WO2012111906A1 - 이미지 특징 데이터 생성 장치 및 방법 - Google Patents
이미지 특징 데이터 생성 장치 및 방법 Download PDFInfo
- Publication number
- WO2012111906A1 WO2012111906A1 PCT/KR2011/008600 KR2011008600W WO2012111906A1 WO 2012111906 A1 WO2012111906 A1 WO 2012111906A1 KR 2011008600 W KR2011008600 W KR 2011008600W WO 2012111906 A1 WO2012111906 A1 WO 2012111906A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- feature
- feature point
- image
- points
- point
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
Definitions
- the present invention relates to an apparatus and method for generating image feature data, and more particularly, to an apparatus and method capable of properly determining a feature point representing a feature of an image and efficiently generating feature data describing the feature point.
- a feature point is a point that can represent a feature of an image, and is a point or point that well describes a feature of the image regardless of changes in the scale, rotation, or distortion of the image.
- These feature points vary depending on the size and content of a given image and also the type of feature point extraction / determination method. For example, thousands to tens of thousands of feature points may be extracted per picture.
- These feature points are widely used in the field of image processing or computer vision. For example, the feature points are extracted and the feature data of the extracted feature points are used to find the corresponding parts in the two images. It is used in various ways.
- the conventional feature point extraction / decision method there are many limitations in that a large number of feature points are acquired in a given image, so that the amount of data to be processed is excessive in the subsequent post-processing process, and thus a large amount of computation time is required. Doing.
- the feature data formed for each extracted / determined feature point also has the disadvantage that the amount and processing time are excessively required.
- the present invention has been made in view of the above, and an object thereof is to provide a method and apparatus capable of effectively determining a feature point from an image.
- the present invention provides a method and an apparatus capable of generating feature data of a feature point that can explain the feature of each feature point of the image well and can significantly reduce the amount of data, processing and computation time. Another purpose.
- an apparatus for generating feature data of an image comprising: a feature point determiner for determining a feature point from an image and extracting feature point information of the determined feature point; A feature point filter to determine at least one or more feature points as final feature points among the feature points determined by the feature point determiner; And a feature data generator configured to generate feature data of the image based on the final feature point determined by the feature point filter and the feature point information of the final feature point.
- the feature point information extracted from the feature point determiner includes the strength of the feature point
- the feature point filter part has a point having a greater intensity than the points located in the peripheral region of the feature point based on the strength of the feature point as the final feature point. Can be configured to make decisions.
- the feature point filtering unit among the points located in the area around the feature point, Where c i is the i th feature point, f (c i ) is the intensity of the i th feature point, R 1 (c i ) is the set of points in a predetermined peripheral region of the feature point, Determines R 2 (C i ) , which are important points that satisfy the formula of the maximum value of intensity, T 1 is the threshold value among R 1 (c i ) , (Where # is an operator for obtaining the size of the set, T 2 is a threshold) may be configured to determine as the final feature point.
- a method for generating feature data of an image comprising: determining a feature point from an image and extracting feature point information of the determined feature point; Determining at least one feature point as the final feature point among the determined feature points; And generating feature data of the image based on the determined final feature point and feature point information of the final feature point.
- an apparatus for generating feature data of an image comprising: a feature point determiner configured to determine a feature point from an image and extract feature point information of the determined feature point; A feature point direction estimator for estimating direction information on each of the feature points determined by the feature point determiner; And a feature data generation unit for generating a binary feature vector for each feature point determined by the feature point determiner based on the feature point information and the direction information, and generating feature data of an image including the binary feature vector.
- a data generating device can be provided.
- the feature point direction estimating unit may be configured to estimate the direction of the feature point by calculating a gradient around each point for all points of a certain area around the feature point and obtaining an average value of the direction.
- the feature data generator may generate a binary feature vector for each feature point determined by the feature point determiner based on the feature point information and the direction information, and generate feature data of an image including the generated binary feature vector. It can also be configured to.
- the feature data generator generates a peripheral image area including the feature point for each feature point, aligns the generated areas in the same direction, and then divides the aligned peripheral image areas into subregions, respectively. And generate a binary feature vector based on an average value of brightness values of the subregions.
- the binary feature vector may be generated by at least one selected from a difference vector and a difference vector of an average value of brightness values of the sub-regions.
- selecting at least one of the difference vector and the difference vector of the average value of the brightness values of the sub-regions may be configured to be selected corresponding to each bit of the binary feature vector.
- it may be configured to determine the value of the corresponding bit of the binary feature vector by calculating a linear combination or a nonlinear combination for the difference vector and the difference vectors selected corresponding to each bit and comparing the result with a threshold.
- alignment may be performed according to a preset criterion for each bit value of the binary feature vector.
- the feature data of the image may be configured to further include at least one or more of the location information, size information, direction information of the feature point.
- the feature point determiner may further include a feature filter unit that finally determines at least one or more feature points among the determined feature points as feature points.
- the feature point information extracted by the feature point determiner includes the strength of the feature point, and the feature point filter part has a point having a greater intensity than the points located in the peripheral region of the feature point based on the strength of the feature point. It can also be configured to make a decision.
- the feature point filtering unit among the points located in the area around the feature point, Where c i is the i th feature point, f (c i ) is the intensity of the i th feature point, R 1 (c i ) is the set of points in a predetermined peripheral region of the feature point, Determines R 2 (C i ) , which are important points that satisfy the formula of the maximum value of intensity, T 1 is the threshold value among R 1 (c i ) , (Where # is an operator for obtaining the size of the set, and T 2 is a threshold) can be configured to finally determine as a feature point.
- a method for generating feature data of an image comprising: determining a feature point from an image and extracting feature point information of the determined feature point; Estimating direction information for each of the determined feature points; And generating a binary feature vector for each of the determined feature points based on the feature point information and the direction information, and generating feature data of an image including a binary feature vector.
- the present invention it is possible to provide a method and apparatus capable of effectively determining a feature point from an image.
- the present invention can provide a method and apparatus for properly selecting the feature points that can well represent the features of the image when there are too many feature points of the image.
- the present invention has the effect of generating feature data of a feature point that can explain the feature well for each feature point of the image and can significantly reduce the amount of data, processing and computation time.
- the present invention has the effect of generating the characteristic data of the image that is robust to various changes such as scaling, rotation, observation angle, etc. while reducing data generation and processing time.
- the processing and calculation time are greatly reduced as compared to the conventional technology, so that the feature data of the image can be quickly and efficiently Relevant work can be performed, and feature data of an image having robust characteristics can be generated and used regardless of various changes such as scaling, rotation, and viewing angle of the image, thereby greatly improving the overall working time and efficiency.
- FIG. 1 is a block diagram showing the configuration of an image feature data generating apparatus 10 according to an embodiment of the present invention.
- FIG. 2 is a flowchart illustrating an embodiment of a method for generating feature data of an image implemented by the image feature data generating apparatus 10 of FIG. 1.
- FIG. 3 is a diagram showing the configuration of an image feature data generating apparatus 20 according to another embodiment of the present invention.
- FIG. 4 is a screen illustrating an actual example of a process of generating a binary feature vector.
- FIG. 5 is a flowchart illustrating a process of generating a binary feature vector.
- FIG. 6 is a flowchart illustrating an example of a detailed process of generating a binary feature vector based on a difference vector and a difference vector.
- FIG. 7 is a diagram showing the configuration of an image characteristic data generation device 30 according to another embodiment of the present invention.
- FIG. 1 is a block diagram showing the configuration of an image feature data generating apparatus 10 according to an embodiment of the present invention.
- the image feature data generating apparatus 10 includes a feature point determiner 11, a feature filter 12, and a feature data generator 13. Determine and generate feature data from the determined feature points.
- the feature point determiner 11 determines a feature point from an image and extracts feature point information of the determined feature point.
- the image refers to still image data, for example, digital data represented by a file format such as jpg, bmp, tif, or the like.
- an image point (interest point, feature point) refers to points that can express the features of the image better than other points in the image, generally scaling (scaling) ), It is common to determine the feature points that can always be detected equally on the image, regardless of changes in rotation, observation angle, etc.
- the feature point determiner 11 may use the feature point extraction / decision method known in the art as it is. As described above, the maximum / minimum value in the scale space of a Laplacian of Gaussian (LoG) filter or a Difference of Gaussians (DoG) filter is described. By using the method, using the Hessian matrix determinant (determinant), etc. can be used to determine the points that can be a feature point in a given image.
- LiG Laplacian of Gaussian
- DoG Difference of Gaussians
- Scale-Invariant Feature Transform as disclosed in US Patent No. 6,711,293 (David G. Lowe) or the Speed Up Robust Features (SURF) algorithm disclosed in US Patent Publication No. US 2009/0238460 may be used.
- SIFT Scale-Invariant Feature Transform
- SURF Speed Up Robust Features
- the feature point determination unit 11 in the present invention can use any feature point extraction / determination method known in the prior art as such, which is not essential to the present invention, and thus detailed description thereof will be omitted.
- the feature point determiner 11 extracts the feature points of the image and also extracts other feature point information such as the strength of the feature point and the size of the feature point. Since the type and specific details of the feature point information may vary depending on the feature point extraction / determination method used, the feature point information is selectively extracted according to data used in post-processing such as image matching, object tracking, image comparison, and the like.
- the strength of the feature point may vary depending on the feature point extraction / determination method used. For example, when using a Laglacian of Gaussian filter, a Laplacian operator can be used as the strength of the feature point. Given image f (x, y) is a Gaussian kernel for a given scale t When convolved by, the LoG scale-space is Where the Laplacian operator The Laplace operator results in a large value at the dark and light points of the image, so you can basically determine whether the image can be used as a feature point. Can be used as an indicator of strength as a feature point, depending on the magnitude of its value.
- the result of the Laplace operator can be used as the feature point strength.
- the determinant value of the Hessian matrix may be used as the feature point strength. In this way, the intensity of the feature point can use information based on a discriminant used to extract / determine the feature point of the image by the prior art.
- the size of the feature point of the image represents the information of the area occupied by the feature point in the image, for example, in the case of a rectangle may be represented as the length of each side, the length of the radius in the case.
- the size of these feature points may also be used in the prior art.
- the scale t or the maximum intensity of the feature point
- Value can be used, such as k times (where k is any constant, such as 4,6, etc.).
- the feature point filtering unit 12 performs a function of determining at least one or more feature points as the final feature points among the feature points determined by the feature point determiner 11 as described above. Since the feature points determined by the feature point determiner 11 may typically reach from tens to many thousands and tens of thousands per image, the feature points having more distinctive features may be selected from among the feature points for large-capacity and high-speed processing. It is necessary, the feature point filtering unit 12 performs a function of selecting the feature points having a more distinct and clear features than the other feature points determined from the feature point determination unit 11 and select them as the final feature point.
- the feature point filter 12 may select the feature points by the following method. For example, for an image having a size W ⁇ H, the intensity of the i-th point for the feature points c 1 , c 2 ,..., C N determined by the feature point determiner 11 is denoted by f (c i ) , If a set of points belonging to the periphery of each point, for example, min (W, H) / 10 radius is R 1 (c i ) , then whether or not the point c i can be finally selected as a feature point is first Calculate the relatively important points R 2 (C i ) around The feature points satisfying the following equations defined by and using them can be determined as the final feature points.
- T 1 and T 2 are optimizable thresholds.
- R 2 (C i ) which are important points around arbitrary feature points, are found and their strengths are used, the feature has a relatively large intensity compared to the strength of surrounding important points.
- the feature point can be finally selected as the final feature point. That is, even for images with large local changes or images with complex textures, it is possible to select feature points that represent the entire area well while stably having a small number, and determine the number of feature points determined by the feature point determiner 11 in several tens. Can be reduced to hundreds.
- the strength may be a value of a formula used to determine whether a feature point is used in an algorithm used in the feature point determiner 11, such as a Laplacian operator, as described above in the feature point determiner 11.
- the feature data generator 13 generates a feature data of the image based on the final feature point determined by the feature filter 12 and feature point information of the final feature point.
- Feature data generally refers to data that describes information related to feature points extracted / determined for a given image.
- Such feature data may be used in post-processing processes such as object tracking, image matching, image identity determination, and the like. It is useful.
- Such feature data can be generated in various forms according to the type, condition or algorithm of the post-processing process.
- the feature data generation unit 13 in the embodiment of FIG. 1 can use the feature data generation method known by the prior art as it is.
- the feature data may include coordinates of a feature point, orientation information, a feature vector, and the like.
- the feature data generation unit 13 of the embodiment of FIG. 1 may use the feature data generation method used in the following embodiments, which will be described in detail with reference to FIG. 3 and below.
- FIG. 2 is a flowchart illustrating an embodiment of a method for generating feature data of an image implemented by the image feature data generating apparatus 10 of FIG. 1.
- the feature point determiner 11 determines a feature point from a given image and extracts feature point information of the determined feature point as described above (S100).
- the feature point filtering unit 12 determines at least one or more feature points from the feature points determined by the feature point determiner 11 as final feature points (S110).
- the feature point determining unit 11 uses the equation used to determine the feature point to define the strength of the feature point and based on the strength of the feature point to determine the final feature point with a feature point that is higher in strength than the surrounding points. It can be made by.
- the feature data generator 13 When the final feature point is determined, the feature data generator 13 generates the feature data of the image based on the feature information of the final feature point and the final feature point determined by the feature filter 12 (S120).
- FIG. 3 is a diagram showing the configuration of an image feature data generating apparatus 20 according to another embodiment of the present invention.
- the image feature data generation device 20 of the present embodiment includes a feature point determiner 21, a feature point direction estimator 22, and a feature data generator 23.
- the feature point determiner 21 has the same structure and operation as the feature point determiner 11 described in the embodiment of FIGS. 1 to 2, detailed description thereof will be omitted.
- the feature point filtration unit 12 is omitted in comparison with the embodiment of FIGS. 1 and 2.
- 1 to 2 are characterized in that the feature points are selected by an appropriate number, this embodiment generates feature data for the feature points determined once the feature points are determined, regardless of the process of determining and selecting the feature points. Since the present invention relates to a method of doing so, the feature filter 12 of FIGS. 1 to 2 is not necessarily used. Of course, it may be configured to include the feature point filtering unit 12, and the embodiment of the case in which the feature point filtering unit 12 is included will be described later separately.
- the feature point direction estimator 22 performs a function of estimating direction information on each of the feature points determined by the feature point determiner 22. This may use a variety of gradient based methods known in the art. For example, the direction information of the feature point can be estimated by calculating a gradient around each point and averaging the directions of all points of a certain area around the feature point. According to this method, the original direction can be estimated even when the feature point undergoes any rotational transformation. The method for estimating the direction of the feature point in the feature point direction estimator 22 may also use a known method known in the prior art, and thus, detailed description thereof will be omitted.
- the feature data generator 23 generates a binary feature vector for each of the feature points determined by the feature point determiner 21 based on the feature point information and the direction information, and generates the feature data of the image including the generated binary feature vector. It performs the function.
- the feature point information is extracted and generated by the feature point determiner 21 and the direction information is generated by the feature point direction estimator 22 described above.
- the feature data generator 23 can be processed quickly in a post-processing process such as image matching, object tracking, image comparison, etc. using the generated feature data.
- a binary feature vector is generated to include the feature data, and finally, the feature data is generated.
- Such a binary feature vector should be robust to each of the feature points and not alter the features unique to the feature points.
- FIGS. 4 and 5 A process of generating a binary feature vector in the feature data generator 23 will be described with reference to FIGS. 4 and 5.
- 4 is a screen illustrating an actual example of a process of generating a binary feature vector
- FIG. 5 is a flowchart illustrating a process of generating a binary feature vector.
- a peripheral image area including the feature points is generated in the form of a rectangle, for example, by using the size and direction information of the feature points. Align in the same direction (S500, S510)).
- generating the surrounding image area including the feature point in the shape of a rectangle may use size information included in the feature point information extracted by the feature point determiner 11.
- a rectangular peripheral image area may be generated using information such as length (for a square), length of a horizontal and vertical side (for a rectangle), radius (for a circle), and the like as size information.
- a square according to the length of the corresponding side may be generated, and if a length of the horizontal and vertical sides is given, a square having the maximum or minimum value of the side may be generated.
- a radius value is given, a square having the radius as the length of the side may be generated.
- the alignment of the generated surrounding image areas in the same direction is to obtain the same feature vector even when the target image is a rotated form of another image.
- a method of generating a quadrangular shape based on coordinates within a predetermined size, for example, ⁇ 10 around the feature point may be used instead of the size information of the feature point.
- each of the generated and aligned image areas is divided into N ⁇ N sub-regions as shown on the right side of FIG. 4 (S520).
- the feature data generator 23 selects at least one or more of the difference vector D (i, j) and the difference vector E (i, j, k, l) defined by the above equation and based on the result.
- a binary feature vector is generated (S540).
- FIG. 6 An example of a specific process of generating a binary feature vector based on the difference vector and the difference vector is shown in FIG. 6.
- the example of FIG. 6 corresponds to a case in which a binary feature vector includes M bits, and each process of FIG. 6 is repeated M times.
- the selection and generation of at least one of the difference vector and the difference vector should be performed over M times, which is the number of bits of the binary feature vector.
- the set of the difference vector and the difference vector should be different from each other. It is preferable to set in advance so that different sets of difference vectors and sets of difference vectors are selected.
- a linear combination calculation is performed on the selected and generated difference vectors and the difference vectors (S542). For example, if the selected and generated difference vectors are four of D (1,2), D (3,4), E (1,2,4,5) and E (3,5,6,7), respectively. Compute a linear combination for the values of (they will each have a difference value and a difference value of the average value of brightness as described above). That is, in the case of a linear combination, a linear combination represented such as aD (1,2) + bD (3,4) + cE (1,2,4,5) + dE (3,5,6,7) can be calculated. Where a, b, c, and d are arbitrary coefficients.
- step S542 a nonlinear combination including a nonlinear operation such as multiplication may be performed in addition to the linear combination, and in some cases, a linear combination and a nonlinear combination may be mixed. Can be.
- the result value is obtained, and it is determined whether the result value is greater than a predetermined threshold value, for example, 0 (S543), and if greater than 0, 1 is assigned to the corresponding bit, i. S544), if smaller than 0, 0 is assigned to the corresponding bit, i.e., the i-th bit (S545). In this way, the value of the i-th bit of the binary feature vector is determined.
- a predetermined threshold value for example, 0 (S543), and if greater than 0, 1 is assigned to the corresponding bit, i. S544), if smaller than 0, 0 is assigned to the corresponding bit, i.e., the i-th bit (S545).
- i M (i.e., the last bit) (S546), and if it is not the last bit, i is incremented (S547) and the above steps S541 to S547 are repeated. If the last bit is terminated (S548).
- a binary feature vector consisting of M bits represented by 0 or 1 for each value is generated for a given feature point.
- the binary feature vector shown on the right side of FIG. 4 is generated through such a process and consists of a total of 6 bits, and 0 or 1 is assigned to each bit.
- an alignment process may be further performed for each bit based on importance.
- the process of sorting the binary feature vectors according to the order that is, the robust order, may be performed. That is, when the process shown in FIG. 6 is performed, when the value of the M bit is expressed as "001010,” it may be arranged as "011001" according to importance, and such a binary feature vector is shown on the right side of FIG. It was.
- the binary feature vectors are arranged in order of importance, the comparison and retrieval of data in the order of importance in the post-processing process can be quickly processed based on the importance.
- the feature data generator 23 finally generates the feature data of the image including the other feature point information of the feature point including the binary feature vector.
- the other feature point information included in the feature data may include at least one of, for example, an x coordinate value, a y coordinate value, size information, and direction information of the feature point. All of this information may be included or only some of them may be selected and configured, which may be set differently depending on the conditions in the post-processing process.
- the final feature data generated when all the other feature point information is configured as described above may be a set of final feature points consisting of (x coordinate, y coordinate, size, direction, binary feature vector) for each feature point. have.
- the difference vector and the difference vector of the average value of the brightness values of the subregions are defined (step S530 of FIG. 5), and any one of them is based on the above-described criteria.
- the difference vector and the difference vector are generated in advance for all the average values of the brightness values of the subregions.
- the difference vector and the difference vector are generated in advance in step S530 of FIG. 5 with respect to the average value of the brightness values of all the subregions, and then the difference vector and the difference generated in step S541 of FIG. 6. Only the selection process is performed based on the criteria described above among the minute vectors. That is, the process of calculating and generating the difference vector and the difference vector in step S541 may be omitted.
- FIG. 7 is a diagram showing the configuration of an image characteristic data generation device 30 according to another embodiment of the present invention.
- the embodiment of FIG. 7 is the same as the embodiment of FIG. 3 except that the feature point determiner 21 further includes the feature point filter 24.
- the feature point filtration part 24 here is the same as what was demonstrated in FIGS. That is, the embodiment of FIG. 7 is the same as the embodiment of FIGS. 3 to 6, except that the feature point filtering unit 24 selects the feature points in an appropriate number as described with reference to FIGS. 1 to 2. 22) and the feature data generator 23 generate the feature data of the image. Therefore, in the embodiment of FIG. 7, since binary feature vectors are generated and feature data are formed for a smaller number of feature points than in the embodiments of FIGS. It has the advantage of being possible. In FIG. 7, other components are the same as described above, and thus a detailed description thereof will be omitted.
- post-processing such as image matching, object tracking, image identity determination, etc. by data having a smaller capacity while well representing the features of the original image using the generated feature data This allows for fast and efficient processing.
- the feature point determiner 21 determines a feature point from an image and extracts feature point information of the determined feature point (S800).
- the feature point direction estimator 22 estimates direction information on each of the determined feature points (S810).
- the feature data generation unit 23 generates a binary feature vector for each of the determined feature points based on the feature point information and the direction information (S820). This can be done through the following process as described above. That is, the peripheral image areas including the feature points are generated, the generated areas are aligned in the same direction, and the aligned peripheral image areas are each divided into sub areas. The average value of the brightness values of the divided sub-regions is obtained, the difference vector and the difference vector are calculated, and at least one of the difference vector and the difference vector is selected corresponding to each bit of the binary feature vector. And, by calculating a linear combination for the selected difference vector and the difference vectors and compares the result with a threshold value can be made through the process of determining the value of the corresponding bit of the binary feature vector.
- the feature data of the image is generated for each feature point including information on coordinates, sizes, directions, etc., including the binary feature vectors (S830). ).
- the present invention can be applied to a moving picture composed of a set of images. That is, since the video may be represented as a set of images that are still images composed of a plurality of frames, the present invention may be applied as it is when each frame constituting the video is regarded as an image of the present invention. In this case, the result is that feature data is generated for each frame of the video.
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
Description
Claims (17)
- 이미지의 특징 데이터를 생성하는 장치에 있어서,이미지로부터 특징점을 결정하고 결정된 특징점의 특징점 정보를 추출하는 특징점 결정부;상기 특징점 결정부에서 결정된 특징점 중에서 적어도 하나 이상의 특징점을 최종 특징점으로 결정하는 특징점 여과부; 및상기 특징점 여과부에서 결정된 최종 특징점과 최종 특징점의 특징점 정보에 기초하여 이미지의 특징 데이터를 생성하는 특징 데이터 생성부를 포함하는 이미지 특징 데이터 생성 장치.
- 제1항에 있어서,상기 특징점 결정부에서 추출되는 특징점 정보는 특징점의 강도를 포함하며,상기 특징점 여과부는 상기 특징점의 강도에 기초하여 특징점의 주변 영역에 위치하는 점들에 비하여 큰 강도를 갖는 점을 최종 특징점으로 결정하는 것을 특징으로 하는 이미지 특징 데이터 생성 장치.
- 이미지의 특징 데이터를 생성하는 방법에 있어서,이미지로부터 특징점을 결정하고 결정된 특징점의 특징점 정보를 추출하는 단계;상기 결정된 특징점 중에서 적어도 하나 이상의 특징점을 최종 특징점으로 결정하는 단계; 및상기 결정된 최종 특징점과 최종 특징점의 특징점 정보에 기초하여 이미지의 특징 데이터를 생성하는 단계를 구비하는 이미지의 특징 데이터 생성 방법.
- 이미지의 특징 데이터를 생성하는 장치에 있어서,이미지로부터 특징점을 결정하고 결정된 특징점의 특징점 정보를 추출하는 특징점 결정부;상기 특징점 결정부에서 결정된 특징점들 각각에 대한 방향 정보를 추정하는 특징점 방향 추정부; 및상기 특징점 결정부에서 결정된 특징점 각각에 대하여, 상기 특징점 정보와 상기 방향 정보에 기초하여 이진 특징 벡터를 생성하고, 이진 특징 벡터를 포함하는 이미지의 특징 데이터를 생성하는 특징 데이터 생성부를 포함하는 이미지 특징 데이터 생성 장치.
- 제5항에 있어서,상기 특징점 방향 추정부는,특징점 주변 일정 영역의 모든 점들에 대해서 각 점 주변의 그래디언트를 계산하고 그 방향의 평균값을 구함으로써 특징점의 방향을 추정하는 것을 특징으로 하는 이미지 특징 데이터 생성 장치.
- 제5항에 있어서,상기 특징 데이터 생성부는,상기 특징점 결정부에서 결정된 특징점 각각에 대하여, 상기 특징점 정보와 상기 방향 정보에 기초하여 이진 특징 벡터를 생성하고, 생성된 이진 특징 벡터를 포함하는 이미지의 특징 데이터를 생성하는 것을 특징으로 하는 이미지 특징 데이터 생성 장치.
- 제7항에 있어서,상기 특징 데이터 생성부는,각각의 특징점에 대하여 특징점을 포함하는 주변 이미지 영역을 생성하고 생성된 영역들을 동일 방향으로 정렬한 후, 정렬된 주변 이미지 영역들을 각각 부영역들로 분할하고, 분할된 부영역들의 밝기값의 평균값에 기초하여 이진 특징 벡터를 생성하는 것을 특징으로 하는 이미지 특징 데이터 생성 장치.
- 제8항에 있어서,상기 이진 특징 벡터는,상기 부영역들의 밝기값의 평균값의 차분 벡터와 차차분 벡터 중에서 선택된 적어도 어느 하나에 의해 생성되는 것을 특징으로 하는 이미지 특징 데이터 생성 장치.
- 제9항에 있어서,상기 부영역들의 밝기값의 평균값의 차분 벡터와 차차분 벡터 중에서 적어도 어느 하나를 선택하는 것은 이진 특징 벡터의 각 비트에 대응하여 선택되는 것을 특징으로 하는 이미지 특징 데이터 생성 장치.
- 제10항에 있어서,상기 각 비트에 대응하여 선택된 차분 벡터와 차차분 벡터들에 대해 선형 조합 또는 비선형 조합을 계산하고 그 결과값을 임계치와 비교함으로써 이진 특징 벡터의 해당 비트의 값을 결정하는 것을 특징으로 하는 이미지 특징 데이터 생성 장치.
- 제11항에 있어서,상기 이진 특징 벡터의 각 비트값에 대해 미리 설정해둔 기준에 따라 정렬을 수행하는 것을 특징으로 하는 이미지 특징 데이터 생성 장치.
- 제5항에 있어서,상기 이미지의 특징 데이터는 특징점의 위치 정보, 크기 정보, 방향 정보 중 적어도 어느 하나 이상을 더 포함하는 것을 특징으로 하는 이미지 특징 데이터 생성 장치.
- 제5항에 있어서,상기 특징점 결정부는,결정된 특징점 중에서 적어도 하나 이상의 특징점을 최종적으로 특징점으로서 결정하는 특징점 여과부를 더 포함하는 것을 특징으로 하는 이미지 특징 데이터 생성 장치.
- 제5항에 있어서,상기 특징점 결정부에서 추출되는 특징점 정보는 특징점의 강도를 포함하며,상기 특징점 여과부는 상기 특징점의 강도에 기초하여 특징점의 주변 영역에 위치하는 점들에 비하여 큰 강도를 갖는 점을 최종적으로 특징점으로서 결정하는 것을 특징으로 하는 이미지 특징 데이터 생성 장치.
- 이미지의 특징 데이터를 생성하는 방법에 있어서,이미지로부터 특징점을 결정하고 결정된 특징점의 특징점 정보를 추출하는 단계;상기 결정된 특징점들 각각에 대한 방향 정보를 추정하는 단계; 및상기 결정된 특징점 각각에 대하여, 상기 특징점 정보와 상기 방향 정보에 기초하여 이진 특징 벡터를 생성하고, 이진 특징 벡터를 포함하는 이미지의 특징 데이터를 생성하는 단계를 포함하는 이미지 특징 데이터 생성 방법.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US13/985,129 US8983199B2 (en) | 2011-02-14 | 2011-11-11 | Apparatus and method for generating image feature data |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR1020110012741A KR101165357B1 (ko) | 2011-02-14 | 2011-02-14 | 이미지 특징 데이터 생성 장치 및 방법 |
KR10-2011-0012741 | 2011-02-14 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2012111906A1 true WO2012111906A1 (ko) | 2012-08-23 |
Family
ID=46672789
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/KR2011/008600 WO2012111906A1 (ko) | 2011-02-14 | 2011-11-11 | 이미지 특징 데이터 생성 장치 및 방법 |
Country Status (3)
Country | Link |
---|---|
US (1) | US8983199B2 (ko) |
KR (1) | KR101165357B1 (ko) |
WO (1) | WO2012111906A1 (ko) |
Families Citing this family (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101165359B1 (ko) | 2011-02-21 | 2012-07-12 | (주)엔써즈 | 이미지와 이미지 또는 이미지와 동영상 사이의 상호 관계 분석 장치 및 방법 |
WO2013073273A1 (ja) * | 2011-11-18 | 2013-05-23 | 日本電気株式会社 | 局所特徴量抽出装置、局所特徴量抽出方法、及びプログラム |
KR101491461B1 (ko) * | 2013-08-02 | 2015-02-23 | 포항공과대학교 산학협력단 | 공분산 기술자를 이용하는 물체 인식 방법 및 장치 |
EP4057215A1 (en) * | 2013-10-22 | 2022-09-14 | Eyenuk, Inc. | Systems and methods for automated analysis of retinal images |
US20150271514A1 (en) * | 2014-03-18 | 2015-09-24 | Panasonic Intellectual Property Management Co., Ltd. | Prediction image generation method, image coding method, image decoding method, and prediction image generation apparatus |
JP6986721B2 (ja) * | 2014-03-18 | 2021-12-22 | パナソニックIpマネジメント株式会社 | 復号装置及び符号化装置 |
US11159798B2 (en) * | 2018-08-21 | 2021-10-26 | International Business Machines Corporation | Video compression using cognitive semantics object analysis |
CN110895699B (zh) * | 2018-09-12 | 2022-09-13 | 北京字节跳动网络技术有限公司 | 用于处理图像的特征点的方法和装置 |
CN110889904B (zh) * | 2019-11-20 | 2024-01-19 | 广州智能装备研究院有限公司 | 一种图像特征精简方法 |
CN111047650B (zh) * | 2019-12-02 | 2023-09-01 | 北京深测科技有限公司 | 一种用于飞行时间相机的参数标定方法 |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20000054305A (ko) * | 2000-06-01 | 2000-09-05 | 이성환 | 적은 수의 특징점을 이용한 얼굴 영상 압축과 손상된 얼굴영상의 복원 방법 및 장치 |
KR20080088778A (ko) * | 2007-03-30 | 2008-10-06 | 한국전자통신연구원 | Svd 기반의 영상 비교시스템 및 방법 |
KR20100097297A (ko) * | 2009-02-26 | 2010-09-03 | 인천대학교 산학협력단 | 마커를 사용하지 않는 증강공간 제공 장치 |
Family Cites Families (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6711293B1 (en) | 1999-03-08 | 2004-03-23 | The University Of British Columbia | Method and apparatus for identifying scale invariant features in an image and use of same for locating an object in an image |
IT1311443B1 (it) * | 1999-11-16 | 2002-03-12 | St Microelectronics Srl | Metodo di classificazione di immagini digitali in base al lorocontenuto. |
JP4492036B2 (ja) * | 2003-04-28 | 2010-06-30 | ソニー株式会社 | 画像認識装置及び方法、並びにロボット装置 |
JP2006268825A (ja) * | 2005-02-28 | 2006-10-05 | Toshiba Corp | オブジェクト検出装置、学習装置、オブジェクト検出システム、方法、およびプログラム |
US20070160308A1 (en) * | 2006-01-11 | 2007-07-12 | Jones Michael J | Difference of sum filters for texture classification |
EP1850270B1 (en) | 2006-04-28 | 2010-06-09 | Toyota Motor Europe NV | Robust interest point detector and descriptor |
KR100866792B1 (ko) * | 2007-01-10 | 2008-11-04 | 삼성전자주식회사 | 확장 국부 이진 패턴을 이용한 얼굴 기술자 생성 방법 및장치와 이를 이용한 얼굴 인식 방법 및 장치 |
JP4988408B2 (ja) * | 2007-04-09 | 2012-08-01 | 株式会社デンソー | 画像認識装置 |
FR2931277B1 (fr) * | 2008-05-19 | 2010-12-31 | Ecole Polytech | Procede et dispositif de reconnaissance invariante-affine de formes |
US8457409B2 (en) * | 2008-05-22 | 2013-06-04 | James Ting-Ho Lo | Cortex-like learning machine for temporal and hierarchical pattern recognition |
US8150170B2 (en) * | 2008-05-30 | 2012-04-03 | Microsoft Corporation | Statistical approach to large-scale image annotation |
CN101609506B (zh) * | 2008-06-20 | 2012-05-23 | 索尼株式会社 | 用于识别图像中的模型对象的装置及方法 |
KR101622110B1 (ko) * | 2009-08-11 | 2016-05-18 | 삼성전자 주식회사 | 특징점 추출 방법 및 추출 장치, 이를 이용한 영상 기반 위치인식 방법 |
US8625902B2 (en) * | 2010-07-30 | 2014-01-07 | Qualcomm Incorporated | Object recognition using incremental feature extraction |
US20140093142A1 (en) * | 2011-05-24 | 2014-04-03 | Nec Corporation | Information processing apparatus, information processing method, and information processing program |
US9082235B2 (en) * | 2011-07-12 | 2015-07-14 | Microsoft Technology Licensing, Llc | Using facial data for device authentication or subject identification |
US8453075B2 (en) * | 2011-09-02 | 2013-05-28 | International Business Machines Corporation | Automated lithographic hot spot detection employing unsupervised topological image categorization |
US8774508B2 (en) * | 2012-02-27 | 2014-07-08 | Denso It Laboratory, Inc. | Local feature amount calculating device, method of calculating local feature amount, corresponding point searching apparatus, and method of searching corresponding point |
IL226219A (en) * | 2013-05-07 | 2016-10-31 | Picscout (Israel) Ltd | Efficient comparison of images for large groups of images |
-
2011
- 2011-02-14 KR KR1020110012741A patent/KR101165357B1/ko active IP Right Grant
- 2011-11-11 US US13/985,129 patent/US8983199B2/en active Active
- 2011-11-11 WO PCT/KR2011/008600 patent/WO2012111906A1/ko active Application Filing
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20000054305A (ko) * | 2000-06-01 | 2000-09-05 | 이성환 | 적은 수의 특징점을 이용한 얼굴 영상 압축과 손상된 얼굴영상의 복원 방법 및 장치 |
KR20080088778A (ko) * | 2007-03-30 | 2008-10-06 | 한국전자통신연구원 | Svd 기반의 영상 비교시스템 및 방법 |
KR20100097297A (ko) * | 2009-02-26 | 2010-09-03 | 인천대학교 산학협력단 | 마커를 사용하지 않는 증강공간 제공 장치 |
Also Published As
Publication number | Publication date |
---|---|
US8983199B2 (en) | 2015-03-17 |
US20140050411A1 (en) | 2014-02-20 |
KR101165357B1 (ko) | 2012-07-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2012111906A1 (ko) | 이미지 특징 데이터 생성 장치 및 방법 | |
US10872262B2 (en) | Information processing apparatus and information processing method for detecting position of object | |
WO2012115332A1 (ko) | 이미지와 이미지 또는 이미지와 동영상 사이의 상호 관계 분석 장치 및 방법 | |
CN106952269B (zh) | 近邻可逆的视频前景物体序列检测分割方法及*** | |
CN108121991B (zh) | 一种基于边缘候选区域提取的深度学习舰船目标检测方法 | |
CN109918971B (zh) | 监控视频中人数检测方法及装置 | |
JP2018514024A5 (ko) | ||
Fu et al. | Learning to detect specular highlights from real-world images | |
CN102982539B (zh) | 一种基于图像复杂性的特征自适应图像共分割方法 | |
CN108510496B (zh) | 基于图像dct域的svd分解的模糊检测方法 | |
WO2014036813A1 (zh) | 图像特征的提取方法及装置 | |
EP3678046A1 (en) | Hand detection method and system, image detection method and system, hand segmentation method, storage medium, and device | |
WO2016108327A1 (ko) | 차량검출방법, 차량검출을 위한 데이터베이스의 구조, 및 차량검출을 위한 데이터베이스 구축방법 | |
CN111310768B (zh) | 基于鲁棒性背景先验和全局信息的显著性目标检测方法 | |
CN108038458B (zh) | 基于特征摘要图的视频中的户外场景文本自动获取方法 | |
CN111833347A (zh) | 一种输电线路防振锤缺陷检测方法及相关装置 | |
Sahoo et al. | A fast valley-based segmentation for detection of slowly moving objects | |
WO2024016632A1 (zh) | 亮点定位方法、亮点定位装置、电子设备及存储介质 | |
CN115984341B (zh) | 海洋水质微生物检测方法、装置、设备及存储介质 | |
CN106611417B (zh) | 将视觉元素分类为前景或背景的方法及装置 | |
CN109741370B (zh) | 一种目标跟踪方法和装置 | |
CN113221606B (zh) | 一种基于ims视频会议登录的人脸识别方法 | |
CN105184244B (zh) | 视频人脸检测方法及装置 | |
CN113705660A (zh) | 目标识别方法及相关设备 | |
KR20060064801A (ko) | 영상 인식 장치 및 방법 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 11858559 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
WWE | Wipo information: entry into national phase |
Ref document number: 13985129 Country of ref document: US |
|
32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 13-12-2013) |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 11858559 Country of ref document: EP Kind code of ref document: A1 |