumination changes on video segmentation, and then the color difference histogram algorithm is used for video segmentation. .So as to ensure that the background update and foreground target segmentation can still be carried out robustly in video scenes with large illumination changes.

Description

A Segmentation Method of Tobacco Field Collection Image

技术领域 technical field

本发明涉及图像处理领域，更具体地，涉及一种烟田采集图像分割方法。 The invention relates to the field of image processing, and more particularly, to a method for segmenting images collected from tobacco fields.

背景技术 Background technique

对采集烟田的图像进行处理的过程中，如何处理光照变化所带来的影响是一个重要难题。尽管针对场景中含有光照变化的情况已提出很多算法，然而光照变化的任意性和快速性仍然使得这些检测算法存在较高的误检率。近年来，提出了一种新的学习机制改进了高斯混合算法对背景的适应能力，并利用基于帧差的启发式方法辅助速率控制方法减少光照变化下的前景误检率，还有用3个阈值对像素属于前景还是背景进行分类的新方法，在一定程度上解决不同光照条件对视频分割的影响；以及提出尺度不变的局部3元模式算子解决光照变化的影响等.通过仿真对比发现这些方法提出的颜色差分直方图视频分割算法与上述算法相比具有目标分割稳健和运算速度快等特点。由于这些算法是利用图像和其背景图像颜色差分建立颜色差分直方图，然后设定分割阈值，最后判断图像各像素属于背景还是前景目标，因而当场景中光照变化幅度较大时，不可避免的将光照变化引起的图像和其背景图像对应像素的差异统计到颜色差分直方图中，影响分割阈值的设定，从而造成对图像分割失效，对其背景图像更新失败.如果背景更新持续失效，必将会影响后续视频分割效果. In the process of processing the image of the collected tobacco field, how to deal with the influence of the illumination change is an important problem. Although many algorithms have been proposed for the scene with illumination changes, the randomness and rapidity of illumination changes still make these detection algorithms have a high false detection rate. In recent years, a new learning mechanism has been proposed to improve the adaptability of the Gaussian mixture algorithm to the background, and a heuristic method based on frame difference is used to assist the rate control method to reduce the foreground false detection rate under illumination changes, and three thresholds are used A new method of classifying whether a pixel belongs to the foreground or the background solves the influence of different lighting conditions on video segmentation to a certain extent; and proposes a scale-invariant local 3-element pattern operator to solve the influence of lighting changes, etc. Through simulation and comparison, it is found that these Compared with the above algorithm, the color difference histogram video segmentation algorithm proposed by the method has the characteristics of robust object segmentation and fast operation speed. Because these algorithms use the color difference between the image and its background image to establish a color difference histogram, then set the segmentation threshold, and finally judge whether each pixel of the image belongs to the background or the foreground object. The difference between the corresponding pixels of the image and its background image caused by illumination changes is counted into the color difference histogram, which affects the setting of the segmentation threshold, resulting in the failure of image segmentation and the failure of updating its background image. If the background update continues to fail, it will It will affect the subsequent video segmentation effect.

发明内容 Contents of the invention

本发明提供一种烟田采集图像分割方法，该方法能抑制光照变化对图像分割稳健性的影响，可稳健的进行图像背景更新和前景目标分割。 The invention provides a tobacco field collection image segmentation method, which can suppress the impact of illumination changes on image segmentation robustness, and can stably perform image background update and foreground target segmentation.

为了达到上述技术效果，本发明的技术方案如下： In order to achieve the above-mentioned technical effect, the technical scheme of the present invention is as follows:

一种烟田采集图像分割方法，包括以下步骤： A tobacco field acquisition image segmentation method, comprising the following steps:

S1：统计采样的烟田图像帧数Nf，采用基于概率的背景抽取算法抽取背景图像f_B(x,y)； S1: Statistically sampled tobacco field image frame number Nf, using the probability-based background extraction algorithm to extract the background image f _B (x, y);

S2：分别提取第i帧图像f_i(x,y)的红色、绿色和蓝色值Ri(x,y)、Gi(x,y)和Bi(x,y)，i＝0,1,···，Nf； S2: Extract the red, green and blue values Ri(x,y), Gi(x,y) and Bi(x,y) of the i-th frame image f _i (x,y) respectively, i=0,1, ···, Nf;

S3：将RGB色彩空间的背景图像f_B(x,y)和第i帧图像f_i(x,y)转化到HSV色彩空间，第i帧图像及其背景图像的亮度分量分别为fv_i(x,y)和fv_B(x,y)，判断第i帧图像及其背景图像光照是否有变化，若无变化就将第i帧图像进行分割，若有变化就进一步判断第i帧图像是否存在光照昏暗； S3: Convert the background image f _B (x, y) of the RGB color space and the i-th frame image f _i (x, y) to the HSV color space, and the brightness components of the i-th frame image and its background image are respectively fv _i ( x, y) and fv _B (x, y), to determine whether there is any change in the illumination of the i-th frame image and its background image, if there is no change, the i-th frame image is segmented, and if there is a change, it is further judged There is dim light;

S4：若存在光照昏暗就分别对fv_i(x,y)和fv_B(x,y)进行光照补偿，并将光照补偿后的fv_i(x,y)和fv_B(x,y)进行光照抑制再结合色调、饱和度分量转回到RGB空间生成f’_i(x,y)和f’_B(x,y)；若不存在光照昏暗就直接对fv_i(x,y)和fv_B(x,y)进行光照抑制再结合色调、饱和度分量转回到RGB空间生成f”_i(x,y)和f”_B(x,y)； S4: If there is dim light, perform light compensation on fv _i (x, y) and fv _B (x, y) respectively, and perform light compensation on fv _i (x, y) and fv _B (x, y) Illumination suppression is combined with hue and saturation components to return to RGB space to generate f' _i (x, y) and f' _B (x, y); if there is no dim light, directly fv _i (x, y) and fv _B (x, y) performs light suppression and then combines the hue and saturation components to return to the RGB space to generate f” _i (x, y) and f” _B (x, y);

S5：对f’_i(x,y)、f’_B(x,y)或f”_i(x,y)、f”_B(x,y)重复步骤S3-S5的过程进行背景更新和分割。 S5: Repeat steps S3-S5 for f' _i (x, y), f' _B (x, y) or f" _i (x, y), f" _B (x, y) for background update and segmentation .

进一步地，所述步骤S3中判断第i帧图像及其背景图像光照是否有变化的方法如下： Further, in the step S3, the method for judging whether the illumination of the i-th frame image and its background image has changed is as follows:

式中x和y分别代表大小为M×N的一帧图像的像素坐标，Th₁＝c₁×M×N,c₁为常系数。 In the formula, x and y respectively represent the pixel coordinates of a frame image whose size is M×N, Th ₁ =c ₁ ×M×N, and c ₁ is a constant coefficient.

进一步地，所述步骤S3中对第i帧图像进行分割的过程如下： Further, the process of segmenting the i-th frame image in the step S3 is as follows:

S31：利用f_i(x,y)和f_B(x,y)分别计算颜色差分直方图值 S31: Use f _i (x, y) and f _B (x, y) to calculate the color difference histogram value respectively

ΔR(x,y)＝R_i(x,y)-R_B(x,y) ΔR(x,y)=R _i (x,y)-R _B (x,y)

ΔG(x,y)＝G_i(x,y)-G_B(x,y)x＝1,2,…,M；y＝1,2…,N； ΔG(x,y)=G _i (x,y)-G _B (x,y)x=1,2,...,M; y=1,2...,N;

ΔB(x,y)＝B_i(x,y)-B_B(x,y) ΔB(x,y)=B _i (x,y)-B _B (x,y)

S32：对颜色差分直方图值进行平滑处理： S32: smoothing the color difference histogram values:

S33：计算分割左右阈值SR_i、SL_i： S33: Calculate the segmentation left and right thresholds SR _i , SL _i :

S34：分割图像： S34: Segment the image:

S35：若QB(x,y)为0就将图像f_i(x,y)中对应位置的像素信息更新到其背景图像fv_B(x,y)中： S35: If QB(x, y) is 0, update the pixel information of the corresponding position in the image f _i (x, y) to its background image fv _B (x, y):

若QB(x,y)不为0，就去除前景阴影并将相应的QB(x,y)置为0； If QB(x,y) is not 0, remove the foreground shadow and set the corresponding QB(x,y) to 0;

S36：对QB(x,y)采用约束行程算法和形态学方法填补前景目标的孔去除孤立像素和极小目标获得最终的图像分割。 S36: For QB(x,y), the constrained stroke algorithm and the morphological method are used to fill the holes of the foreground target, remove isolated pixels and extremely small targets, and obtain the final image segmentation.

进一步地，所述步骤S3中判断第i帧图像是否存在光照昏暗的方法如下： Further, in the step S3, the method for judging whether the i-th frame image has dim light is as follows:

式中x和y分别代表大小为M×N的一帧图像的像素坐标，Th₂＝c₂×M×N,c₂为常系数。 In the formula, x and y respectively represent the pixel coordinates of a frame image with a size of M×N, Th ₂ =c ₂ ×M×N, and c ₂ is a constant coefficient.

进一步地，所述步骤S4中进行光照抑制的过程如下： Further, the process of light suppression in the step S4 is as follows:

进一步地，所述步骤S4中进行光照补偿的过程如下： Further, the process of performing illumination compensation in the step S4 is as follows:

与现有技术相比，本发明技术方案的有益效果是： Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

本发明方法在检测到场景中光照有较大变化时，利用同态滤波算法抑制光照变化对视频分割的影响，然后采用颜色差分直方图算法进行视频分割，从而保证在含有较大光照变化的视频场景中，仍可稳健的进行背景更新和前景目标分割。 When the method of the present invention detects that there is a large change in the illumination in the scene, the homomorphic filtering algorithm is used to suppress the influence of the illumination change on the video segmentation, and then the color difference histogram algorithm is used to perform video segmentation, thereby ensuring that the video with a large illumination change In the scene, the background update and foreground object segmentation can still be performed robustly.

附图说明 Description of drawings

图1为本发明方法的流程图。 Fig. 1 is the flowchart of the method of the present invention.

具体实施方式 detailed description

附图仅用于示例性说明，不能理解为对本专利的限制； The accompanying drawings are for illustrative purposes only and cannot be construed as limiting the patent;

为了更好说明本实施例，附图某些部件会有省略、放大或缩小，并不代表实际产品的尺寸； In order to better illustrate this embodiment, some parts in the drawings will be omitted, enlarged or reduced, and do not represent the size of the actual product;

对于本领域技术人员来说，附图中某些公知结构及其说明可能省略是可以理解的。 For those skilled in the art, it is understandable that some well-known structures and descriptions thereof may be omitted in the drawings.

下面结合附图和实施例对本发明的技术方案做进一步的说明。 The technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

实施例1 Example 1

步骤S3中判断第i帧图像及其背景图像光照是否有变化的方法如下： In step S3, the method for judging whether the illumination of the i-th frame image and its background image has changed is as follows:

步骤S3中对第i帧图像进行分割的过程如下： The process of segmenting the i-th frame image in step S3 is as follows:

ΔR(x,y)＝R_i(x,y)-R_B(x,y) ΔR(x,y)=R _i (x,y)-R _B (x,y)

ΔB(x,y)＝B_i(x,y)-B_B(x,y) ΔB(x,y)=B _i (x,y)-B _B (x,y)

S34：分割图像： S34: Segment the image:

步骤S3中判断第i帧图像是否存在光照昏暗的方法如下： In step S3, the method for judging whether there is dim light in the i-th frame image is as follows:

步骤S4中进行光照抑制的过程如下： The process of light suppression in step S4 is as follows:

本发明方法在检测到场景中光照有较大变化时，利用同态滤波算法抑制光照变化对视频分割的影响，然后采用颜色差分直方图算法进行视频分割.从而保证在含有较大光照变化的视频场景中，仍可稳健的进行背景更新和前景目标分割。 When the method of the present invention detects that there is a large change in the illumination in the scene, the homomorphic filtering algorithm is used to suppress the impact of the illumination change on the video segmentation, and then the color difference histogram algorithm is used to perform video segmentation. Thereby ensuring that the video with a large illumination change In the scene, the background update and foreground object segmentation can still be performed robustly.

相同或相似的标号对应相同或相似的部件； The same or similar reference numerals correspond to the same or similar components;

附图中描述位置关系的用于仅用于示例性说明，不能理解为对本专利的限制； The positional relationship described in the drawings is only for illustrative purposes and cannot be construed as a limitation to this patent;

显然，本发明的上述实施例仅仅是为清楚地说明本发明所作的举例，而并非是对本发明的实施方式的限定。对于所属领域的普通技术人员来说，在上述说明的基础上还可以做出其它不同形式的变化或变动。这里无需也无法对所有的实施方式予以穷举。凡在本发明的精神和原则之内所作的任何修改、等同替换和改进等，均应包含在本发明权利要求的保护范围之内。 Apparently, the above-mentioned embodiments of the present invention are only examples for clearly illustrating the present invention, rather than limiting the implementation of the present invention. For those of ordinary skill in the art, other changes or changes in different forms can be made on the basis of the above description. It is not necessary and impossible to exhaustively list all the implementation manners here. All modifications, equivalent replacements and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the claims of the present invention.

Claims

1. a tobacco field acquisition image segmentation method, is characterized in that, comprises the following steps:

S1: Statistically sampled tobacco field image frame number Nf, using the probability-based background extraction algorithm to extract the background image f _B (x, y);

S2: Extract the red, green and blue values Ri(x,y), Gi(x,y) and Bi(x,y) of the i-th frame image f _i (x,y) respectively, i=0,1, ···, Nf;

S3: Convert the background image f _B (x, y) of the RGB color space and the i-th frame image f _i (x, y) to the HSV color space, and the brightness components of the i-th frame image and its background image are respectively fv _i ( x, y) and fv _B (x, y), to determine whether there is any change in the illumination of the i-th frame image and its background image, if there is no change, the i-th frame image is segmented, and if there is a change, it is further judged There is dim light;

S4: If there is dim light, perform light compensation on fv _i (x, y) and fv _B (x, y) respectively, and perform light compensation on fv _i (x, y) and fv _B (x, y) Illumination suppression is combined with hue and saturation components to return to RGB space to generate f' _i (x, y) and f' _B (x, y); if there is no dim light, directly fv _i (x, y) and fv _B (x, y) performs light suppression and then combines the hue and saturation components to return to the RGB space to generate f” _i (x, y) and f” _B (x, y);

S5: Repeat steps S3-S5 for f' _i (x, y), f' _B (x, y) or f" _i (x, y), f" _B (x, y) for background update and segmentation .

2. the tobacco field collection image segmentation method according to claim 1, is characterized in that, in the described step S3, the method for judging whether the i-th frame image and the background image illumination thereof has changed is as follows:

{Σ Σ}_{x x = = 11}^{M m} {Σ Σ}_{y the y = = 11}^{N N} | | {fv fv}_{i i} ((x x,, y the y)) - - {fv fv}_{B B} ((x x,, y the y)) | | > > {Th Th}_{11}

In the formula, x and y respectively represent the pixel coordinates of a frame image whose size is M×N, Th ₁ =c ₁ ×M×N, and c ₁ is a constant coefficient.

3. tobacco field acquisition image segmentation method according to claim 1, is characterized in that, in described step S3, the process that the i-th frame image is segmented is as follows:

S31: Use f _i (x, y) and f _B (x, y) to calculate the color difference histogram value respectively

ΔR(x,y)=R _i (x,y)-R _B (x,y)

ΔG(x,y)=G _i (x,y)-G _B (x,y)x=1,2,...,M; y=1,2...,N;

ΔB(x,y)=B _i (x,y)-B _B (x,y)

S32: smoothing the color difference histogram values:

{{{HS HS}_{j j}^{c c} = = \frac{{Σ Σ}_{k k = = - - 22}^{22} {Hist Hist}_{i i}^{c c} ((j j + + k k))}{55}}}_{c c = = R R,, G G,, B B}^{j j = = - - 253253,, - - 252252,, ... ...,, 253253};;

S33: Calculate the segmentation left and right thresholds SR _i , SL _i :

{L L}_{t t}^{c c} = = \underset{k k &Element; &Element; ((- - 252252,, {μ μ}_{t t}^{c c} - - 11))}{m m a a x x} ((i i n no d d))

{R R}_{t t}^{c c} = = \underset{k k &Element; &Element; (({μ μ}_{t t}^{c c} + + 11,, 252252))}{m m i i n no} ((i i n no d d))

i i n no d d \overset{Δ Δ}{= =} {{k k | | {HS HS}_{t t}^{c c} ((k k)) \leq \leq {HS HS}_{t t}^{c c} ((k k - - 11)) a a n no d d {HS HS}_{t t}^{c c} ((k k)) \leq \leq {HS HS}_{t t}^{c c} ((k k + + 11))}};;

{SR SR}_{i i} = = {R R}_{i i}^{R R} + + {R R}_{i i}^{G G} + + {R R}_{i i}^{B B}

{SL SL}_{i i} = = {L L}_{i i}^{R R} + + {L L}_{i i}^{G G} + + {L L}_{i i}^{B B}

S34: Segment the image:

Q Q B B ((x x,, y the y)) = = \{\begin{matrix} 00,, i i f f {D D.}_{F f B B} ((x x,, y the y)) \leq \leq S S {R R}_{i i} o o r r {D D.}_{F f B B} ((x x,, y the y)) \\ 11,, o o t t h h e e r r w w i i s the s e e \end{matrix}

S35: If QB(x, y) is 0, update the pixel information of the corresponding position in the image f _i (x, y) to its background image fv _B (x, y):

{R R}_{B B} ((x x,, y the y)) = = \frac{{R R}_{B B} ((x x,, y the y)) \times \times 77 + + {R R}_{S S} ((x x,, y the y))}{88}

{G G}_{B B} ((x x,, y the y)) = = \frac{{G G}_{B B} ((x x,, y the y)) \times \times 77 + + {G G}_{S S} ((x x,, y the y))}{88}

{B B}_{B B} ((x x,, y the y)) = = \frac{{B B}_{B B} ((x x,, y the y)) \times \times 77 + + {B B}_{S S} ((x x,, y the y))}{88}

If QB(x,y) is not 0, remove the foreground shadow and set the corresponding QB(x,y) to 0;

S36: For QB(x,y), the constrained stroke algorithm and the morphological method are used to fill the holes of the foreground target, remove isolated pixels and extremely small targets, and obtain the final image segmentation.

4. the tobacco field collection image segmentation method according to claim 1, is characterized in that, in the described step S3, the method for judging whether the i-th frame image has dim light is as follows:

{Σ Σ}_{x x = = 11}^{M m} {Σ Σ}_{y the y = = 11}^{N N} {fv fv}_{i i} ((x x,, y the y)) < < {Th Th}_{22}

In the formula, x and y respectively represent the pixel coordinates of a frame image with a size of M×N, Th ₂ =c ₂ ×M×N, and c ₂ is a constant coefficient.

5. The tobacco field acquisition image segmentation method according to claim 1, characterized in that, the process of light suppression in the step S4 is as follows:

\begin{matrix} {H h}_{11} ((u u,, v v)) = = (({γ γ}_{H h} - - {γ γ}_{L L})) [[11 - - exp exp (({- - c c}_{00} {((\frac{D D. ((u u,, v v))}{{D D.}_{00}}))}^{22}))]] + + {γ γ}_{L L} \\ D D. ((u u,, v v)) = = {[[{((u u - - \frac{M m}{22}))}^{22} + + {((v v - - \frac{N N}{22}))}^{22}]]}^{\frac{11}{22}} \end{matrix} . .

6. The tobacco field acquisition image segmentation method according to claim 1, wherein the process of performing illumination compensation in the step S4 is as follows:

\begin{matrix} {H h}_{11} ((u u,, v v)) = = (({γ γ}_{H h} - - {γ γ}_{L L})) [[11 - - exp exp ((- - {c c}_{00} {((\frac{D D. ((u u,, v v))}{{D D.}_{00}^{33}}))}^{22}))]] + + {γ γ}_{L L} \\ D D. ((u u,, v v)) = = {[[{((u u - - \frac{M m}{22}))}^{22} + + {((v v - - \frac{N N}{22}))}^{22}]]}^{\frac{11}{22}} \end{matrix} . .