CN103218822B - Based on the image characteristic point automatic testing method of disappearance importance - Google Patents

Abstract

The present invention relates to a kind of image characteristic point automatic testing method based on disappearance importance, comprising: gather image, input computing machine and be translated into gray level image; The average disappearance importance at each point place in computed image; The standard deviation disappearance importance at each point place in computed image; Utilize average to lack importance and carry out marginal point mark; Utilize standard deviation to lack importance and carry out unique point mark; Export the image characteristic point marked.Compared to existing method, method provided by the invention has clear superiority in positioning precision.

Description

Automatic Detection Method of Image Feature Points Based on Missing Importance

technical field

The invention relates to the field of automatic detection of image features in computer vision, in particular to an automatic detection method of image feature points.

Background technique

Feature point automatic detection technology has important applications in many fields such as image retrieval, object recognition, video tracking and augmented reality. In recent years, many feature point automatic detection algorithms have emerged, and the more representative algorithms include: (1) the SUSAN method proposed in the paper "SUSAN–A New Approach to Low Level Image Processing. International Journal of Computer Vision. 1997, 23(1): 45-78"; (2 ) The CSS method proposed in the paper "RobustImageCornerDetectionThroughCurvatureScaleSpace.IEEETransonPatternAnalysisandMachineIntelligence.1998,20(12):1376-1381"; :2932-2942" proposed LOD method.

Among the above methods, the SUSAN method is more sensitive to image noise and threshold selection; although the LOD method has high accuracy in detecting feature points, it is inefficient due to its many steps and the need for data fitting; the CCS algorithm is due to its excellent It is the most commonly used detection method at present. The basic algorithm steps of the CSS method are: Step 1, use the Canny edge detection operator to perform edge detection; Step 2, connect intermittent and incomplete edges into complete edges on the edge graph; Detect the maximum point of curvature; step 4, find the precise position of the feature point by tracking in the scale space.

The main problem of this method is the problem caused by the use of the Canny operator in step 1: (1) The Gaussian filter used by the Canny edge detection operator causes the image edge feature position to shift, so step 4 is required to track the feature points in the multi-scale space Accurate position, on the one hand, is relatively complex to achieve, and on the other hand, the position obtained by tracking is still not very accurate; (2) Gaussian filtering performed during the execution of the Canny operator often results in broken and incomplete edges, so the second step is required. Edges are reconnected, and the connection process often leads to feature point position shifts, missing, errors, and ultimately affects the accuracy of feature detection.

Contents of the invention

The invention mainly solves the problem of automatic detection of feature points in digital images, and aims to provide a simple and higher-accuracy automatic detection method of feature points that does not require Gaussian filtering. To achieve this purpose, the method provided by the invention mainly includes the following steps:

Step S1: collecting images, inputting them into a computer and converting them into grayscale images;

Step S2: Calculate the mean-missing importance at each point in the image;

Step S3: Calculate the missing importance of the standard deviation at each point in the image;

Step S4: mark the edge points by using the mean-missing importance;

Step S5: mark the feature points by using the missing importance of the standard deviation;

Step S6: output the feature points marked in step S5.

The automatic detection method of image feature points based on missing importance provided by the present invention is improved on the basis of inheriting the basic idea of CSS algorithm "first perform edge detection and then perform feature point detection on the edge map". Compared with the CSS algorithm, this method no longer uses the Canny edge detection operator for edge detection through the mean importance and standard deviation missing importance defined based on statistics, and avoids the inaccurate and inaccurate final edge detection results caused by Gaussian filtering. Integrity, and finally guarantee the accuracy and completeness of feature point detection on the edge map. In addition, since it is no longer necessary to track the positions of the feature points in the scale space, compared with the existing CSS method, the method provided by the present invention is simpler and easier to implement.

Description of drawings

FIG. 1 is a flowchart of the automatic detection method of image feature points based on missing importance in the present invention.

detailed description

FIG. 1 is a flow chart of the automatic detection method of image feature points based on missing importance in the present invention. The main steps of the method provided by the invention include: collecting an image, inputting it into a computer and converting it into a grayscale image; calculating the missing importance of the mean value at each point in the image; calculating the missing importance of the standard deviation at each point in the image; using The mean missing importance is used to mark edge points; the standard deviation missing importance is used to mark feature points; the marked image feature points are output. The specific implementation details of each step are as follows:

Step S1: collecting images, inputting them into a computer and converting them into grayscale images;

Step S2: Calculate the mean-missing importance at each point in the image. The specific method is: for any position X(x,y) in the image, first set the point X(x,y) as the center and the circle with radius R The shape area is determined as the support area of X(x,y) and recorded as Ω(X), and then the average value of the gray value of each pixel in Ω(X) is calculated and recorded as m ₁ (X), and then the calculation of Ω( After removing the point X(x,y) in X), the average value of the gray value of each pixel is recorded as m ₂ (X), and finally m(X)=|m ₁ (X)-m ₂ (X)| Defined as the lack of mean value at point X(x,y), the value range of R when defining the support area is 1~3;

Step S3: Calculate the missing importance of the standard deviation at each point in the image. The specific method is: for any position X(x,y) in the image, first determine the support of point X(x,y) according to the method described in step S2 Area Ω(X), then calculate the standard deviation of the gray value of each pixel in Ω(X) and record it as s ₁ (X), and then calculate the pixel points after removing point X(x,y) in Ω(X) The standard deviation of the gray value is recorded as s ₂ (X), and finally s(X)=|s ₁ (X)-s ₂ (X)| is defined as the standard deviation at point X(x,y). When defining the support area, the value of R ranges from 1 to 3;

Step S4: Use the mean missing importance to mark edge points, the specific method is: first calculate the threshold T=k Mean(M), where Mean(M) represents the mean missing importance of each point in the entire image calculated in step S2 Mean value, the value range of k is 2~5, and then, if the mean value loss importance at a certain point in the image is greater than T, the point is marked as an edge point;

Step S5: Use the missing standard deviation importance to mark feature points. The specific method is: for the missing standard deviation importance at each point obtained in step S3, set the missing standard deviation importance corresponding to the position not marked as an edge point in step S4. is 0, then, if the standard deviation missing importance at a point in the image is the maximum value within the 5×5 neighborhood of the point, then mark the point as a feature point;

Step S6: output the feature points marked in step S5.

The automatic detection method of image feature points based on missing importance provided by the present invention is improved on the basis of inheriting the basic idea of CSS algorithm "first perform edge detection and then perform feature point detection on the edge map". Compared with the CSS algorithm, this method no longer uses the Canny edge detection operator for edge detection by defining the importance of missing mean and standard deviation based on statistics, and avoids inaccurate edge detection results caused by Gaussian filtering. And incomplete, and finally ensure the accuracy and completeness of feature point detection on the edge map. In addition, since it is no longer necessary to track the positions of the feature points in the scale space, compared with the existing CSS method, the method provided by the present invention is simpler and easier to implement.

Claims (1)

1., based on an image characteristic point automatic testing method for disappearance importance, it is characterized in that, comprise step:

Step S1: gather image, input computing machine and be converted into gray level image;

Step S2: the average disappearance importance calculating each point place in gray level image, concrete mode is: for any position X (x in gray level image, y), first will with an X (x, y) centered by, radius is that the border circular areas of R is defined as X (x, y) supporting zone is also designated as Ω (X), then calculates the mean value of each pixel gray-scale value in Ω (X) and is designated as m ₁(X), then calculate in Ω (X) remove an X (x, y) afterwards each pixel gray-scale value mean value and be designated as m ₂(X), finally by m (X)=| m ₁(X)-m ₂(X) | be defined as the average disappearance importance at X (x, a y) place, during definition supporting zone, the span of R is 1 ~ 3;

Step S3: the standard deviation disappearance importance calculating each point place in gray level image, concrete mode is: for any position X (x in gray level image, y), first an X (x is determined according to mode described in step S2, y) supporting zone Ω (X), then calculates the standard deviation of each pixel gray-scale value in Ω (X) and is designated as s ₁(X), then calculate in Ω (X) remove an X (x, y) afterwards each pixel gray-scale value standard deviation and be designated as s ₂(X), finally by s (X)=| s ₁(X)-s ₂(X) | be defined as the standard deviation disappearance importance at X (x, a y) place, during definition supporting zone, the span of R is 1 ~ 3;

Step S4: utilize average to lack importance and carry out marginal point mark, concrete mode is: first calculated threshold T=kMean (M), wherein Mean (M) represents the average of each point place average disappearance importance in the whole gray level image that step S2 calculates, and the span of k is 2 ~ 5; Then, if the average disappearance importance at certain some place is greater than T in gray level image, then this point is labeled as marginal point;

Step S5: utilize standard deviation to lack importance and carry out unique point mark, concrete mode is: for the standard deviation disappearance importance at step S3 gained each point place, standard deviation disappearance importance step S4 not being labeled as the position of marginal point corresponding is set to 0, then, if the standard deviation disappearance importance at certain some place is maximal value in 5 × 5 neighborhoods of this point in gray level image, then this point is labeled as unique point;

Step S6: export the unique point that step S5 marks.