CN114549649A - Feature matching-based rapid identification method for scanned map point symbols - Google Patents
Feature matching-based rapid identification method for scanned map point symbols Download PDFInfo
- Publication number
- CN114549649A CN114549649A CN202210447549.XA CN202210447549A CN114549649A CN 114549649 A CN114549649 A CN 114549649A CN 202210447549 A CN202210447549 A CN 202210447549A CN 114549649 A CN114549649 A CN 114549649A
- Authority
- CN
- China
- Prior art keywords
- image
- map
- target
- symbol
- point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a rapid identification method of scanned map point symbols based on feature matching, which comprises the following steps: s1: inputting a target symbol and a target map image; s2: performing map local image block detection on the target map image in the step S1; s3: according to the size characteristics of the target symbol, the map local image block in the step S2 is screened to obtain a suspected point symbol area; s4: carrying out SIFT feature matching on the target symbol and each suspected point symbol area obtained in the step S3; s5: and setting a confidence threshold of the matching number of SIFT feature matching, and when the matching number is greater than the set confidence threshold, confirming that the region is a point symbol region and marking the position of the region in the original target image. The rapid identification method of the scanned map point symbols based on the feature matching can effectively and rapidly identify the point symbols of the scanned map image, reduces manual identification procedures and improves identification efficiency.
Description
Technical Field
The invention relates to the technical field of wireless positioning, in particular to a method for quickly identifying scanned map point symbols based on feature matching, belonging to a wireless positioning algorithm.
Background
Map symbols are important tools for conveying spatial information, and are a special graphic visual language. The map uses map symbols to express graphic elements, and reflects the positions, relationships, attributes and spatial distribution characteristics and evolution rules of various entities or phenomena in the objective world. The point symbol is an important component in a map symbol system, and can represent the meaning of an object by the shape or hue of a color of a symbol, the position of an object by the anchor point of a symbol and the importance level or number of an object by the size of a symbol. The correct understanding of the map is the key to fully exert the functions of the map, and particularly, the understanding of the map point symbols is important. Along with the development of science and technology, the intelligence degree of a computer is greatly enhanced, and the accuracy of image recognition is also greatly improved, so that the computer can possibly read a map, and the recognition of map point symbols by the computer from the perspective of human vision is one of the most important links.
In recent years, a great deal of research has been conducted by many scholars at home and abroad for the problem of map symbol recognition. The method mainly comprises the following steps: (1) template matching: template matching is the earliest method for recognizing map point symbols, but the matching result depends on the similarity calculation standard, the calculation amount is huge, and the recognition precision and efficiency of complex maps are low; (2) the statistical structure is as follows: and carrying out a large amount of statistical analysis on the symbol template to find out the structural characteristics of regularity. The method also achieves better recognition results because the map point symbols have fixed shapes and structures. However, map symbols are various in types, shapes among the symbols are different, the map symbols cannot be distinguished by using one or more simple structural features, more structures must be considered, and high-dimensional features often cause difficulty in classification and decision making; (3) the mathematical morphology is as follows: the method has the greatest characteristic that a large amount of complex image processing operation can be converted into the combined operation of basic displacement and logic operation, so that the algorithm design is more flexible and the efficiency is higher. However, the method is carried out in a binary image (generally a black-and-white image), and map symbol color information is ignored; (4) a neural network: in recent years, the neural network has a good advantage in image recognition, has strong learning ability on characteristics, and has good robustness. However, the power of neural networks relies on a rich set of data. For map images, it is difficult to form a symbol data set having a scale, which limits the application of neural networks to some extent. The above method mainly focuses on the description of global features of the whole map image, however, for some map images with large image resolution (greater than 5000 × 5000) and complicated map elements, efficient and accurate point symbol identification still remains a great challenge.
Chinese patent document (application number: 201510965338.5) discloses an image homonym point extraction method based on a local sorting direction histogram descriptor, which comprises the steps of calculating the direction derivatives of eight directions of each pixel point in a feature neighborhood, taking the direction positions of the largest derivative and the second largest derivative as feature elements, partitioning the neighborhood, and counting the distribution of the feature elements to form descriptor vectors. According to the technical scheme, the color information is ignored, the size of the target symbol is not considered, only matching is carried out, recognition is not carried out, and the matching efficiency is low.
Chinese patent document (application number: 201710006658.7) discloses a block fast matching algorithm based on a key point descriptor, namely an algorithm for achieving efficient image feature matching by feature point selection of block local information. Firstly, using an SIFT feature extraction method to obtain image feature points; then, carrying out rapid and accurate matching on SIFT feature points of the image; the computation complexity and the time loss of matching are reduced by carrying out local quick matching on non-matching points through a Neighbor-select algorithm based on angle hypothesis and rejecting mismatching points in a confirmed matching geometric neighborhood. The technical scheme ignores the color information, and the integrity of the elements can be possibly damaged by adopting a circular target segmentation method and a rectangular target segmentation method.
Chinese patent document (application number: 20171046969. X) discloses a rapid image SIFT feature matching method based on GPU and cascade hash, and establishes a GPU, memory and hard disk three-level exchange mechanism according to GPU global video memory size limitation and memory size limitation; meanwhile, an improved GPU parallel protocol method is adopted to perform Hash mapping with different coding lengths on all SIFT feature points of the image for two times; an upper triangular matrix block reading method is proposed and used; carrying out rough screening through a locality sensitive hash algorithm; fine screening is carried out on the candidate points by calculating the hamming distance; finally, the most suitable matching point is found out by calculating the Euclidean distance between the screening point and the point to be matched; and by using an asynchronous parallel method of the CPU and the GPU, the matching result data of the image pair is copied from a GPU video memory to a memory and stored to a disk while the GPU is operated. The technical scheme ignores the color information, does not consider the size of the target symbol, is the matching processing of the whole image and focuses on the hardware acceleration of the matching process.
Chinese patent document (application number: 201711365742.4) discloses a homologous local replication detection method based on superpixel multi-feature matching, which comprises the steps of firstly carrying out Gaussian smoothing filtering pretreatment on an image to be detected, using a BEMD algorithm to solve the image texture degree proportion, and adaptively initializing the number of superpixel blocks; secondly, acquiring image super-pixel blocks by using a SLIC (linear segmentation algorithm) -based super-pixel segmentation algorithm, and obtaining characteristic information of the super-pixel blocks by using a color lookup table color quantization technology and a texture moment analysis method; then, performing feature matching of super-pixel blocking by utilizing an Rg2NN algorithm and a BBF algorithm; and finally, performing post-processing by utilizing an SIFT feature point extraction algorithm, an RANSAC method, a ZNCC algorithm, a morphological method and the like. According to the technical scheme, color information is ignored, a blocking method of super-pixel segmentation is adopted, segmentation has certain randomness, and the number of segmented block areas is large.
Chinese patent document (application No. 201910719397.2) discloses a wood counting model based on object edge detection and feature matching, comprising the following steps: s1, gray processing; s2, image denoising processing; s3, image blurring processing; s4, edge enhancement processing; s5, HSV conversion processing; s6, edge detection processing; s7, binarization processing; s8, counting the end faces of the wood by using an SIFT algorithm, judging and processing environmental factors with large influence by adding edge detection and feature matching compared with the traditional graying-binarization processing mode, and obtaining a better wood counting result by pattern matching. The technical scheme is that the matching processing of the whole image is low in efficiency; information utilization for external environment (i.e. information elimination of irrelevant elements); the edge information of the whole image is obtained by simply utilizing edge detection without distinguishing degrees.
For the identification problem of the map point symbol, a more common method is feature matching, that is, the image similarity is calculated by constructing the feature expression of the whole image, and the matching result is judged according to the similarity value, so as to achieve the purpose of identifying the symbol. A general feature matching method is based on image representation (such as color, shape, texture, etc.) of global feature descriptors, which can be directly used as image description vectors and applied to the whole image. However, global features lack robustness, and for map images, when symbols are interfered by other map element covers or complex backgrounds, the features can be obviously changed, and recognition results are directly influenced. In addition, since the dot symbols occupy a small area in the map image, other elements (such as roads, water systems, and notes) inevitably participate in the image matching calculation when matching is performed, which greatly reduces the accuracy and efficiency of recognition. If symbol recognition is performed directly from the entire image, the non-dot symbol content occupying the image body can greatly interfere with the recognition result. From the perspective of information retrieval, the content of the non-dot symbol also affects the efficiency of retrieval.
Therefore, it is necessary to provide a method for rapidly identifying scanned map point symbols based on feature matching, which starts with local image blocks and divides the map into blocks by using a bounding box algorithm aiming at the configuration characteristics of the map point symbols, and provides an SIFT feature matching method considering point symbol color information, so as to solve the problem of rapidly identifying the scanned map point symbols; on one hand, the method provided by the invention overcomes the difficulty that global features are easily interfered; on the other hand, the defect of color loss of the traditional SIFT feature is made up by introducing color information.
Disclosure of Invention
The invention provides a rapid identification method of scanned map point symbols based on feature matching, which can effectively and rapidly identify the point symbols of a scanned map image, reduce manual identification procedures and improve identification efficiency.
In order to solve the technical problems, the invention adopts the technical scheme that: the method for quickly identifying the scanned map point symbol based on the feature matching specifically comprises the following steps:
s1: inputting a target symbol and a target map image;
s2: performing map local image block detection on the target map image in the step S1;
s3: according to the size characteristics of the target symbol, screening the local image block of the map in the step S2 to obtain a suspected point symbol area;
s4: performing SIFT feature matching on the target symbol and each suspected point symbol area obtained in the step S3;
s5: and setting a confidence threshold of the matching number of SIFT feature matching, and when the matching number is greater than the set confidence threshold, confirming that the region is a point symbol region and marking the position of the region in the original target image.
By adopting the technical scheme, the edge groups with similar structures are constructed by adopting a clustering idea in the edge extraction process, so that local image blocks are formed; and then, local blocking processing is carried out on the matched image, the method of edge structure clustering adopted during local blocking fully considers the connection structure of the image elements, and the blocking area is screened based on the symbol size characteristics, so that the matching efficiency is improved, simultaneously, the characteristic matching quality is enhanced, the color information constraint is increased by adopting SIFT characteristic matching, the characteristic description capacity is improved, the image matching is completed, and the matching quantity is used as the positioning basis of the map symbols. The rapid identification method of the scanned map point symbol based on the feature matching starts from a local image block and adopts a bounding box algorithm to block a map according to the configuration characteristics of the map point symbol, provides an SIFT feature matching method considering the color information of the point symbol, and is used for solving the problem of rapid identification of the scanned map point symbol; on one hand, the method provided by the invention overcomes the difficulty that global features are easily interfered; on the other hand, the defect of color loss of the traditional SIFT feature is made up by introducing color information.
As a preferred technical solution of the present invention, the step S2 specifically includes:
s21: carrying out gray level conversion on the target map image to obtain a gray level image;
s22: filtering and denoising the gray image in the step S21;
s23: acquiring an edge image of the target map image from the filtered gray level image by using a Canny edge detection algorithm;
s24: grouping the sparse edge images obtained in the step S23 based on a combination strategy to obtain a plurality of edge groups of a target map image; the goal is to group together edge points that approximate a line segment,
s25: and performing similarity calculation on all the edge groups obtained in the step S24, that is, calculating the similarity between every two edge groups, and if the similarity between two edge groups is greater than 0 and less than 1, aggregating the two edge groups, and generating a local image block of the map according to the aggregated result.
As a preferred technical solution of the present invention, the specific steps of the Canny edge detection algorithm in step S23 to obtain the edge image of the target map image are as follows:
s231: calculating the gradient and the direction of the gray image by adopting finite difference of first-order partial derivatives;
s232: performing non-maximum suppression on the gradient;
s233: and controlling edge connection through high and low thresholds to obtain an edge image of the target map image.
As a preferred technical solution of the present invention, the step S24 includes the following steps:
s241: selecting any edge point as a starting point, and traversing other edge points in eight neighborhoods of the starting point;
s242: calculating the sum of the direction angle differences between every two edge points until the sum is greater than 0.5 pi, and defining the edge point set into an edge group;
s243: and repeating the operations of the steps S241-S242 for the rest edge points until all the edge points are calculated, and obtaining a plurality of edge groups of the target map image.
As a preferred technical solution of the present invention, the calculation formula for calculating the similarity between every two edge groups in step S25 is as follows:
in the formula (I), the compound is shown in the specification,representing the similarity (also called affinity) of the two edge groups,andthe mean included angle is shown as the angle of the mean,mean azimuthal angle is indicated.
As a preferred technical solution of the present invention, in the step S22, a two-dimensional gaussian kernel is adopted to perform a convolution on the grayscale image in the step S21, and the weighted average is performed on the grayscale values of each pixel and the field pixels thereof, so as to complete filtering and denoising of the grayscale image of the target map image.
As a preferred technical solution of the present invention, the step S3 includes the following steps:
s31: calculating the length and width of the target symbol;
s32: calculating the length and width of each map local image block in the step S2;
s33: screening the local image blocks of the map in the step S2 to obtain suspected point symbol areas; the screening formula is as follows:
in the formula (I), the compound is shown in the specification,andrespectively representing the width and height of the template image,andrespectively representing the width and the height of the local image block; and when the local image block of the map meets the screening formula at the same time, judging that the local image block of the map is the suspected point symbol area.
As a preferred technical solution of the present invention, the step S4 includes the following steps:
s41: converting the target symbol and the target map image from RGB representation to HSV representation; the conversion formula is as follows:
in the formula, min and max represent the minimum value and the maximum value in rgb respectively, and h, s and v represent hue, saturation and brightness respectively;
s42: by setting a brightness thresholdObtaining the position of pixel point in the template of the target symbol, and calculating the H value (hue) range of the target symbol;
S43: calculating the target symbol and the SIFT feature points of each suspected point symbol area in the step S3 respectively;
s44: according to the H value range obtained in the step S42, SIFT feature points meeting the conditions;
s45: respectively calculating the euclidean distance between the SIFT feature vector in the template of the target symbol and the SIFT feature description vector of each suspected point symbol region in the step S3, and judging whether the two SIFT feature points are matched according to the ratio of the closest distance to the next closest distance.
As a preferred technical solution of the present invention, the step S45 includes the following steps:
s451: assume SIFT feature description vector in template of target symbol asThe SIFT feature description vector of the suspected point symbol region is;
S452: calculating the Euclidean distance between the target symbol and the SIFT feature description vector of each suspected point symbol region, wherein the calculation formula is as follows:
in the formula (I), the compound is shown in the specification,representing the Euclidean distance of two feature description vectors;a feature description vector representing a region of the suspected point symbol;a feature description vector representing a target symbol;
s453: judging whether the two feature points are matched or not according to the ratio relation of the nearest distance and the next nearest distance; the formula is as follows:
in the formula (I), the compound is shown in the specification,the closest distance is indicated and is,the next closest distance is represented as the distance between,represents a distance threshold; if it isThe ratio is less than or equal to the distance thresholdIf so, the matching is successful, otherwise, the matching is failed.
As a preferred embodiment of the present invention, the formula of the step S21 for performing the grayscale conversion on the target map image is as follows:
in the formula, Gray represents the Gray value of each pixel point in the converted target map image, and R, G, B represents the R, G, B value of the pixel point in the original target image.
Compared with the prior art, the method for rapidly identifying the scanned map point symbol based on the feature matching has the beneficial effects that:
(1) in the edge extraction process, an edge group with a similar structure is constructed by adopting a clustering idea, so that a local image block is formed; the edge structure clustering method adopted in the local block fully considers the connection structure of the image elements, and screens the block areas based on the symbol size characteristics, so that the matching efficiency is improved, and the characteristic matching quality is enhanced; the idea of image local blocking is adopted to solve the influence of irrelevant elements on the identification of the scanning map point symbols;
(2) the problem of rapid identification of the scanned map point symbols is solved by adopting a color-complemented SIFT feature matching method; color information constraint is added by adopting SIFT feature matching, and feature description capability is improved;
(3) the rapid identification method for scanning map point symbols based on feature matching breaks through the dependence of the traditional map image point symbol identification method on prior knowledge, the adopted feature description method has the characteristics of high identification degree, strong robustness and the like, especially for large-size map images, the adopted local blocking method greatly improves the efficiency and the precision of point symbol identification, and can provide effective support for map image information retrieval.
Drawings
FIG. 1 is a flow chart of a method for rapid identification of scanned map point symbols based on feature matching in accordance with the present invention;
FIG. 2 is a schematic diagram of an edge image obtained by the method for rapidly identifying a scanned map point symbol based on feature matching according to the present invention;
FIG. 3 is a schematic diagram of an extraction result of a suspected point symbol area obtained by the rapid identification method of scanned map point symbols based on feature matching according to the present invention;
fig. 4 is a schematic diagram of a fast recognition result of a scanned map point symbol obtained by the fast recognition method of a scanned map point symbol based on feature matching according to the present invention.
Detailed Description
The technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the drawings of the embodiments of the present invention.
Example (b): as shown in fig. 1, the method for rapidly identifying a scanned map point symbol based on feature matching specifically includes the following steps:
s1: inputting a target symbol and a target map image;
s2: performing map local image block detection on the target map image in the step S1;
the step S2 specifically includes:
s21: carrying out gray level conversion on the target map image to obtain a gray level image; the formula of the step S21 for performing the gray scale conversion on the target map image is as follows:
in the formula, Gray represents the Gray value of each pixel point in the converted target map image, and R, G, B represents the R, G, B value of the pixel point in the original target image;
s22: filtering and denoising the gray image in the step S21;
in the step S22, performing a convolution on the grayscale image in the step S21 by using a two-dimensional gaussian kernel, performing weighted average on the grayscale values of each pixel and the field pixels thereof, effectively filtering high-frequency noise in the image, and completing filtering and denoising of the grayscale image of the target map image;
s23: acquiring an edge image of the target map image from the filtered gray level image by using a Canny edge detection algorithm;
the specific steps of acquiring the edge image of the target map image by the Canny edge detection algorithm in the step S23 are as follows:
s231: calculating the gradient and the direction of the gray image by adopting finite difference of first-order partial derivatives;
s232: performing non-maximum suppression on the gradient;
s233: controlling edge connection through high and low thresholds to obtain an edge image of the target map image, as shown in fig. 2;
s24: grouping the sparse edge images obtained in the step S23 based on a combination strategy to obtain a plurality of edge groups of a target map image; the goal is to group together edge points that approximate a line segment,
the specific steps of step S24 are:
s241: selecting any edge point as a starting point, and traversing other edge points in eight neighborhoods of the starting point;
s242: calculating the sum of the direction angle differences between every two edge points until the sum is greater than 0.5 pi, and defining the edge point set into an edge group;
s243: repeating the operations of the steps S241-S242 for the rest edge points until all the edge points are calculated, and obtaining a plurality of edge groups of the target map image;
s25: performing similarity calculation on all edge groups obtained in the step S24, that is, calculating the similarity between every two edge groups, if the similarity value between two edge groups is greater than 0 and less than 1, aggregating the two edge groups, and generating a local image block of the map according to the aggregated result;
the calculation formula for calculating the similarity between each two edge groups in step S25 is as follows:
in the formula (I), the compound is shown in the specification,representing the similarity (also called affinity) of the two edge groups,andthe mean included angle is shown as the angle of the mean,representing the average azimuth angle;
s3: according to the size characteristics of the target symbol, the map local image blocks in the step S2 are screened, and by controlling the length and width of the local image blocks, image blocks with excessively large or small size and large difference in length and width are removed, so as to obtain a suspected point symbol area, as shown in fig. 3;
different from other elements on the map, the size of the point symbol has certain similarity with the length-width ratio of the circumscribed rectangle; the specific steps of step S3 are:
s31: calculating the length and width of the target symbol;
s32: calculating the length and width of each map local image block in the step S2;
s33: screening the local image blocks of the map in the step S2 to obtain suspected point symbol areas; the screening formula is as follows:
in the formula (I), the compound is shown in the specification,andrespectively representing the width and height of the template image,andrespectively representing the width and the height of the local image block; when the local image block of the map meets the screening formula at the same time, judging that the local image block of the map is a suspected point symbol area;
s4: performing SIFT feature matching on the target symbol and each suspected point symbol area obtained in the step S3;
the specific steps of step S4 are:
s41: converting the target symbol and the target map image from RGB representation to HSV representation; the conversion formula is as follows:
wherein min and max represent the minimum and maximum values in rgb, respectively; an HSV (Hue, Saturation, Value) color space, also called a hexagonal pyramid model, is a color model facing visual perception, and describes colors from Hue (H), Saturation (S), and brightness (V) according to their intuitive characteristics; according to the invention, color information is added into the SIFT matching process according to the design concept of map point symbols, so that the defect of SIFT color information loss is overcome;
s42: by setting a brightness thresholdObtaining the pixel point position in the template of the target symbol, and calculating the H value range of the target symbol;
S43: calculating the target symbol and the SIFT feature points of each suspected point symbol area in the step S3 respectively;
s44: according to the H value range obtained in the step S42, SIFT feature points meeting the conditions;
s45: respectively calculating Euclidean distances between SIFT feature vectors in the template of the target symbol and SIFT feature description vectors of each suspected point symbol region in the step S3, and judging whether the two SIFT feature points are matched according to a ratio relation between the closest distance and the next closest distance;
the specific steps of step S45 are:
s451: assume SIFT feature description vector in template of target symbol asThe SIFT feature description vector of the suspected point symbol region is;
S452: calculating the Euclidean distance between the target symbol and the SIFT feature description vector of each suspected point symbol region, wherein the calculation formula is as follows:
in the formula (I), the compound is shown in the specification,representing the Euclidean distance of two feature description vectors;a feature description vector representing a region of the suspected point symbol;a feature description vector representing a target symbol;
s453: judging whether the two feature points are matched or not according to the ratio relation of the nearest distance to the next nearest distance; the formula is as follows:
in the formula (I), the compound is shown in the specification,the closest distance is indicated and is,the next closest distance is represented as the distance between,represents a distance threshold; if it isThe ratio is less than or equal to the distance thresholdIf so, matching is successful, otherwise, matching is failed; common threshold fetching(ii) a Finally, if the matching number of the target symbol template and the suspected point symbol area is more, the area is judged to have higher confidence coefficient of the point symbol area;
s5: setting a confidence threshold of the matching number of SIFT feature matching, and when the matching number is greater than the set confidence threshold, determining that the region is a point symbol region and marking the position of the region in the original target image; as shown in fig. 4.
The above description is only exemplary of the present invention and should not be taken as limiting the invention, as any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A rapid identification method of scanned map point symbols based on feature matching is characterized by comprising the following steps:
s1: inputting a target symbol and a target map image;
s2: performing map local image block detection on the target map image in the step S1;
s3: according to the size characteristics of the target symbol, screening the local image block of the map in the step S2 to obtain a suspected point symbol area;
s4: carrying out SIFT feature matching on the target symbol and each suspected point symbol area obtained in the step S3;
s5: and setting a confidence threshold of the matching number of SIFT feature matching, and when the matching number is greater than the set confidence threshold, confirming that the region is a point symbol region and marking the position of the region in the original target image.
2. The method for rapidly identifying a scanned map point symbol based on feature matching according to claim 1, wherein the step S2 specifically comprises:
s21: carrying out gray level conversion on the target map image to obtain a gray level image;
s22: filtering and denoising the gray image in the step S21;
s23: acquiring an edge image of the target map image from the filtered gray level image by using a Canny edge detection algorithm;
s24: grouping the edge images obtained in the step S23 based on a combination strategy to obtain a plurality of edge groups of the target map image;
s25: and performing similarity calculation on all the edge groups obtained in the step S24, that is, calculating the similarity between every two edge groups, and if the similarity between two edge groups is greater than 0 and less than 1, aggregating the two edge groups, and generating a local image block of the map according to the aggregated result.
3. The method for rapidly identifying the scanned map point symbol based on the feature matching as claimed in claim 2, wherein the specific steps of the Canny edge detection algorithm in the step S23 to obtain the edge image of the target map image are as follows:
s231: calculating the gradient and the direction of the gray image by adopting finite difference of first-order partial derivatives;
s232: performing non-maximum suppression on the gradient;
s233: and controlling edge connection through high and low thresholds to obtain an edge image of the target map image.
4. The method for rapidly identifying the scanned map point symbol based on the feature matching as claimed in claim 2, wherein the specific steps of the step S24 are as follows:
s241: selecting any edge point as a starting point, and traversing other edge points in eight neighborhoods of the starting point;
s242: calculating the sum of the direction angle differences between every two edge points until the sum is greater than 0.5 pi, and defining the edge point set into an edge group;
s243: and repeating the operations of the steps S241-S242 for the rest edge points until all the edge points are calculated, and obtaining a plurality of edge groups of the target map image.
5. The method for rapidly identifying the scanned map point symbol based on the feature matching as claimed in claim 2, wherein the calculation formula for calculating the similarity between every two edge groups in the step S25 is as follows:
6. The method as claimed in claim 2, wherein in step S22, the gray scale image in step S21 is convolved once by using a two-dimensional gaussian kernel, and the gray scale values of each pixel and its field pixels are weighted-averaged to complete filtering and de-noising of the gray scale image of the target map image.
7. The method for rapidly identifying the scanned map point symbol based on the feature matching as claimed in claim 2, wherein the specific steps of the step S3 are as follows:
s31: calculating the length and width of the target symbol;
s32: calculating the length and width of each map local image block in the step S2;
s33: screening the local image blocks of the map in the step S2 to obtain suspected point symbol areas; the screening formula is as follows:
in the formula (I), the compound is shown in the specification,andrespectively representing the width and height of the template image,andrespectively representing the width and the height of the local image block; and when the local image block of the map meets the screening formula at the same time, judging that the local image block of the map is the suspected point symbol area.
8. The method for rapidly identifying the scanned map point symbol based on the feature matching as claimed in claim 7, wherein the specific steps of the step S4 are as follows:
s41: converting the target symbol and the target map image from RGB representation to HSV representation; the conversion formula is as follows:
in the formula, min and max represent the minimum value and the maximum value in rgb respectively, and h, s and v represent hue, saturation and brightness respectively;
s42: by setting a brightness thresholdObtaining the pixel point position in the template of the target symbol, and calculating the H value range of the target symbol;
S43: calculating the target symbol and the SIFT feature points of each suspected point symbol area in the step S3 respectively;
s44: according to the H value range obtained in the step S42, SIFT feature points meeting the conditions;
s45: respectively calculating the euclidean distance between the SIFT feature vector in the template of the target symbol and the SIFT feature description vector of each suspected point symbol region in the step S3, and judging whether the two SIFT feature points are matched according to the ratio of the closest distance to the next closest distance.
9. The method for rapidly identifying the scanned map point symbol based on the feature matching as claimed in claim 8, wherein the specific steps of the step S45 are as follows:
s451: assume SIFT feature description vector in template of target symbol asThe SIFT feature description vector of the suspected point symbol region is;
S452: calculating the Euclidean distance between the SIFT feature vector in the template of the target symbol and the SIFT feature description vector of each suspected point symbol region, wherein the calculation formula is as follows:
in the formula (I), the compound is shown in the specification,representing the Euclidean distance of two feature description vectors;a feature description vector representing a region of the suspected point symbol;a feature description vector representing a target symbol;
s453: judging whether the two feature points are matched or not according to the ratio relation of the nearest distance and the next nearest distance; the formula is as follows:
in the formula (I), the compound is shown in the specification,the closest distance is indicated and is,the next closest distance is represented as the distance between,represents a distance threshold; if it isThe ratio is less than or equal to the distance thresholdIf so, the matching is successful, otherwise, the matching is failed.
10. The method for rapidly identifying a symbol of a scanned map point based on feature matching as claimed in claim 6, wherein the formula of the step S21 for performing gray scale conversion on the target map image is as follows:
in the formula, Gray represents the Gray value of each pixel point in the converted target map image, and R, G, B represents the R, G, B value of the pixel point in the original target image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210447549.XA CN114549649A (en) | 2022-04-27 | 2022-04-27 | Feature matching-based rapid identification method for scanned map point symbols |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210447549.XA CN114549649A (en) | 2022-04-27 | 2022-04-27 | Feature matching-based rapid identification method for scanned map point symbols |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114549649A true CN114549649A (en) | 2022-05-27 |
Family
ID=81667581
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210447549.XA Pending CN114549649A (en) | 2022-04-27 | 2022-04-27 | Feature matching-based rapid identification method for scanned map point symbols |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114549649A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117079166A (en) * | 2023-10-12 | 2023-11-17 | 江苏智绘空天技术研究院有限公司 | Edge extraction method based on high spatial resolution remote sensing image |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103065147A (en) * | 2012-12-25 | 2013-04-24 | 天泽信息产业股份有限公司 | Vehicle monitoring method based on image matching and recognition technology |
CN104809463A (en) * | 2015-05-13 | 2015-07-29 | 大连理工大学 | High-precision fire flame detection method based on dense-scale invariant feature transform dictionary learning |
CN105825203A (en) * | 2016-03-30 | 2016-08-03 | 大连理工大学 | Ground arrowhead sign detection and identification method based on dotted pair matching and geometric structure matching |
-
2022
- 2022-04-27 CN CN202210447549.XA patent/CN114549649A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103065147A (en) * | 2012-12-25 | 2013-04-24 | 天泽信息产业股份有限公司 | Vehicle monitoring method based on image matching and recognition technology |
CN104809463A (en) * | 2015-05-13 | 2015-07-29 | 大连理工大学 | High-precision fire flame detection method based on dense-scale invariant feature transform dictionary learning |
CN105825203A (en) * | 2016-03-30 | 2016-08-03 | 大连理工大学 | Ground arrowhead sign detection and identification method based on dotted pair matching and geometric structure matching |
Non-Patent Citations (2)
Title |
---|
C.LAWRENCE ZITNICK等: "Edge Boxes: Locating Object Proposals from Edges", 《ECCV》 * |
杨涛等: "一种基于HSV颜色空间和SIFT特征的车牌提取算法", 《计算机应用研究》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117079166A (en) * | 2023-10-12 | 2023-11-17 | 江苏智绘空天技术研究院有限公司 | Edge extraction method based on high spatial resolution remote sensing image |
CN117079166B (en) * | 2023-10-12 | 2024-02-02 | 江苏智绘空天技术研究院有限公司 | Edge extraction method based on high spatial resolution remote sensing image |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109086714B (en) | Form recognition method, recognition system and computer device | |
CN107609549B (en) | Text detection method for certificate image in natural scene | |
JP5008572B2 (en) | Image processing method, image processing apparatus, and computer-readable medium | |
CN101398894B (en) | Automobile license plate automatic recognition method and implementing device thereof | |
CN108121991B (en) | Deep learning ship target detection method based on edge candidate region extraction | |
CN105046252B (en) | A kind of RMB prefix code recognition methods | |
CN107038416B (en) | Pedestrian detection method based on binary image improved HOG characteristics | |
CN110334762B (en) | Feature matching method based on quad tree combined with ORB and SIFT | |
JP3353968B2 (en) | Image processing device | |
CN108491786B (en) | Face detection method based on hierarchical network and cluster merging | |
CN110309808B (en) | Self-adaptive smoke root node detection method in large-scale space | |
CN104778701A (en) | Local image describing method based on RGB-D sensor | |
CN110647795A (en) | Form recognition method | |
CN110738106A (en) | optical remote sensing image ship detection method based on FPGA | |
WO2021253633A1 (en) | Recognition method and terminal for batch of qr codes | |
WO2022121025A1 (en) | Certificate category increase and decrease detection method and apparatus, readable storage medium, and terminal | |
CN111783773A (en) | Correction method for angle-oriented inclined wire pole signboard | |
CN110969164A (en) | Low-illumination imaging license plate recognition method and device based on deep learning end-to-end | |
JP2011248702A (en) | Image processing device, image processing method, image processing program, and program storage medium | |
Mei et al. | A novel framework for container code-character recognition based on deep learning and template matching | |
CN111915628A (en) | Single-stage instance segmentation method based on prediction target dense boundary points | |
CN111027564A (en) | Low-illumination imaging license plate recognition method and device based on deep learning integration | |
CN114549649A (en) | Feature matching-based rapid identification method for scanned map point symbols | |
CN114387592A (en) | Character positioning and identifying method under complex background | |
CN104268845A (en) | Self-adaptive double local reinforcement method of extreme-value temperature difference short wave infrared image |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220527 |
|
RJ01 | Rejection of invention patent application after publication |