Method for detecting a corner point and device therefor

Abstract

The application discloses a kind of corner point detection method and its device, the detection method includes the following steps: obtaining the picture to be detected, initializes angle set;The edge of the picture to be detected is detected and the edge graph is obtained, and curve is extracted from the edge graph;Each curve is processed as follows: fill the gap in curve, find several T joints in curve and add all the T joints found to angle set;The curve c j It is handled, the corner point in curve is identified, and the corner point identified is added to angle set, and the corner point in angle set S is the corner point in the picture to be detected.The detection method has the advantages of consistent with real curvature, angle convex / concave small and low computational complexity.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and apparatus for detecting corner points. Background Technology

[0002] Corner detection is an important task in many image analysis applications, such as shape representation, image matching and registration, and camera calibration. Many corner detectors have been developed, which can be broadly classified into two categories: intensity-based methods and contour-based methods.

[0003] Intensity-based methods detect corners by utilizing the intensity values ​​of pixels in the input image. The Harris detector uses the autocorrelation matrix of local intensity variations to measure the difference between corners and non-corners. Smith and Brady developed a corner measurement method called the Minimum Single-Value Segment Assimilation Kernel (SUSAN), which identifies corners based on the number of pixels similar to the center pixel within a circular template. Among these, chord-point distance-based methods are promising, representing a class of discrete curvature models for processing digital curves that have attracted significant attention due to their robustness to noise and ease of encoding. However, existing models of this type typically encounter three problems: inconsistency with true curvature (i.e., curvature as defined mathematically), convex / concave corner losses, and high computational complexity. Summary of the Invention

[0004] In view of this, the main objective of the present invention is to provide a method and apparatus for detecting corner points.

[0005] To achieve the above objectives, the technical solution of the present invention is implemented as follows: a corner detection method, comprising the following steps: acquiring an image to be detected (Image), initializing a corner set. Edge detection is performed on the image to be detected to obtain an edge map, and curves are extracted from the edge map. , ,..., Where M is a natural number; for any curve All were processed as follows: Fill curve Find the curve in the gap. Several T-junctions are found and all T-junctions are added to the corner set S; for the curve Smoothing is performed on the curve. Represented as point ,point ,...,point Calculate any point Corresponding curvature product Then, based on the curvature product , ,..., Identify curves The corner points in the set are identified and added to the corner set S, where... i and j are both natural numbers; 1 ≤ i ≤ j , 1≤j≤M; angle set The corner points in the image are all corner points in the image to be detected.

[0006] As an improvement to this embodiment of the invention, the step of performing edge detection on the image to be detected and obtaining an edge map specifically includes: performing edge detection on the image to be detected based on a Canny detector and obtaining an edge map. An edge detector is a high-pass filter that can enhance high-frequency components and suppress low-frequency components. Since edges and noise are both high-frequency components, edge detectors often amplify noise.

[0007] As an improvement to an embodiment of the present invention, the calculation of any point Corresponding curvature product Specifically, this includes: use Measurement points DCPD curvature value ,use Measurement points DCPD curvature value ,use Measurement points DCPD curvature value Based on DCPD curvature value , and Calculation points Corresponding curvature product ,in, , and All are natural numbers, 1≤ , and <i, function , , and Points , and x-coordinate , and Points , and The ordinate.

[0008] As an improvement to this embodiment of the invention, the DCPD curvature value-based , and Calculation points Corresponding curvature product Specifically, it includes: .

[0009] As an improvement to an embodiment of the present invention, the curvature product-based , ,..., Identify curves The corner points specifically include: through the product of curvature , ,..., Non-maximum suppression is performed to obtain several corner points with local maximum curvature values. Then, corner points with curvature values ​​greater than a predetermined threshold T are selected as curves. The corner points in the middle.

[0010] This invention also provides a corner detection device, comprising the following step modules: an initialization module, used to acquire the image to be detected (Image) and initialize the corner set. The processing module is used to perform edge detection on the image to be detected (Image) and obtain an edge map, and extract curves from the edge map. , ,..., Where M is a natural number; for any curve All were processed as follows: Fill curve Find the curve in the gap. Several T-junctions are found and all T-junctions are added to the corner set S; for the curve Smoothing is performed on the curve. Represented as point ,point ,...,point Calculate any point Corresponding curvature product Then, based on the curvature product , ,..., Identify curves The corner points in the set are identified and added to the corner set S, where... i and j are both natural numbers; 1 ≤ i ≤ j , 1≤j≤M; Corner module, used for corner sets The corner points in the image are all corner points in the image to be detected.

[0011] As an improvement to an embodiment of the present invention, the processing module is further configured to: perform edge detection on the image to be detected Image based on the Canny detector and obtain an edge map.

[0012] As an improvement to this embodiment of the invention, the processing module is further configured to: use Measurement points DCPD curvature value ,use Measurement points DCPD curvature value ,use Measurement points DCPD curvature value Based on DCPD curvature value , and Calculation points Corresponding curvature product ,in, , and All are natural numbers, 1≤ , and <i, function , , and Points , and x-coordinate , and Points , and The ordinate.

[0013] As an improvement to this embodiment of the invention, the processing module is further configured to: .

[0014] As an improvement to this embodiment of the invention, the processing module is further configured to: multiply the curvature product , ,..., Non-maximum suppression is performed to obtain several corner points with local maximum curvature values. Then, corner points with curvature values ​​greater than a predetermined threshold T are selected as curves. The corner points in the middle.

[0015] The corner detection method and apparatus provided in this invention have the following advantages: This invention discloses a corner detection method and apparatus, which includes the following steps: acquiring an image to be detected and initializing a corner set; performing edge detection on the image to be detected and obtaining an edge map, and extracting curves from the edge map; performing the following processing on each curve: filling the gaps in the curve, finding several T-junctions in the curve and adding all the found T-junctions to the corner set; and processing the curves... The process involves smoothing the curve, identifying corner points, and adding the identified corners to a corner set. The corner points in the image are all corner points in the image to be detected. This detection method has the advantages of consistent curvature, small corner convexity / concavity, and low computational complexity. Attached Figure Description

[0016] Figure 1 , Figure 2 , Figure 3A , Figure 3B , Figure 4 , Figure 5 , Figure 6A , Figure 6B , Figure 7A , Figure 7B , Figure 8 , Figure 9 , Figure 10 , Figure 11 , Figure 12 , Figure 13A , Figure 13B , Figure 14A , Figure 14B , Figure 14C , Figure 14D , Figure 14E , Figure 15A and Figure 15B This is a schematic diagram of the corner detection method in the embodiment;

[0017] Figure 16 , Figure 17 , Figure 18 , Figure 19 , Figure 20 , Figure 21 and Figure 22 The figure shows the experimental results of the corner detection method in the embodiment;

[0018] Figure 23 This is a flowchart illustrating the corner detection method in the embodiment. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the embodiments shown in the accompanying drawings. However, these embodiments do not limit the present invention, and any structural, methodological, or functional modifications made by those skilled in the art based on these embodiments are included within the scope of protection of the present invention.

[0020] The following description and accompanying drawings fully illustrate specific embodiments described herein to enable those skilled in the art to practice them. Some embodiments may include or substitute parts and features of other embodiments. The scope of the embodiments herein encompasses the entire scope of the claims and all available equivalents thereof. Throughout this document, the terms “first,” “second,” etc., are used only to distinguish one element from another without requiring or implying any actual relationship or order between the elements. Indeed, a first element can also be referred to as a second element, and vice versa. Furthermore, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a structure, apparatus, or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a structure, apparatus, or device. Without further limitation, an element defined by the phrase “comprising one…” does not exclude the presence of other identical elements in the structure, apparatus, or device that includes said element. The various embodiments described herein are presented in a progressive manner, with each embodiment focusing on its differences from other embodiments; similar or identical parts between embodiments can be referred to interchangeably.

[0021] The terms "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer" used in this document to indicate orientation or positional relationships are based on the orientation or positional relationships shown in the accompanying drawings and are used only for the convenience of describing this document and simplifying the description. They do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. In the description herein, unless otherwise specified and limited, the terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two elements, or direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.

[0022] Embodiment 1 of the present invention provides a method for detecting corner points, such as... Figure 23 As shown, it includes the following steps:

[0023] Step 2301: Obtain the image to be detected and initialize the angle set. ;

[0024] Step 2302: Perform edge detection on the image to be detected and obtain an edge map, then extract curves from the edge map. , ,..., Where M is a natural number; for any curve All were processed as follows: Fill curve Find the curve in the gap. Several T-junctions are found and all T-junctions are added to the corner set S; for the curve Smoothing is performed on the curve. Represented as point ,point ,...,point Calculate any point Corresponding curvature product Then, based on the curvature product , ,..., Identify curves The corner points in the set are identified and added to the corner set S, where... i and j are both natural numbers; 1 ≤ i ≤ j , 1≤j≤M;

[0025] Step 2303: Angle Set The corner points in the image are all corner points in the image to be detected.

[0026] The basic idea of ​​chord point distance is to move along the input curve across a fixed number of points, and then measure the discrete curvature of the curve based on the perpendicular distances from these points on the arc to their associated chords (i.e., the straight line segments connecting the two ends of the arc).

[0027] like Figure 1 As shown, in the discrete case, the input curve is formed by a set of points. To represent this, the chord of the curve is a straight line segment whose endpoints are... and All points lie on curve L, separated by a distance of L pixels. Let L denote the chord, for measuring points... The curvature at that point, the chord L moves along the curve while maintaining... Let k be an interior point, meaning 1 ≤ k ≤ L-1. Then, calculate the values ​​of k for k = 1, 2, ..., L-1 respectively. The distance to chord L is denoted as Finally, click. Maximum chord distance (ACDM) at a given location: Represented as The calculation is as follows: .

[0028] Chord-to-point distance accumulation (CPDA) moves the chord along the curve, and then the CPDA curvature is calculated as follows: The sum, that is .

[0029] To improve the robustness of CPDA, this detection method calculates CPDA at three different scales, i.e., using three different parameter L values, to estimate the curvature of the curve at each point. This represents the value of the parameter used (j=1, 2, and 3). At the th tick... The CPDA curvature measured at the location was first normalized: j are 1, 2, and 3 respectively. Next, the product of the normalized CPDA values ​​is... .

[0030] The curve chord ratio (CCR) will be the ratio of the points... and The ratio of chord length to arc length is used as an estimate of discrete curvature, such as... Figure 2 As shown. CCR can be represented as follows: ,in, .

[0031] Here, we first examine the problems encountered by existing chord-to-point distance models and then explain the inventors' motivation. Next, we propose a new curvature model that fully addresses the problems of existing models.

[0032] The motivation is that a reasonable discrete curvature model should exhibit behavior consistent with true curvature (i.e., curvature as defined in mathematics). A key characteristic of true curvature is that small-angle bends have higher curvature than large-angle bends. Unfortunately, this fundamental property is frequently violated by existing models. Figure 3A and Figure 3B This demonstrates the issue regarding CPDA. For example... Figure 3A As shown, the input curve used for testing has three corners, A, B, and C, where the angle value of corner C is smaller than the other two corners. However, when using CPDA, the discrete curvature measured at corner C is lower than the curvature at A and B, as shown... Figure 3B As shown. This problem makes it difficult to identify significant corners, resulting in poor corner detection performance.

[0033] Typical problems with existing discrete curvature models: such as Figure 3A The test curves shown are marked with a dot at the starting point and an arrow indicating the direction of curve tracing. Figure 3B As shown, the curve's discrete curvature was measured using CPDA. The problem is that, although the angle value of corner "C" is small, its measured curvature is lower than that of the other two corners "A" and "B".

[0034] To further investigate the above problem, a model containing only a single corner was used. Sharp corner patterns, such as Figure 4 As shown. At the corner Discrete curvature measured at [location] The calculation is as follows: ,in, This is the angle value of the corner, and L is a parameter in CPDA used to determine the arc length (in pixels) required to measure curvature. Without loss of generality, assume... Then, the angle value was studied. The measured curvature as it increases How it changes. Figure 5 Drawn and The relationship diagram is shown, where parameter L is set to 10 pixels. Although, according to the aforementioned mathematical properties, the actual curvature of the corner should vary with the angle value... The curvature decreases monotonically as it increases, but it can be seen that the measured curvature exhibits inconsistent and illogical behavior—that is, it increases first and then decreases. Figure 5 Demonstrates the use of CPDA in Figure 4 The discrete curvature measured at the corner point of the sharp corner pattern shown is plotted for different angle values.

[0035] A second key characteristic of true curvature is that a flat curve should have low curvature at every point. Unfortunately, this property is often violated by existing chord-to-point distance models. To illustrate this, Figure 6A and Figure 6B An illustrative example is given. Figure 6A The highlighted curve in the edge plot shown is quite flat, so its true curvature is close to zero at every point. However, the measured discrete curvature varies greatly from one point to another, and moreover, the curvature values ​​at many points are extremely high (e.g., close to 1), such as... Figure 6B As shown, the measured curvature values ​​range between [0, 1]; that is, 1 is the highest curvature that can be achieved, which should represent a significant bend in the curve.

[0036] Concavity / convexity is an important characteristic of corners, playing a crucial role in many applications such as shape matching and 3D reconstruction. Mathematically, the concavity / convexity of a corner can be determined by the sign of its curvature value. That is, if the curvature is positive, the corner is convex; otherwise, it is concave. As a third issue, some existing models also lack this desirable property. For example, the discrete curvature measured by CCR is always greater than 0, thus losing information about the corner's concavity / convexity.

[0037] Furthermore, it is worth noting that if existing chord-to-point distance models utilize only a single scale (i.e., arc length L) to measure discrete curvature, their performance may become even more inconsistent with the perception of the human visual system (HVS). To illustrate this, the inventors tested... Figure 7A The diagram shows a curve. On this curve, a true corner "A" and a spurious (i.e., meaningless) corner "B" are marked. These two corners have the same angle value, but their straight arm lengths differ. For HVS, corner "A" is significant, while "B" is negligible. However, if a single-scale model CCR is used, these two points will have the same discrete curvature, as shown... Figure 7B As shown, they will all be detected as corners or missed – the former will result in a false positive, and the latter in a false negative.

[0038] In this embodiment, the step of performing edge detection on the image to be detected and obtaining an edge map specifically includes: performing edge detection on the image to be detected and obtaining an edge map based on the Canny detector.

[0039] In this embodiment, the calculation of any point Corresponding curvature product Specifically, it includes:

[0040] use Measurement points DCPD curvature value ,use Measurement points DCPD curvature value ,use Measurement points DCPD curvature value Based on DCPD curvature value , and Calculation points Corresponding curvature product ,in, , and All are natural numbers, 1≤ , and <i, function , , and Points , and x-coordinate , and Points , and The ordinate.

[0041] To address the aforementioned problems simultaneously, the detection method in this embodiment of the invention proposes a novel discrete curvature model. The basic idea of ​​this detection method is to develop a curvature metric that is highly consistent with true curvature and human visual system (HVS) perception. In the discrete case, the input curve is represented by a series of connected points. .set up This refers to the point where the discrete curvature needs to be measured. In this detection method, the chord is taken as the connection point. and The straight line segment, where r is a predetermined parameter defining the radius of the Region of Support (ROS, i.e., the arc) used to measure curvature. Then, the point... The discrete curvature at point directly passes through the point To the string Vertical distance (see attached) Figure 8 ) is used for measurement, and r is used for normalization.

[0042] The curvature is called the Direct Chord to Point Distance (DCPD) curvature value. Where r takes the following values: , and r, as a normalization factor, makes the measured curvature The range of values ​​is [ Between 1 and 1].

[0043] Compared with existing chord-to-point distance-based models, the proposed DCPD has the following multiple advantages:

[0044] (1) The first advantage of this detection method is that DCPD can exhibit behavior consistent with actual curvature, i.e., corners with smaller angle values ​​have higher curvature. To verify this, Figure 9 Demonstrates the use of DCPD in Figure 3A The discrete curvature measured on the test curve is shown. It can be seen that the corner point "C" with the smaller angle value has a higher discrete curvature than the other two corner points. Figure 10 Further demonstrated in Figure 4 Discrete curvature measured on the sharp corner point pattern shown and compare it with the angle value A comparison was made. It can be seen that, with... The increase, It decreases monotonically, exhibiting the same behavior as the actual curvature. In fact, and There is a close relationship between them, as described in the following theorem.

[0045] Theorem 1: For an angle , The sharp corner point, its DCPD (discrete curvature point distance) curvature The value measured at this corner point satisfies .

[0046] Proof 1: Let For point To the string The vertical distance. According to the definition of DCPD (Discrete Curvature Point Distance), we have For the sharp corner points considered, it is easy to see... (refer to Figure 4 ).therefore, .

[0047] More broadly, a rounded corner pattern was studied, such as Figure 11 As shown. At the corner of this pattern. The curve has a constant curvature around it, denoted as . The following theorem shows that, Discrete curvature and true curvature at a given point The relationship between them is roughly linear.

[0048] Theorem 2: For a point at a corner Surrounded by constant curvature The rounded corners, their DCPD (discrete distance to curvature) curvature satisfy Where r is the radius of the ROS (receiver operating region), which is equal to half the arc length. The DCPD curvature is measured at that corner point.

[0049] Proof 2: Reference Figure 11 Let O be the center of curvature. The radius of curvature is, i.e. .have and ,in It is a vector and The angle between them. Therefore, it can be deduced that .

[0050] Through the exist Perform a second-order Taylor series expansion at the point as Approximation (i.e.) ),get .

[0051] Through the derivation of Theorems 1 and 2 above, it is mathematically proven that the DCPD (Discrete Curvature Point Distance) in the embodiments of this invention can exhibit behavior consistent with true curvature. In particular, prominent corner points will have high DCPD curvature and can therefore be easily detected. This means that DCPD will produce fewer false negatives (i.e., missed detections) compared to using existing problematic chord-to-point distance-based models.

[0052] (2) A second advantage of this detection method is that it introduces fewer false positives (i.e., false detections) because, according to the two theorems mentioned above, the curvature of the flat curve is almost zero at every point. Therefore, the second problem of existing models (i.e., the potential for extremely large curvature values ​​on the flat curve) has also been solved. Figure 6A The test curve given shows a maximum discrete curvature of only 0.0468 measured using DCPD. Therefore, when this detection method is used for corner detection, it will not detect true corners from this curve.

[0053] (3) The third advantage of this detection method is that the concavity and convexity of each corner point can be easily identified based on the sign of the measured DCPD curvature. Finally, it is worth noting that DCPD is more efficient than existing models in terms of computational complexity. Assume that the input curve has N points. The computational burden of different models relative to the number of square root operations (the main overhead of calculating discrete curvature) is recorded in Table 1. CPDA (see paper: JHHan, T. Poston, Chord-to-point distance accumulation and planar curvature: a new approach to discrete curvature, PatternRecog. Lett. 520 22(10)(2001)1133–1144) and ACDM (see paper: T.-Y. Phillips, A. Rosenfeld, A method of curve partitioning using arcchord distance, PatternRecog. Lett. 5(4)(1987)285–288) require L-2 square root operations when processing each point of the curve, where L is the length of the chord used. Therefore, these two models require a total of The CCR (see paper: SWTeng, RMNSadat, G. Lu, Effective and efficient contour-based cornerdetectors, Pattern Recog. 48(7)(2015) 2185–2197) requires N-1 square root operations to calculate the arc length, and another N-1 square root operations to calculate the chord length. Therefore, the total number of square root operations is 2N-2. In contrast, the DCPD model has a lower computational burden—it only requires N square root operations, each used to calculate the discrete curvature at a point on the curve.

[0054] Table 1: Comparison of computation costs for different chord-point distance-based models, based on the number of square root operations.

[0055]

[0056] In this embodiment, the DCPD curvature value is used as a basis for... , and Calculation points Corresponding curvature product Specifically, it includes: .

[0057] here, Figure 12 This demonstrates how to use DCPD (Discrete Curvature Point Distance) to calculate the discrete curvature of a given point across three scales. First, a set of scale values ​​is selected, denoted as... At the current point Then, construct three chords accordingly, that is... , and Next, we will discuss the functions respectively. calculate DCPD curvature at the location , and Finally, the product of these three DCPD curvature values ​​is calculated and used as... Discrete curvature in a multi-scale sense, i.e. .

[0058] In this embodiment, the curvature product-based , ,..., Identify curves The corner points specifically include: through the product of curvature , ,..., Non-maximum suppression is performed to obtain several corner points with local maximum curvature values. Then, corner points with curvature values ​​greater than a predetermined threshold T are selected as curves. The corner points in the middle.

[0059] With curvature product set , ,..., Then, identify the corner points on the curve as follows. First, for , ,..., Non-maximum suppression is implemented—only points with local maxima of curvature are considered corner candidates. Then, a curvature threshold is used to find “meaningful” corners. That is, only corner candidates with curvature values ​​greater than a predetermined threshold T are considered true corners.

[0060] Figure 13A and Figure 13B This demonstrates how to adjust the parameters used in this detection method. Generally, a small curvature threshold T leads to many false positives (misidentifying non-corner points as corner points), while a large T may result in a large number of false negatives (missing real corner points). Therefore, it is necessary to determine an appropriate threshold. To this end, the inventors constructed five image datasets and used these datasets to adjust the value of T. Figure 13A This demonstrates the variation of the F-score (a metric for corner detection performance) under different T values. Accordingly, the default value for T is set to 0.002, at which point the F-score reaches its highest value. Similarly, as... Figure 13B As shown, scale set The default settings have been adjusted to {2, 4, 6}.

[0061] Because of the use of multi-scale curvature, the distinction between significant corner points and trivial curve structures becomes more pronounced, making them easier to detect, such as... Figure 14A , Figure 14B , Figure 14C , Figure 14D and Figure 14E As shown. First, the DCPD used measures at low, medium, and high scales. Figure 14A The discrete curvatures of the test curves shown are respectively as follows: Figure 14B , Figure 14C and Figure 14D As shown. It can be seen that in these single-scale cases, the DCPD curvature exhibits many redundant local maxima, making it difficult to identify the true corner points. Then, the product of these three curvatures is calculated, as shown... Figure 14E As shown, six meaningful corner points can be easily detected from them.

[0062] The identification curve Another advantage of the corner point method is that the convexity or concavity of the corner point can be identified using the sign of H(i). Specifically, if the sign is positive, the corner point is convex; if the sign is negative, it is concave. Figure 15A and Figure 15B An illustrative example is given. Figure 15A The curve in the image has seven convex angles and seven concave angles, and this is used to identify the curve. Discrete curvature measured by the corner point method, such as Figure 15B As shown. Based on the sign of the measured multi-scale curvature, Figure 15A The convexity and concavity of each convex corner are marked with dots, and concave corners are marked with squares.

[0063] experiment

[0064] In the literature, the lack of authoritative image datasets for conducting corner detection experiments with known evaluation metrics (precision, recall, and F-score) is a major issue, primarily because manually labeling real corners on images is extremely time-consuming. To address this critical problem, the inventors meticulously constructed five datasets containing a sufficient number and variety of real-world images from various image and vision tasks, including corner detection, shape matching, image segmentation, and edge detection. These five datasets are labeled Dataset-1 through Dataset-5, as shown in Table 2. For each test image in these datasets, the inventors carefully and objectively identified the actual corners, such as... Figure 16 and Figure 17 The points are shown in the table. The dataset is publicly available online. In addition, two other datasets without real corner reference solutions (labeled Dataset-A and Dataset-B in Table 2) were used to evaluate corner detection performance using metrics such as repeatability and localization error, which are widely used in existing work.

[0065] Table 2: Image dataset used in the inventor's experiments.

[0066]

[0067] Page 1 of 1:(1) F.Mokhtarian,R.Suomela,Robust image corner detectionthrough curvature scale space,IEEE Trans.Pattern Anal.Mach.Intell.20(12)(1998)1376–1381(2)G.Medioni,Y.Yasumoto,Corner detection and curverepresentation using cubic b-splines,Comput.Vis.,Graphics,Image Process.39(3)(1987)267–39. 278(3)A.Kalvin,E.Schonberg,JTSchwartz,M.Sharir,Two-dimensional,model-based,boundary matching using footprints,Int.J.of Robotics Res.5(4)(1986)38–55.

[0068] Appendix 2: D.Sharvit,J.Chan,H.Tech,BBChemistry,Symmetry-based indexingof image databases,J.Visual Commun.Image Represent.9(4)(1998)366–380.

[0069] Chapter 3: X. Sun, B. Zhong, Multi-scale corner detection based ondirect chordto-point distance (data, codes, and supplementary materials).

[0070] Figure 4 and 1:(1) F.Mokhtarian,R.Suomela,Robust image corner detectionthrough curvature scale space,IEEE Trans.Pattern Anal.Mach.Intell.20(12)(1998)1376–1381(2)G.-S.Xia,J.Delon,Y.Gousseau,Accurate junction detectionand characterization in natural images,Int.J.of Comput.Vis.106(1)(2014)31–56.

[0071] Applications 5:F.Kor,WF ̈orstner,eTRIMS image database for interpreting images of man-made scenes,http: / / www.ipb.uni-bonn.de / projects / etrims_db / 。

[0072] Chapter 6: LJLatecki,R.Lakamper,T.Eckhardt,Shape descriptors fornon-rigid shapes with a single closed contour,in: Proc.IEEEConf.Comput.Vis.Pattern Recog.,Vol.

[0073] Page 7 of 7: M. Awrangjeb, G. Lu, Robust image corner detection based on the chordto-point distance accumulation technique, IEEE Trans.Multimedia 10(6)(2008)1059–1072.

[0074] To test the robustness of corner detection under different conditions, five types of image quality attacks were performed on each image in the seven datasets mentioned above, with different parameter settings for each type, as detailed below. Therefore, a total of 87,344 images were tested, including the original selected images, 3,708 rotated images, 52,530 scaled images, 24,720 mixed-attack images, 4,120 JPEG compressed images, and 2,060 noisy images.

[0075] Rotation: Each image is rotated at 10° intervals within the range of [-90°, 90°] by different angles (excluding 0°, as it corresponds to the original image).

[0076] Scaling: Each image is scaled uniformly and non-uniformly in the range [0.5, 2] at intervals of 0.1. The scaling factor is... and Independent and separate selection (excluding) (The situation).

[0077] Hybrid attack: Apply a hybrid attack of rotation and scaling to each image, with rotation angles sampled at 10° intervals (excluding 0°) within the range of [-30°, 30°], and scaling factor... Sampling is performed independently and separately within the range [0.8, 1.2] at intervals of 0.1, and... .

[0078] Lossy compression: Each image is compressed using JPEG lossy compression, with the quality factor sampled at 5 intervals in the range [5, 100].

[0079] Noise: Gaussian white noise with a mean of zero is added to each image. The variance of the Gaussian noise is selected in increments of 0.005 within the range of [0.005, 0.05].

[0080] Evaluation Metrics. Six evaluation metrics were used in the objective evaluation, including the arithmetic mean of precision and recall (APR), F-score, average repetition rate (AR), and localization error (LE). The first two metrics require the use of true corner points. The average repetition rate measures the average ratio of repeated corner points between the original and degraded images. Localization error is defined as the average pixel deviation of repeated corner points, specifically as follows: , Where No and Nt represent the number of corners detected in the input image before and after a given image quality attack, respectively; Nm is the number of corners that match between them; and These are the positions of the i-th matching corner points before and after the attack. Here, "receptively" may be a misspelling of "respectively," meaning "separately."

[0081] Objective evaluation

[0082] It is well known that there is a trade-off between precision and recall in evaluation metrics. That is, for example, one method can easily achieve very high precision by sacrificing recall. Therefore, to make fair comparisons, the inventors tuned the parameters of each comparison corner detector to achieve the highest average F-score (based on the initially selected images) across Dataset-1 through Dataset-5, and then applied these parameters to all subsequent experiments.

[0083] True Corner

[0084] First, objective evaluations were performed on the images initially selected from Dataset-1 to Dataset-5, and the results are recorded in Table 3. Note that Dataset-5 represents a more challenging task compared to the other four datasets because, for this dataset, only corners meaningful to the human visual system are labeled as true corners, while unimportant corners (such as those on tree and cloud contour curves) are ignored. Therefore, on Dataset-5, each corner detector in the comparison tends to exhibit poor performance. The existing chord-to-point distance-based corner detector CPDA performs poorly due to utilizing only a single scale. Its multi-scale version, MCPDA, and another chord-to-point distance-based method, CCR, achieve competitive performance compared to other types of corner detectors. In contrast, our proposed detection method performs best on every evaluation metric across every dataset used. This clearly validates the potential of the proposed discrete curvature model and multi-scale corner detector.

[0085] Then, images subjected to different attacks were tested, and the results are reported in Table 4, where the performance of each corner detector is measured by the F-score. It can be observed that, out of 25 comparison cases (i.e., 5 datasets multiplied by 5 types of attacks), the method achieved the highest F-score in 17 cases. Finally, for each attack case, it provided the best performance in terms of average F-score across all used datasets.

[0086] Table 3: Comparison of different corner detectors in terms of APR (item 1) and F score (item 2) on the first 5 datasets.

[0087]

[0088] Table 4: Comparison of F-scores of different corner detectors under different image attacks, including rotation (item 1), scaling (item 2), hybrid attack (item 3), JPEG lossy compression (item 4), and Gaussian noise (item 5).

[0089]

[0090] The robustness of our detection method was tested on Dataset-A and Dataset-B and compared with existing methods. Since there is no reference solution for real corner points on these two datasets, the evaluation was based on average repeatability and localization error, which were calculated by comparing the sets of corner points detected before and after different attacks. Results are as follows... Figure 18 As shown, Figure 18 This section compares the repeatability (subplot a) and localization error (subplot b) of different corner detectors for Dataset-A and Dataset-B under different types of attacks. Note that higher average repeatability and lower localization error indicate better performance. It can be seen that, on average, this detection method achieves the best repeatability and competitive localization error performance.

[0091] Subjective evaluation

[0092] Five test images were selected from each of the first five datasets for subjective evaluation. The results are as follows: Figure 19 , Figure 20 and Figure 21 As shown. For each method used, the detected corner points can be compared with... Figure 16 and Figure 17 The performance of this detection method is evaluated by comparing it with real corner points shown in the image. It can be seen that the proposed method performs better than existing methods because it produces the fewest false alarms (i.e., false alerts) and false negatives (i.e., missed corner points). Figure 19 Subjective comparison of three binary images from Dataset-1, Dataset-2, and Dataset-3 using different corner detectors. Figure 20 Subjective comparison of different corner detectors on grayscale images from Dataset-4. Figure 21 Subjective comparison of color images from Dataset-5 using different corner detectors.

[0093] exist Figure 16 The image shows test images from three datasets (Binary), with the ground manually labeled. (a) dataset-1, 12 commonly used corner detection images; (b) dataset-2, 25 images from the Kimia25 dataset; (c) dataset-3, 36 images of Arabic numerals and Latin letters.

[0094] computational efficiency

[0095] To compare the computational efficiency of different corner detectors, these detectors were implemented on the first five datasets using the same machine (Intel CPU, 3.1 GHz, 16 GB RAM). Performance points between average CPU runtime and average F-score are shown below. Figure 22 As shown.

[0096] As can be seen, although this detection method performs best in terms of F-score, its CPU runtime is only slightly longer than SuperPoint and CCR. Note that SuperPoint requires a GPU for accelerated computation, while CCR is a single-scale corner detector. Therefore, it can be concluded that the multi-scale corner detector developed by the inventors has high computational efficiency. In fact, the detection method in this embodiment can achieve real-time performance—the processing time for each image on the dataset used is only about 80 milliseconds.

[0097] in conclusion

[0098] Based on a detailed study of existing chord-to-point distance-based models, a novel discrete curvature metric, the Direct String-to-Point Distance (DCPD), is proposed. DCPD directly measures the curvature of a digital curve using the maximum chord-to-point distance. Mathematical derivation demonstrates a close relationship between the detection method presented in this embodiment and the true curvature. Therefore, DCPD exhibits behavior consistent with the true curvature. Furthermore, compared to existing chord-to-point distance-based models, DCPD has the lowest computational complexity.

[0099] Building upon the proposed DCPD, the inventors further developed an efficient multi-scale corner detector. Given the lack of authoritative image datasets for corner detection experiments, another significant contribution of the inventors' work is the creation of an image dataset containing a sufficient number and variety of images collected from various image processing tasks. Specifically, the inventors labeled real corners on five image datasets for public access. To test the robustness of each comparative corner detector, the inventors' test dataset was further augmented with an additional 87,138 images by applying five different types of attacks (such as compression) and imposing varying degrees of each attack. The evaluation results obtained demonstrate that the proposed corner detector significantly outperforms several state-of-the-art detection methods.

[0100] Embodiment 2 of the present invention provides a corner detection device, including the following step modules: an initialization module, used to acquire the image to be detected (Image) and initialize the corner set. The processing module is used to perform edge detection on the image to be detected (Image) and obtain an edge map, and extract curves from the edge map. , ,..., Where M is a natural number; for any curve All were processed as follows: Fill curve Find the curve in the gap. Several T-junctions are found and all T-junctions are added to the corner set S; for the curve Smoothing is performed on the curve. Represented as point ,point ,...,point Calculate any point Corresponding curvature product Then, based on the curvature product , ,..., Identify curves The corner points in the set are identified and added to the corner set S, where... i and j are both natural numbers; 1 ≤ i ≤ j , 1≤j≤M; Corner module, used for corner sets The corner points in the image are all corner points in the image to be detected.

[0101] In this embodiment, the processing module is further configured to: perform edge detection on the image to be detected based on the Canny detector and obtain an edge map.

[0102] In this embodiment, the processing module is further configured to: use Measurement points DCPD curvature value ,use Measurement points DCPD curvature value ,use Measurement points DCPD curvature value Based on DCPD curvature value , and Calculation points Corresponding curvature product ,in, , and All are natural numbers, 1≤ , and <i, function , , and Points , and x-coordinate , and Points , and The ordinate.

[0103] In this embodiment, the processing module is further configured to: .

[0104] In this embodiment, the processing module is further configured to: multiply the curvature product , ,..., Non-maximum suppression is performed to obtain several corner points with local maximum curvature values. Then, corner points with curvature values ​​greater than a predetermined threshold T are selected as curves. The corner points in the middle.

[0105] It should be noted that although the steps are described in a specific order above, it does not mean that the steps must be executed in the above specific order. In fact, some of these steps can be executed concurrently or even in a different order, as long as the required function can be achieved.

[0106] This invention can be a system, method, and / or computer program product. A computer program product may include a readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the invention.

[0107] A readable storage medium can be a tangible device that holds and stores instructions for use by an instruction execution device. Readable storage media can include, for example, but not limited to, electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof.

[0108] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for detecting corner points, characterized in that, Includes the following steps: Obtain the image to be detected (Image) and initialize the angle set. ; Based on the Canny detector, edge detection is performed on the image to be detected to obtain an edge map, and curves are extracted from the edge map. , ,..., Where M is a natural number; for any curve All were processed as follows: Fill curve Find the curve in the gap. Several T-junctions are found and all T-junctions are added to the corner set S; for the curve Smoothing is performed on the curve. Represented as point ,point ,...,point ,use Measurement points DCPD curvature value ,use Measurement points DCPD curvature value ,use Measurement points DCPD curvature value , ,in, , and All are natural numbers, 1≤ , and <i, function , , and Points , and x-coordinate , and Points , and The ordinate, then, through the curvature product , ,..., Non-maximum suppression is performed to obtain several corner points with local maximum curvature values. Then, corner points with curvature values ​​greater than a predetermined threshold T are selected as curves. The corner points in the set are identified and added to the corner set S, where... i and j are both natural numbers; 1 ≤ i ≤ j , 1≤j≤M; Angle set The corner points in the image are all corner points in the image to be detected.

2. A corner detection device, characterized in that, Includes the following modules: The initialization module is used to obtain the image to be detected and initialize the angle set. ; The processing module is used to perform edge detection on the image to be detected (Image) based on the Canny detector and obtain an edge map, and extract curves from the edge map. , ,..., Where M is a natural number; for any curve All were processed as follows: Fill curve Find the curve in the gap. Several T-junctions are found and all T-junctions are added to the corner set S; for the curve Smoothing is performed on the curve. Represented as point ,point ,...,point ,use Measurement points DCPD curvature value ,use Measurement points DCPD curvature value ,use Measurement points DCPD curvature value , ,in, , and All are natural numbers, 1≤ , and <i, function , , and Points , and x-coordinate , and Points , and The ordinate, then, through the curvature product , ,..., Non-maximum suppression is performed to obtain several corner points with local maximum curvature values. Then, corner points with curvature values ​​greater than a predetermined threshold T are selected as curves. The corner points in the set are identified and added to the corner set S, where... i and j are both natural numbers; 1 ≤ i ≤ j , 1≤j≤M; Corner module, used for corner sets The corner points in the image are all corner points in the image to be detected.

Images