CN105844651A - Image analyzing apparatus - Google Patents

Abstract

The invention discloses an image analyzing apparatus which comprises a pre-treatment module for images, an image extreme point detection module, an image characteristic point positioning module, a principal direction determining module and a characteristics extracting module. The image characteristic point positioning module determines the extreme points of characteristics points by sorting out the low contrast ratio points sensitive to noises and the unstable marginal points. The principal direction determining module connects any two adjacent peak values in the histograms of oriented gradients for the characteristic points to form a plurality of sub-line sections wherein adjacent sub-line sections with similar gradients are consolidated in the length direction to form line sections. The direction of the most optimal one of the line sections is regarded as the principal direction of the characteristic points. According to the invention, high identifying accuracy and high speed can be achieved.

Description

Image analysis device

Technical Field

The invention relates to the field of image analysis, in particular to an image analysis device.

Background

In the related art, the object to be detected in the image analysis device is often a face image of a person. However, when a wide range of contents stored in the memory is targeted, it is desirable to target a wide variety of object images such as vehicles, animals, buildings, graphics, and various articles. In addition, in order to process large-scale data, it is necessary to improve the analysis processing efficiency and accuracy.

Disclosure of Invention

In view of the above problems, the present invention provides an image analysis device with high recognition and detection speed and high accuracy.

The purpose of the invention is realized by adopting the following technical scheme:

an image analysis device for performing recognition detection on an image is provided, which includes:

(1) the image preprocessing module comprises an image conversion submodule for converting the color image into a gray image and an image filtering submodule for filtering the gray image, wherein the image gray conversion formula of the image conversion submodule is as follows:

$\begin{array}{c}I(x,y)=\frac{max\left(R\right(x,y),G(x,y),B(x,y\left)\right)+min\left(R\right(x,y),G(x,y),B(x,y))}{2}\\ +2\[\max \left(R\right(x,y),G(x,y),B(x,y\left)\right)-\min \left(R\left(x,y\right),G\left(x,y\right),B\left(x,y\right)\right)\]\end{array}$

wherein, R (x, y), G (x, y) and B (x, y) respectively represent the red, green and blue intensity values at the pixel point (x, y), and I (x, y) represents the gray value at the pixel point (x, y);

(2) the image extreme point detection module detects the position of each extreme point through a Gaussian difference scale space of an image established by convolution of a Gaussian difference operator and the image, when the values of a sampling point relative to 18 points corresponding to 8 adjacent points and upper and lower adjacent scales of the sampling point with the same scale are all large, the sampling point is a maximum point, when the values of the sampling point relative to 18 points corresponding to 8 adjacent points and upper and lower adjacent scales with the same scale are all small, the sampling point is a minimum point, and the simplified calculation formula of the Gaussian difference scale space is as follows:

D(x,y,σ)＝(G(x,kσ)-G(x,σ))*I'(x,y)+(G(y,kσ)-G(y,σ))*I'(x,y)

here, the

$G(x,\mathrm{\σ})=\frac{1}{\sqrt{2\mathrm{\π}}\mathrm{\σ}}{e}^{-x^2/2{\mathrm{\σ}}^2},G(y,\mathrm{\σ})=\frac{1}{\sqrt{2\mathrm{\π}}\mathrm{\σ}}{e}^{-y^2/2{\mathrm{\σ}}^2}$

Wherein D (x, y, σ) represents a gaussian difference scale space function, I' (x, y) is an image function output by the image transformation submodule,. indicates a convolution operation,. sigma.represents a scale space factor, G (x, σ), G (y, σ) are defined scale-variable gaussian functions, and k is a constant multiplication factor;

(3) the image feature point positioning module determines extreme points serving as feature points by eliminating low-contrast points sensitive to noise and unstable edge points in the extreme points, and comprises a first positioning submodule, a second positioning submodule and a third positioning submodule, wherein the first positioning submodule, the second positioning submodule and the third positioning submodule are sequentially connected and are used for accurately positioning the extreme points, the second positioning submodule is used for removing the low-contrast points, and the third positioning submodule is used for removing the unstable edge points, wherein:

a. the first positioning submodule performs quadratic Taylor expansion on the Gaussian difference scale space function and obtains the accurate position of an extreme point through derivation, and the scale space function of the extreme point is as follows:

$D\left(\hat{X}\right)=D(x,y,\mathrm{\σ})+\frac{\∂D{(x,y,\mathrm{\σ})}^T}{\∂x}\hat{X}$

wherein,scale space function representing extreme points, D (x, y, σ)^TIs an offset from the extreme point,representing the exact location of the extreme point;

b. the second positioning sub-module sequentially performs gray scale enhancement and normalization processing on the image output by the image conversion sub-module and then eliminates the low-contrast points, wherein the enhanced gray scale values are as follows:

here, the

The determination formula of the low-contrast point is as follows:

$D\left(\hat{X}\right)<T_1,T_1\&Element;\&lsqb;0.01,0.06\&rsqb;$

where I "(x, y) represents the gray value enhanced image function,for the correction coefficients containing local information, M is the maximum gray-scale value of the pixel, which is 255, M_HIs the mean of all pixels in the image with a gray value above 128, m_LIs the average of all pixels with a gray value below 128, # x (y) is the image processed by the image filtering sub-module, T₁Is a set threshold value;

c. the third positioning submodule obtains the main curvature of the extreme point by calculating a Hessian matrix H with the position scale of the extreme point being 2 × 2, and eliminates the main curvature ratio larger than a set threshold value T₂To eliminate said unstable edge points, wherein the threshold value T is₂Has a value range of [10,15 ]]The principal curvature ratio is determined by comparing ratios between eigenvalues of the matrix H;

preferably, the image analysis apparatus further includes:

(1) the main direction determining module comprises a connecting sub-module, a merging sub-module and a processing sub-module which are sequentially connected, wherein the connecting sub-module is used for connecting any two adjacent peak values in a gradient direction histogram of the feature points to form a plurality of sub-line segments, the merging sub-module is used for merging the sub-line segments which have similar slopes and are adjacent in the length direction to form a line segment, the processing sub-module is used for taking the direction of the optimal line segment in the line segments as the main direction of the feature points, and the judging formula of the optimal line segment is as follows:

$L_Y=L_{{\stackrel{\&OverBar;}{g}}_{\max }},{\stackrel{\&OverBar;}{g}}_{\max }=\max \left({\stackrel{\&OverBar;}{g}}_{L_n}\right),{\stackrel{\&OverBar;}{g}}_{L_n}=\frac{1}{k}\underset{k=1}{\overset{k}{\&Sigma;}}{g}_k,L_n\&Element;L_{\mathrm{\&upsi;}})$

wherein L is_YThe optimal line segment is represented by a line segment,is an average gradient value ofThe line segment of (a) is,is the average gradient value of the nth line segment of the plurality of line segments, g_kIs the k-th sub-line segment, L, in the n-th line segment_υThe line segment length in the plurality of line segments is larger than the average line segment length;

(2) and the feature extraction module rotates the neighborhood of the feature point according to the main direction and describes the feature point according to the rotated neighborhood so as to generate the descriptor of the feature point.

Further, the sub-line segments with similar slopes have slope differences smaller than a preset threshold T₃The sub-line segment of, the threshold value T₃Has a value range of (0, 0.1)]。

The invention has the beneficial effects that:

1. the set image preprocessing module considers the visual habit and the nonlinear relation of the perceptibility of human eyes to different colors and the color intensity, and can describe the image more accurately;

2. a simplified calculation formula of a Gaussian difference scale space is provided, the calculation amount is reduced, the calculation speed is improved, and the speed of image analysis is further improved;

3. the set image characteristic point positioning module removes low-contrast points and unstable edge points from extreme points, so that the effectiveness of the characteristic points is guaranteed, the gray value of the image is enhanced, the stability of the image can be greatly improved, the low-contrast points are removed more accurately, and the accuracy of image analysis is further improved;

4. the method comprises the steps of setting a main direction determining module, providing a judging formula of an optimal line segment, taking the direction of the optimal line segment in the line segment formed by connecting any two adjacent peak values in a gradient direction histogram of the characteristic point as the main direction of the characteristic point, wherein the line segment is more stable relative to the point, so that a descriptor of the characteristic point corresponding to an image has repeatability, the accuracy of the descriptor of the characteristic is improved, the image can be identified and detected more quickly and accurately, and the robustness is high.

Drawings

The invention is further illustrated by means of the attached drawings, but the embodiments in the drawings do not constitute any limitation to the invention, and for a person skilled in the art, other drawings can be obtained on the basis of the following drawings without inventive effort.

Fig. 1 is a schematic diagram of the connection of modules of the present invention.

Detailed Description

The invention is further described with reference to the following examples.

Example 1

Referring to fig. 1, the image analysis apparatus of the present embodiment includes:

(1) the image preprocessing module comprises an image conversion submodule for converting the color image into a gray image and an image filtering submodule for filtering the gray image, wherein the image gray conversion formula of the image conversion submodule is as follows:

$\begin{array}{c}I(x,y)=\frac{max\left(R\right(x,y),G(x,y),B(x,y\left)\right)+min\left(R\right(x,y),G(x,y),B(x,y))}{2}\\ +2\&lsqb;\max \left(R\right(x,y),G(x,y),B(x,y\left)\right)-\min \left(R\left(x,y\right),G\left(x,y\right),B\left(x,y\right)\right)\&rsqb;\end{array}$

wherein, R (x, y), G (x, y) and B (x, y) respectively represent the red, green and blue intensity values at the pixel point (x, y), and I (x, y) represents the gray value at the pixel point (x, y);

(2) the image extreme point detection module detects the position of each extreme point through a Gaussian difference scale space of an image established by convolution of a Gaussian difference operator and the image, when the values of a sampling point relative to 18 points corresponding to 8 adjacent points and upper and lower adjacent scales of the sampling point with the same scale are all large, the sampling point is a maximum point, when the values of the sampling point relative to 18 points corresponding to 8 adjacent points and upper and lower adjacent scales with the same scale are all small, the sampling point is a minimum point, and the simplified calculation formula of the Gaussian difference scale space is as follows:

D(x,y,σ)＝(G(x,kσ)-G(x,σ))*I'(x,y)+(G(y,kσ)-G(y,σ))*I'(x,y)

here, the

$G(x,\mathrm{\&sigma;})=\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;}}{e}^{-x^2/2{\mathrm{\&sigma;}}^2},G(y,\mathrm{\&sigma;})=\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;}}{e}^{-y^2/2{\mathrm{\&sigma;}}^2}$

Wherein D (x, y, σ) represents a gaussian difference scale space function, I' (x, y) is an image function output by the image transformation submodule,. indicates a convolution operation,. sigma.represents a scale space factor, G (x, σ), G (y, σ) are defined scale-variable gaussian functions, and k is a constant multiplication factor;

(3) the image feature point positioning module determines extreme points serving as feature points by eliminating low-contrast points sensitive to noise and unstable edge points in the extreme points, and comprises a first positioning submodule, a second positioning submodule and a third positioning submodule, wherein the first positioning submodule, the second positioning submodule and the third positioning submodule are sequentially connected and are used for accurately positioning the extreme points, the second positioning submodule is used for removing the low-contrast points, and the third positioning submodule is used for removing the unstable edge points, wherein:

a. the first positioning submodule performs quadratic Taylor expansion on the Gaussian difference scale space function and obtains the accurate position of an extreme point through derivation, and the scale space function of the extreme point is as follows:

$D\left(\hat{X}\right)=D(x,y,\mathrm{\&sigma;})+\frac{\&part;D{(x,y,\mathrm{\&sigma;})}^T}{\&part;x}\hat{X}$

wherein,scale space function representing extreme points, D (x, y, σ)^TIs an offset from the extreme point,representing the exact location of the extreme point;

b. the second positioning sub-module sequentially performs gray scale enhancement and normalization processing on the image output by the image conversion sub-module and then eliminates the low-contrast points, wherein the enhanced gray scale values are as follows:

here, the

The determination formula of the low-contrast point is as follows:

$D\left(\hat{X}\right)<T_1,T_1\&Element;\&lsqb;0.01,0.06\&rsqb;$

where I "(x, y) represents the gray value enhanced image function,for the correction coefficients containing local information, M is the maximum gray-scale value of the pixel, which is 255, M_HIs the mean of all pixels in the image with a gray value above 128, m_LIs the average of all pixels with a gray value below 128, # x (y) is the image processed by the image filtering sub-module, T₁Is a set threshold value;

c. the third positioning submodule obtains the main curvature of the extreme point by calculating a Hessian matrix H with the position scale of the extreme point being 2 × 2, and eliminates the main curvature ratio larger than a set threshold value T₂To eliminate said unstable edge points, wherein the threshold value T is₂Has a value range of [10,15 ]]The principal curvature ratio is determined by comparing ratios between eigenvalues of the matrix H;

preferably, the image analysis apparatus further includes:

(1) the main direction determining module comprises a connecting sub-module, a merging sub-module and a processing sub-module which are sequentially connected, wherein the connecting sub-module is used for connecting any two adjacent peak values in a gradient direction histogram of the feature points to form a plurality of sub-line segments, the merging sub-module is used for merging the sub-line segments which have similar slopes and are adjacent in the length direction to form a line segment, the processing sub-module is used for taking the direction of the optimal line segment in the line segments as the main direction of the feature points, and the judging formula of the optimal line segment is as follows:

$L_Y=L_{{\stackrel{\&OverBar;}{g}}_{\max }},{\stackrel{\&OverBar;}{g}}_{\max }=\max \left({\stackrel{\&OverBar;}{g}}_{L_n}\right),{\stackrel{\&OverBar;}{g}}_{L_n}=\frac{1}{k}\underset{k=1}{\overset{k}{\&Sigma;}}{g}_k,L_n\&Element;L_{\mathrm{\&upsi;}})$

wherein L is_YThe optimal line segment is represented by a line segment,is an average gradient value ofThe line segment of (a) is,is the average gradient value of the nth line segment of the plurality of line segments, g_kIs the k-th sub-line segment, L, in the n-th line segment_υThe line segment length in the plurality of line segments is larger than the average line segment length;

(2) and the feature extraction module rotates the neighborhood of the feature point according to the main direction and describes the feature point according to the rotated neighborhood so as to generate the descriptor of the feature point.

Further, the sub-line segments with similar slopes have slope differences smaller than a preset threshold T₃The sub-line segment of, the threshold value T₃Has a value range of (0, 0.1)]。

The image preprocessing module provided by the embodiment considers the visual habit and the nonlinear relation between the perceptibility of human eyes to different colors and the color intensity, and can describe the image more accurately; a simplified calculation formula of a Gaussian difference scale space is provided, the calculation amount is reduced, the calculation speed is improved, and the speed of image analysis is further improved; the set image characteristic point positioning module removes low-contrast points and unstable edge points from extreme points to ensure characteristic pointsThe method has the advantages that the gray value of the image is enhanced, so that the stability of the image can be greatly improved, low-contrast points can be removed more accurately, and the accuracy of image analysis is improved; the method comprises the steps that a main direction determining module is set, a judging formula of an optimal line segment is provided, the direction of the optimal line segment in the line segment formed by connecting any two adjacent peak values in a gradient direction histogram of the feature point is used as the main direction of the feature point, the line segment is more stable relative to the point, so that a descriptor of the feature point corresponding to an image has repeatability, the accuracy of the feature descriptor is improved, the image can be identified and detected more quickly and accurately, and the robustness is high; this embodiment takes the threshold value T₁＝0.01，T₂＝10，T₃The accuracy of the image analysis increased by 2% and the speed increased by 1% at 0.1.

Example 2

Referring to fig. 1, the image analysis apparatus of the present embodiment includes:

(1) the image preprocessing module comprises an image conversion submodule for converting the color image into a gray image and an image filtering submodule for filtering the gray image, wherein the image gray conversion formula of the image conversion submodule is as follows:

$\begin{array}{c}I(x,y)=\frac{max\left(R\right(x,y),G(x,y),B(x,y\left)\right)+min\left(R\right(x,y),G(x,y),B(x,y))}{2}\\ +2\&lsqb;\max \left(R\right(x,y),G(x,y),B(x,y\left)\right)-\min \left(R\left(x,y\right),G\left(x,y\right),B\left(x,y\right)\right)\&rsqb;\end{array}$

wherein, R (x, y), G (x, y) and B (x, y) respectively represent the red, green and blue intensity values at the pixel point (x, y), and I (x, y) represents the gray value at the pixel point (x, y);

(2) the image extreme point detection module detects the position of each extreme point through a Gaussian difference scale space of an image established by convolution of a Gaussian difference operator and the image, when the values of a sampling point relative to 18 points corresponding to 8 adjacent points and upper and lower adjacent scales of the sampling point with the same scale are all large, the sampling point is a maximum point, when the values of the sampling point relative to 18 points corresponding to 8 adjacent points and upper and lower adjacent scales with the same scale are all small, the sampling point is a minimum point, and the simplified calculation formula of the Gaussian difference scale space is as follows:

D(x,y,σ)＝(G(x,kσ)-G(x,σ))*I'(x,y)+(G(y,kσ)-G(y,σ))*I'(x,y)

here, the

$G(x,\mathrm{\&sigma;})=\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;}}{e}^{-x^2/2{\mathrm{\&sigma;}}^2},G(y,\mathrm{\&sigma;})=\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;}}{e}^{-y^2/2{\mathrm{\&sigma;}}^2}$

Wherein D (x, y, σ) represents a gaussian difference scale space function, I' (x, y) is an image function output by the image transformation submodule,. indicates a convolution operation,. sigma.represents a scale space factor, G (x, σ), G (y, σ) are defined scale-variable gaussian functions, and k is a constant multiplication factor;

(3) the image feature point positioning module determines extreme points serving as feature points by eliminating low-contrast points sensitive to noise and unstable edge points in the extreme points, and comprises a first positioning submodule, a second positioning submodule and a third positioning submodule, wherein the first positioning submodule, the second positioning submodule and the third positioning submodule are sequentially connected and are used for accurately positioning the extreme points, the second positioning submodule is used for removing the low-contrast points, and the third positioning submodule is used for removing the unstable edge points, wherein:

a. the first positioning submodule performs quadratic Taylor expansion on the Gaussian difference scale space function and obtains the accurate position of an extreme point through derivation, and the scale space function of the extreme point is as follows:

$D\left(\hat{X}\right)=D(x,y,\mathrm{\&sigma;})+\frac{\&part;D{(x,y,\mathrm{\&sigma;})}^T}{\&part;x}\hat{X}$

wherein,scale space function representing extreme points, D (x, y, σ)^TIs an offset from the extreme point,representing the exact location of the extreme point;

b. the second positioning sub-module sequentially performs gray scale enhancement and normalization processing on the image output by the image conversion sub-module and then eliminates the low-contrast points, wherein the enhanced gray scale values are as follows:

here, the

The determination formula of the low-contrast point is as follows:

$D\left(\hat{X}\right)<T_1,T_1\&Element;\&lsqb;0.01,0.06\&rsqb;$

where I "(x, y) represents the gray value enhanced image function,for the correction coefficients containing local information, M is the maximum gray-scale value of the pixel, which is 255, M_HIs the mean of all pixels in the image with a gray value above 128, m_LIs the average of all pixels with a gray value below 128, # x (y) is the image processed by the image filtering sub-module, T₁Is a set threshold value;

c. the third positioning submodule obtains the main curvature of the extreme point by calculating a Hessian matrix H with the position scale of the extreme point being 2 × 2, and eliminates the main curvature ratio larger than a set threshold value T₂To eliminate said unstable edge points, wherein the threshold value T is₂Has a value range of [10,15 ]]The principal curvature ratio is determined by comparing ratios between eigenvalues of the matrix H;

preferably, the image analysis apparatus further includes:

(1) the main direction determining module comprises a connecting sub-module, a merging sub-module and a processing sub-module which are sequentially connected, wherein the connecting sub-module is used for connecting any two adjacent peak values in a gradient direction histogram of the feature points to form a plurality of sub-line segments, the merging sub-module is used for merging the sub-line segments which have similar slopes and are adjacent in the length direction to form a line segment, the processing sub-module is used for taking the direction of the optimal line segment in the line segments as the main direction of the feature points, and the judging formula of the optimal line segment is as follows:

$L_Y=L_{{\stackrel{\&OverBar;}{g}}_{\max }},{\stackrel{\&OverBar;}{g}}_{\max }=\max \left({\stackrel{\&OverBar;}{g}}_{L_n}\right),{\stackrel{\&OverBar;}{g}}_{L_n}=\frac{1}{k}\underset{k=1}{\overset{k}{\&Sigma;}}{g}_k,L_n\&Element;L_{\mathrm{\&upsi;}})$

wherein L is_YThe optimal line segment is represented by a line segment,is an average gradient value ofThe line segment of (a) is,is the average gradient value of the nth line segment of the plurality of line segments, g_kIs the k-th sub-line segment, L, in the n-th line segment_υThe line segment length in the plurality of line segments is larger than the average line segment length;

(2) and the feature extraction module rotates the neighborhood of the feature point according to the main direction and describes the feature point according to the rotated neighborhood so as to generate the descriptor of the feature point.

Further, the sub-line segments with similar slopes have slope differences smaller than a preset threshold T₃The sub-line segment of, the threshold value T₃Has a value range of (0, 0.1)]。

The image preprocessing module provided by the embodiment considers the visual habit and the nonlinear relation between the perceptibility of human eyes to different colors and the color intensity, and can describe the image more accurately; proposes a Gaussian differenceThe simplified calculation formula of the scale space reduces the calculation amount, improves the calculation speed and further improves the speed of image analysis; the set image characteristic point positioning module removes low-contrast points and unstable edge points from extreme points, so that the effectiveness of the characteristic points is guaranteed, the gray value of the image is enhanced, the stability of the image can be greatly improved, the low-contrast points are removed more accurately, and the accuracy of image analysis is further improved; the method comprises the steps that a main direction determining module is set, a judging formula of an optimal line segment is provided, the direction of the optimal line segment in the line segment formed by connecting any two adjacent peak values in a gradient direction histogram of the feature point is used as the main direction of the feature point, the line segment is more stable relative to the point, so that a descriptor of the feature point corresponding to an image has repeatability, the accuracy of the feature descriptor is improved, the image can be identified and detected more quickly and accurately, and the robustness is high; this embodiment takes the threshold value T₁＝0.02，T₂＝11，T₃The accuracy of the image analysis increased by 1% and the speed increased by 1.5% for 0.08.

Example 3

Referring to fig. 1, the image analysis apparatus of the present embodiment includes:

(1) the image preprocessing module comprises an image conversion submodule for converting the color image into a gray image and an image filtering submodule for filtering the gray image, wherein the image gray conversion formula of the image conversion submodule is as follows:

$\begin{array}{c}I(x,y)=\frac{max\left(R\right(x,y),G(x,y),B(x,y\left)\right)+min\left(R\right(x,y),G(x,y),B(x,y))}{2}\\ +2\&lsqb;\max \left(R\right(x,y),G(x,y),B(x,y\left)\right)-\min \left(R\left(x,y\right),G\left(x,y\right),B\left(x,y\right)\right)\&rsqb;\end{array}$

wherein, R (x, y), G (x, y) and B (x, y) respectively represent the red, green and blue intensity values at the pixel point (x, y), and I (x, y) represents the gray value at the pixel point (x, y);

(2) the image extreme point detection module detects the position of each extreme point through a Gaussian difference scale space of an image established by convolution of a Gaussian difference operator and the image, when the values of a sampling point relative to 18 points corresponding to 8 adjacent points and upper and lower adjacent scales of the sampling point with the same scale are all large, the sampling point is a maximum point, when the values of the sampling point relative to 18 points corresponding to 8 adjacent points and upper and lower adjacent scales with the same scale are all small, the sampling point is a minimum point, and the simplified calculation formula of the Gaussian difference scale space is as follows:

D(x,y,σ)＝(G(x,kσ)-G(x,σ))*I'(x,y)+(G(y,kσ)-G(y,σ))*I'(x,y)

here, the

$G(x,\mathrm{\&sigma;})=\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;}}{e}^{-x^2/2{\mathrm{\&sigma;}}^2},G(y,\mathrm{\&sigma;})=\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;}}{e}^{-y^2/2{\mathrm{\&sigma;}}^2}$

Wherein D (x, y, σ) represents a gaussian difference scale space function, I' (x, y) is an image function output by the image transformation submodule,. indicates a convolution operation,. sigma.represents a scale space factor, G (x, σ), G (y, σ) are defined scale-variable gaussian functions, and k is a constant multiplication factor;

(3) the image feature point positioning module determines extreme points serving as feature points by eliminating low-contrast points sensitive to noise and unstable edge points in the extreme points, and comprises a first positioning submodule, a second positioning submodule and a third positioning submodule, wherein the first positioning submodule, the second positioning submodule and the third positioning submodule are sequentially connected and are used for accurately positioning the extreme points, the second positioning submodule is used for removing the low-contrast points, and the third positioning submodule is used for removing the unstable edge points, wherein:

a. the first positioning submodule performs quadratic Taylor expansion on the Gaussian difference scale space function and obtains the accurate position of an extreme point through derivation, and the scale space function of the extreme point is as follows:

$D\left(\hat{X}\right)=D(x,y,\mathrm{\&sigma;})+\frac{\&part;D{(x,y,\mathrm{\&sigma;})}^T}{\&part;x}\hat{X}$

wherein,scale space function representing extreme points, D (x, y, σ)^TIs an offset from the extreme point,representing the exact location of the extreme point;

b. the second positioning sub-module sequentially performs gray scale enhancement and normalization processing on the image output by the image conversion sub-module and then eliminates the low-contrast points, wherein the enhanced gray scale values are as follows:

here, the

The determination formula of the low-contrast point is as follows:

$D\left(\hat{X}\right)<T_1,T_1\&Element;\&lsqb;0.01,0.06\&rsqb;$

where I "(x, y) represents the gray value enhanced image function,for the correction coefficients containing local information, M is the maximum gray-scale value of the pixel, which is 255, M_HIs the mean of all pixels in the image with a gray value above 128, m_LIs the average of all pixels with a gray value below 128, # x (y) is the image processed by the image filtering sub-module, T₁Is a set threshold value;

c. the third positioning submodule obtains the main curvature of the extreme point by calculating a Hessian matrix H with the position scale of the extreme point being 2 × 2, and eliminates the main curvature ratio larger than a set threshold value T₂To eliminate said unstable edge points, wherein the threshold value T is₂Has a value range of [10,15 ]]The principal curvature ratio is determined by comparing ratios between eigenvalues of the matrix H;

preferably, the image analysis apparatus further includes:

(1) the main direction determining module comprises a connecting sub-module, a merging sub-module and a processing sub-module which are sequentially connected, wherein the connecting sub-module is used for connecting any two adjacent peak values in a gradient direction histogram of the feature points to form a plurality of sub-line segments, the merging sub-module is used for merging the sub-line segments which have similar slopes and are adjacent in the length direction to form a line segment, the processing sub-module is used for taking the direction of the optimal line segment in the line segments as the main direction of the feature points, and the judging formula of the optimal line segment is as follows:

$L_Y=L_{{\stackrel{\&OverBar;}{g}}_{\max }},{\stackrel{\&OverBar;}{g}}_{\max }=\max \left({\stackrel{\&OverBar;}{g}}_{L_n}\right),{\stackrel{\&OverBar;}{g}}_{L_n}=\frac{1}{k}\underset{k=1}{\overset{k}{\&Sigma;}}{g}_k,L_n\&Element;L_{\mathrm{\&upsi;}})$

wherein L is_YThe optimal line segment is represented by a line segment,is an average gradient value ofThe line segment of (a) is,is the average gradient value of the nth line segment of the plurality of line segments, g_kIs the k-th sub-line segment, L, in the n-th line segment_υThe line segment length in the plurality of line segments is larger than the average line segment length;

(2) and the feature extraction module rotates the neighborhood of the feature point according to the main direction and describes the feature point according to the rotated neighborhood so as to generate the descriptor of the feature point.

Further, the sub-line segments with similar slopes have slope differences smaller than a preset threshold T₃The sub-line segment of, the threshold value T₃Has a value range of (0, 0.1)]。

The image preprocessing module provided by the embodiment considers the visual habit and the nonlinear relation between the perceptibility of human eyes to different colors and the color intensity, and can describe the image more accurately; a simplified calculation formula of a Gaussian difference scale space is provided, the calculation amount is reduced, the calculation speed is improved, and the speed of image analysis is further improved; the set image characteristic point positioning module removes low-contrast points and unstable edge points from extreme points, so that the effectiveness of the characteristic points is guaranteed, the gray value of the image is enhanced, the stability of the image can be greatly improved, the low-contrast points are removed more accurately, and the accuracy of image analysis is further improved; the method comprises the steps that a main direction determining module is set, a judging formula of an optimal line segment is provided, the direction of the optimal line segment in the line segment formed by connecting any two adjacent peak values in a gradient direction histogram of the feature point is used as the main direction of the feature point, the line segment is more stable relative to the point, so that a descriptor of the feature point corresponding to an image has repeatability, the accuracy of the feature descriptor is improved, the image can be identified and detected more quickly and accurately, and the robustness is high; this embodiment takes the threshold value T₁＝0.03，T₂＝12，T₃The accuracy of the image analysis increased by 2.5% and the speed increased by 3% at 0.06%.

Example 4

Referring to fig. 1, the image analysis apparatus of the present embodiment includes:

(1) the image preprocessing module comprises an image conversion submodule for converting the color image into a gray image and an image filtering submodule for filtering the gray image, wherein the image gray conversion formula of the image conversion submodule is as follows:

$\begin{array}{c}I(x,y)=\frac{max\left(R\right(x,y),G(x,y),B(x,y\left)\right)+min\left(R\right(x,y),G(x,y),B(x,y))}{2}\\ +2\&lsqb;\max \left(R\right(x,y),G(x,y),B(x,y\left)\right)-\min \left(R\left(x,y\right),G\left(x,y\right),B\left(x,y\right)\right)\&rsqb;\end{array}$

wherein, R (x, y), G (x, y) and B (x, y) respectively represent the red, green and blue intensity values at the pixel point (x, y), and I (x, y) represents the gray value at the pixel point (x, y);

(2) the image extreme point detection module detects the position of each extreme point through a Gaussian difference scale space of an image established by convolution of a Gaussian difference operator and the image, when the values of a sampling point relative to 18 points corresponding to 8 adjacent points and upper and lower adjacent scales of the sampling point with the same scale are all large, the sampling point is a maximum point, when the values of the sampling point relative to 18 points corresponding to 8 adjacent points and upper and lower adjacent scales with the same scale are all small, the sampling point is a minimum point, and the simplified calculation formula of the Gaussian difference scale space is as follows:

D(x,y,σ)＝(G(x,kσ)-G(x,σ))*I'(x,y)+(G(y,kσ)-G(y,σ))*I'(x,y)

here, the

$G(x,\mathrm{\&sigma;})=\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;}}{e}^{-x^2/2{\mathrm{\&sigma;}}^2},G(y,\mathrm{\&sigma;})=\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;}}{e}^{-y^2/2{\mathrm{\&sigma;}}^2}$

Wherein D (x, y, σ) represents a gaussian difference scale space function, I' (x, y) is an image function output by the image transformation submodule,. indicates a convolution operation,. sigma.represents a scale space factor, G (x, σ), G (y, σ) are defined scale-variable gaussian functions, and k is a constant multiplication factor;

(3) the image feature point positioning module determines extreme points serving as feature points by eliminating low-contrast points sensitive to noise and unstable edge points in the extreme points, and comprises a first positioning submodule, a second positioning submodule and a third positioning submodule, wherein the first positioning submodule, the second positioning submodule and the third positioning submodule are sequentially connected and are used for accurately positioning the extreme points, the second positioning submodule is used for removing the low-contrast points, and the third positioning submodule is used for removing the unstable edge points, wherein:

a. the first positioning submodule performs quadratic Taylor expansion on the Gaussian difference scale space function and obtains the accurate position of an extreme point through derivation, and the scale space function of the extreme point is as follows:

$D\left(\hat{X}\right)=D(x,y,\mathrm{\&sigma;})+\frac{\&part;D{(x,y,\mathrm{\&sigma;})}^T}{\&part;x}\hat{X}$

wherein,scale space function representing extreme points, D (x, y, σ)^TIs an offset from the extreme point,representing the exact location of the extreme point;

b. the second positioning sub-module sequentially performs gray scale enhancement and normalization processing on the image output by the image conversion sub-module and then eliminates the low-contrast points, wherein the enhanced gray scale values are as follows:

here, the

The determination formula of the low-contrast point is as follows:

$D\left(\hat{X}\right)<T_1,T_1\&Element;\&lsqb;0.01,0.06\&rsqb;$

where I "(x, y) represents the gray value enhanced image function,for the correction coefficients containing local information, M is the maximum gray-scale value of the pixel, which is 255, M_HIs the mean of all pixels in the image with a gray value above 128, m_LIs the average of all pixels with a gray value below 128, # x (y) is the image processed by the image filtering sub-module, T₁Is a set threshold value;

c. the third positioning submodule obtains the main curvature of the extreme point by calculating a Hessian matrix H with the position scale of the extreme point being 2 × 2, and eliminates the main curvature ratio larger than a set threshold value T₂To eliminate said unstable edge points, wherein the threshold value T is₂Has a value range of [10,15 ]]The principal curvature ratio is determined by comparing ratios between eigenvalues of the matrix H;

preferably, the image analysis apparatus further includes:

(1) the main direction determining module comprises a connecting sub-module, a merging sub-module and a processing sub-module which are sequentially connected, wherein the connecting sub-module is used for connecting any two adjacent peak values in a gradient direction histogram of the feature points to form a plurality of sub-line segments, the merging sub-module is used for merging the sub-line segments which have similar slopes and are adjacent in the length direction to form a line segment, the processing sub-module is used for taking the direction of the optimal line segment in the line segments as the main direction of the feature points, and the judging formula of the optimal line segment is as follows:

$L_Y=L_{{\stackrel{\&OverBar;}{g}}_{\max }},{\stackrel{\&OverBar;}{g}}_{\max }=\max \left({\stackrel{\&OverBar;}{g}}_{L_n}\right),{\stackrel{\&OverBar;}{g}}_{L_n}=\frac{1}{k}\underset{k=1}{\overset{k}{\&Sigma;}}{g}_k,L_n\&Element;L_{\mathrm{\&upsi;}})$

wherein L is_YThe optimal line segment is represented by a line segment,is an average gradient value ofThe line segment of (a) is,is the average gradient value of the nth line segment of the plurality of line segments, g_kIs the k-th sub-line segment, L, in the n-th line segment_υThe line segment length in the plurality of line segments is larger than the average line segment length;

(2) and the feature extraction module rotates the neighborhood of the feature point according to the main direction and describes the feature point according to the rotated neighborhood so as to generate the descriptor of the feature point.

Further, the sub-line segments with similar slopes have slope differences smaller than a preset threshold T₃The sub-line segment of, the threshold value T₃Has a value range of (0, 0.1)]。

The image preprocessing module provided by the embodiment considers the visual habit and the nonlinear relation between the perceptibility of human eyes to different colors and the color intensity, and can describe the image more accurately; a simplified calculation formula of a Gaussian difference scale space is provided, the calculation amount is reduced, the calculation speed is improved, and the speed of image analysis is further improved; the set image characteristic point positioning module removes low-contrast points and unstable edge points from extreme points, so that the effectiveness of the characteristic points is guaranteed, the gray value of the image is enhanced, the stability of the image can be greatly improved, the low-contrast points are removed more accurately, and the accuracy of image analysis is further improved; the method comprises the steps that a main direction determining module is set, a judging formula of an optimal line segment is provided, the direction of the optimal line segment in the line segment formed by connecting any two adjacent peak values in a gradient direction histogram of the feature point is used as the main direction of the feature point, the line segment is more stable relative to the point, so that a descriptor of the feature point corresponding to an image has repeatability, the accuracy of the feature descriptor is improved, the image can be identified and detected more quickly and accurately, and the robustness is high; this embodiment takes the threshold value T₁＝0.04，T₂＝13，T₃The accuracy of the image analysis increased by 1.5% and the speed increased by 2%, 0.04.

Example 5

Referring to fig. 1, the image analysis apparatus of the present embodiment includes:

(1) the image preprocessing module comprises an image conversion submodule for converting the color image into a gray image and an image filtering submodule for filtering the gray image, wherein the image gray conversion formula of the image conversion submodule is as follows:

$\begin{array}{c}I(x,y)=\frac{max\left(R\right(x,y),G(x,y),B(x,y\left)\right)+min\left(R\right(x,y),G(x,y),B(x,y))}{2}\\ +2\&lsqb;\max \left(R\right(x,y),G(x,y),B(x,y\left)\right)-\min \left(R\left(x,y\right),G\left(x,y\right),B\left(x,y\right)\right)\&rsqb;\end{array}$

wherein, R (x, y), G (x, y) and B (x, y) respectively represent the red, green and blue intensity values at the pixel point (x, y), and I (x, y) represents the gray value at the pixel point (x, y);

(2) the image extreme point detection module detects the position of each extreme point through a Gaussian difference scale space of an image established by convolution of a Gaussian difference operator and the image, when the values of a sampling point relative to 18 points corresponding to 8 adjacent points and upper and lower adjacent scales of the sampling point with the same scale are all large, the sampling point is a maximum point, when the values of the sampling point relative to 18 points corresponding to 8 adjacent points and upper and lower adjacent scales with the same scale are all small, the sampling point is a minimum point, and the simplified calculation formula of the Gaussian difference scale space is as follows:

D(x,y,σ)＝(G(x,kσ)-G(x,σ))*I'(x,y)+(G(y,kσ)-G(y,σ))*I'(x,y)

here, the

$G(x,\mathrm{\&sigma;})=\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;}}{e}^{-x^2/2{\mathrm{\&sigma;}}^2},G(y,\mathrm{\&sigma;})=\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;}}{e}^{-y^2/2{\mathrm{\&sigma;}}^2}$

Wherein D (x, y, σ) represents a gaussian difference scale space function, I' (x, y) is an image function output by the image transformation submodule,. indicates a convolution operation,. sigma.represents a scale space factor, G (x, σ), G (y, σ) are defined scale-variable gaussian functions, and k is a constant multiplication factor;

(3) the image feature point positioning module determines extreme points serving as feature points by eliminating low-contrast points sensitive to noise and unstable edge points in the extreme points, and comprises a first positioning submodule, a second positioning submodule and a third positioning submodule, wherein the first positioning submodule, the second positioning submodule and the third positioning submodule are sequentially connected and are used for accurately positioning the extreme points, the second positioning submodule is used for removing the low-contrast points, and the third positioning submodule is used for removing the unstable edge points, wherein:

a. the first positioning submodule performs quadratic Taylor expansion on the Gaussian difference scale space function and obtains the accurate position of an extreme point through derivation, and the scale space function of the extreme point is as follows:

$D\left(\hat{X}\right)=D(x,y,\mathrm{\&sigma;})+\frac{\&part;D{(x,y,\mathrm{\&sigma;})}^T}{\&part;x}\hat{X}$

wherein,scale space function representing extreme points, D (x, y, σ)^TIs an offset from the extreme point,representing the exact location of the extreme point;

b. the second positioning sub-module sequentially performs gray scale enhancement and normalization processing on the image output by the image conversion sub-module and then eliminates the low-contrast points, wherein the enhanced gray scale values are as follows:

here, the

The determination formula of the low-contrast point is as follows:

$D\left(\hat{X}\right)<T_1,T_1\&Element;\&lsqb;0.01,0.06\&rsqb;$

where I "(x, y) represents the gray value enhanced image function,for the correction coefficients containing local information, M is the maximum gray-scale value of the pixel, which is 255, M_HIs the mean of all pixels in the image with a gray value above 128, m_LIs the average of all pixels with a gray value below 128, # x (y) is the image processed by the image filtering sub-module, T₁Is a set threshold value;

c. the third positioning submodule obtains the main curvature of the extreme point by calculating a Hessian matrix H with the position scale of the extreme point being 2 × 2, and eliminates the main curvature ratio larger than a set threshold value T₂To eliminate said unstable edge points, wherein the threshold value T is₂Has a value range of [10,15 ]]The principal curvature ratio is determined by comparing ratios between eigenvalues of the matrix H;

preferably, the image analysis apparatus further includes:

(1) the main direction determining module comprises a connecting sub-module, a merging sub-module and a processing sub-module which are sequentially connected, wherein the connecting sub-module is used for connecting any two adjacent peak values in a gradient direction histogram of the feature points to form a plurality of sub-line segments, the merging sub-module is used for merging the sub-line segments which have similar slopes and are adjacent in the length direction to form a line segment, the processing sub-module is used for taking the direction of the optimal line segment in the line segments as the main direction of the feature points, and the judging formula of the optimal line segment is as follows:

$L_Y=L_{{\stackrel{\&OverBar;}{g}}_{\max }},{\stackrel{\&OverBar;}{g}}_{\max }=\max \left({\stackrel{\&OverBar;}{g}}_{L_n}\right),{\stackrel{\&OverBar;}{g}}_{L_n}=\frac{1}{k}\underset{k=1}{\overset{k}{\&Sigma;}}{g}_k,L_n\&Element;L_{\mathrm{\&upsi;}})$

wherein L is_YThe optimal line segment is represented by a line segment,is an average gradient value ofThe line segment of (a) is,is the average gradient value of the nth line segment of the plurality of line segments, g_kIs the k-th sub-line segment, L, in the n-th line segment_υFor the length of one of said line segments being greater than the average lengthA line segment set;

(2) and the feature extraction module rotates the neighborhood of the feature point according to the main direction and describes the feature point according to the rotated neighborhood so as to generate the descriptor of the feature point.

Further, the sub-line segments with similar slopes have slope differences smaller than a preset threshold T₃The sub-line segment of, the threshold value T₃Has a value range of (0, 0.1)]。

The image preprocessing module provided by the embodiment considers the visual habit and the nonlinear relation between the perceptibility of human eyes to different colors and the color intensity, and can describe the image more accurately; a simplified calculation formula of a Gaussian difference scale space is provided, the calculation amount is reduced, the calculation speed is improved, and the speed of image analysis is further improved; the set image characteristic point positioning module removes low-contrast points and unstable edge points from extreme points, so that the effectiveness of the characteristic points is guaranteed, the gray value of the image is enhanced, the stability of the image can be greatly improved, the low-contrast points are removed more accurately, and the accuracy of image analysis is further improved; the method comprises the steps that a main direction determining module is set, a judging formula of an optimal line segment is provided, the direction of the optimal line segment in the line segment formed by connecting any two adjacent peak values in a gradient direction histogram of the feature point is used as the main direction of the feature point, the line segment is more stable relative to the point, so that a descriptor of the feature point corresponding to an image has repeatability, the accuracy of the feature descriptor is improved, the image can be identified and detected more quickly and accurately, and the robustness is high; this embodiment takes the threshold value T₁＝0.05，T₂＝14，T₃The accuracy of the image analysis increased by 1.8% and the speed increased by 1.5%, 0.02.

Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the protection scope of the present invention, although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions can be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims (3)

1. An image analysis device for recognizing and detecting an image, comprising:

(1) the image preprocessing module comprises an image conversion submodule for converting the color image into a gray image and an image filtering submodule for filtering the gray image, wherein the image gray conversion formula of the image conversion submodule is as follows:

$\begin{array}{c}I\left(x,y\right)=\frac{max\left(R\right(x,y),G(x,y),B(x,y))+min\left(R\right(x,y),G(x,y),B(x,y))}{2}\\ +2\&lsqb;\max \left(R\right(x,y),G(x,y),B(x,y))+min\left(R\right(x,y),G(x,y),B(x,y))\&rsqb;\end{array}$

wherein, R (x, y), G (x, y) and B (x, y) respectively represent the red, green and blue intensity values at the pixel point (x, y), and I (x, y) represents the gray value at the pixel point (x, y);

(2) the image extreme point detection module detects the position of each extreme point through a Gaussian difference scale space of an image established by convolution of a Gaussian difference operator and the image, when the values of a sampling point relative to 18 points corresponding to 8 adjacent points and upper and lower adjacent scales of the sampling point with the same scale are all large, the sampling point is a maximum point, when the values of the sampling point relative to 18 points corresponding to 8 adjacent points and upper and lower adjacent scales with the same scale are all small, the sampling point is a minimum point, and the simplified calculation formula of the Gaussian difference scale space is as follows:

D(x,y,σ)＝(G(x,kσ)-G(x,σ))*I'(x,y)+(G(y,kσ)-G(y,σ))*I'(x,y)

here, the

$G(x,\mathrm{\&sigma;})=\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;}}{e}^{-x^2/2{\mathrm{\&sigma;}}^2},G(y,\mathrm{\&sigma;})=\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;}}{e}^{-y^2/2{\mathrm{\&sigma;}}^2}$

Wherein D (x, y, σ) represents a gaussian difference scale space function, I' (x, y) is an image function output by the image transformation submodule,. indicates a convolution operation,. sigma.represents a scale space factor, G (x, σ), G (y, σ) are defined scale-variable gaussian functions, and k is a constant multiplication factor;

(3) the image feature point positioning module determines extreme points serving as feature points by eliminating low-contrast points sensitive to noise and unstable edge points in the extreme points, and comprises a first positioning submodule, a second positioning submodule and a third positioning submodule, wherein the first positioning submodule, the second positioning submodule and the third positioning submodule are sequentially connected and are used for accurately positioning the extreme points, the second positioning submodule is used for removing the low-contrast points, and the third positioning submodule is used for removing the unstable edge points, wherein:

a. the first positioning submodule performs quadratic Taylor expansion on the Gaussian difference scale space function and obtains the accurate position of an extreme point through derivation, and the scale space function of the extreme point is as follows:

$D\left(\hat{X}\right)=D(x,y,\mathrm{\&sigma;})+\frac{\&part;D{(x,y,\mathrm{\&sigma;})}^T}{\&part;x}\hat{X}$

wherein,scale space function representing extreme points, D (x, y, σ)^TIs an offset from the extreme point,representing the exact location of the extreme point;

b. the second positioning sub-module sequentially performs gray scale enhancement and normalization processing on the image output by the image conversion sub-module and then eliminates the low-contrast points, wherein the enhanced gray scale values are as follows:

here, the

The determination formula of the low-contrast point is as follows:

$D\left(\hat{X}\right)<T_1,T_1\&Element;\&lsqb;0.01,0.06\&rsqb;$

where I "(x, y) represents the gray value enhanced image function,for the correction coefficients containing local information, M is the maximum gray-scale value of the pixel, which is 255, M_HIs the mean of all pixels in the image with a gray value above 128, m_LIs the average of all pixels with a gray value below 128, # x (y) is the image processed by the image filtering sub-module, T₁Is a set threshold value;

c. the third positioning submodule obtains the main curvature of the extreme point by calculating a Hessian matrix H with the position scale of the extreme point being 2 × 2, and eliminates the main curvature ratio larger than a set threshold value T₂To eliminate said unstable edge points, wherein the threshold value T is₂Has a value range of [10,15 ]]The principal curvature ratio is determined by comparing the ratios between eigenvalues of the matrix H.

2. An image analysis apparatus according to claim 1, further comprising:

(1) the main direction determining module comprises a connecting sub-module, a merging sub-module and a processing sub-module which are sequentially connected, wherein the connecting sub-module is used for connecting any two adjacent peak values in a gradient direction histogram of the feature points to form a plurality of sub-line segments, the merging sub-module is used for merging the sub-line segments which have similar slopes and are adjacent in the length direction to form a line segment, the processing sub-module is used for taking the direction of the optimal line segment in the line segments as the main direction of the feature points, and the judging formula of the optimal line segment is as follows:

$L_Y=L_{{\stackrel{\&OverBar;}{g}}_{max}},{\stackrel{\&OverBar;}{g}}_{max}=max\left({\stackrel{\&OverBar;}{g}}_{L_n}\right),{\stackrel{\&OverBar;}{g}}_{L_n}=\frac{1}{k}\underset{k=1}{\overset{k}{\&Sigma;}}{g}_k,L_n\&Element;L_{\mathrm{\&upsi;}})$

wherein L is_YThe optimal line segment is represented by a line segment,is an average gradient value ofThe line segment of (a) is,is the average gradient value of the nth line segment of the plurality of line segments, g_kIs the k-th sub-line segment, L, in the n-th line segment_υThe line segment length in the plurality of line segments is larger than the average line segment length;

(2) and the feature extraction module rotates the neighborhood of the feature point according to the main direction and describes the feature point according to the rotated neighborhood so as to generate the descriptor of the feature point.

3. The image analysis apparatus according to claim 1, wherein the sub-line segments having similar slopes have a slope difference smaller than a predetermined thresholdT₃The sub-line segment of, the threshold value T₃Has a value range of (0, 0.1)]。