Apparatus and method for generating higher-order LoG scale space for keypoint detection

Abstract

The present invention relates to an apparatus and method for generating a higher-order LoG scale space for feature point detection. The apparatus for generating a higher-order LoG scale space for feature point detection according to the present invention includes a receiving module that receives an input image from an external source, a memory that stores the input image, a processor that generates a Gaussian scale space using the stored input image and generates a higher-order LoG scale space using the Gaussian scale space, and a transmitting module that outputs the higher-order LoG scale space. When using the higher-order LoG scale space according to the present invention for feature point detection, the utilization of the edge portions of the image is increased, and more feature points can be detected than those obtained by directly using a higher-order LoG kernel. Consequently, superior effects can be expected when performing feature point-based image processing.

Description

Technology Field

[0001] The present invention relates to an apparatus and method for generating a high-order LoG (Laplacian of Gaussian) scale space for feature point detection. More specifically, the present invention relates to an apparatus and method for generating a scale space using a high-order LoG kernel that can be used for detecting feature points of an image for image retrieval, panoramic image generation, etc. Background Technology

[0002] A feature point refers to a point within an image or video that can serve as a characteristic—that is, a point containing important information about the image or video (hereinafter referred to as "image"). These points can appear at unique locations within each image and are robust against affine or non-standard transformations, which are geometric transformations that alter the shape of the entire image—such as translation, scaling, and rotation—by changing the arrangement structure of the pixels that constitute the image. Feature points can be used to perform image search by locating desired objects within an image, to track or recognize objects, or to generate panoramic images by matching images.

[0003] Therefore, the process of detecting accurate and unique feature points is crucial for deriving effective results. A wider candidate region within an image where feature points can be detected increases the number of detected feature points, which can be advantageous for performance. To detect these feature points, one can follow a method of applying a Laplacian of Gaussian (LoG) filter after a Gaussian filter—that is, applying partial derivatives of the Gaussian function along each axis—or utilize the Difference of Gaussian (DoG) method, which uses the difference between Gaussian pyramid images obtained by applying different filters to a single image.

[0004] SIFT (Scale Invariant Feature Transform), a representative image feature detection technology based on DoG (Difference of Gaussian), is characterized by constructing a Gaussian scale space and utilizing it to construct a DoG scale space for feature detection.

[0005] Figure 1 shows images of the DoG scale space according to octave and scale number of the SIFT technology. The DoG scale space is generated using images of the Gaussian scale space and can be composed of difference images of two consecutive scale images within the same octave of the Gaussian scale space. For example, an image obtained by subtracting a scale 0 image from a scale 1 image within the same octave of the Gaussian scale space is generated as a scale 0 image of the first-order DoG scale space and can be included in the DoG scale space. Images generated in this way are used to detect feature points within the image.

[0006] However, when using higher-order LoG (Laplacian of Gaussian) kernels, there is a problem in that the existing DoG scale space cannot be used, so a method for generating a higher-order LoG scale space is required. The problem to be solved

[0007] The present invention aims to provide an apparatus and method for generating a higher-order LoG scale space for feature point detection, in accordance with the need for generating the aforementioned higher-order LoG scale space.

[0008] More specifically, the purpose is to provide an apparatus and method that can improve the utilization of image edge regions during feature point detection using a high-order LoG kernel and generate an octave-based scale space.

[0009] Other objects and advantages of the present invention may be understood from the following description and will become more clearly apparent from the embodiments of the present invention. Furthermore, it will be readily apparent that the objects and advantages of the present invention can be realized by the means and combinations thereof set forth in the claims. means of solving the problem

[0010] To achieve the above objective, the device for generating a higher-order LoG scale space for feature point detection according to the present invention comprises a receiving module for receiving an input image from the outside, a memory for storing the input image, a processor for generating a Gaussian scale space using the stored input image and generating a higher-order LoG scale space using the Gaussian scale space, and a transmitting module for outputting the higher-order LoG scale space. Effects of the invention

[0011] The present invention relates to an apparatus and method for generating a higher-order LoG scale space for feature point detection.

[0012] According to the present invention, a higher-order LoG scale space within an octave can be generated and utilized for detecting feature points. Furthermore, by utilizing the scale space, the utilization of the edge portions of the image during feature point detection is increased. Consequently, more feature points can be detected than those obtained by directly using a higher-order LoG kernel, and superior effects can be expected when performing feature point-based image processing using this. Brief explanation of the drawing

[0013] Figure 1 is for the DoG scale space images by octave and scale. Figure 2 is for Gaussian scale space images by octave and scale. FIG. 3 illustrates a Gaussian scale space according to one embodiment of the present invention. FIG. 4 illustrates a higher-order LoG scale space according to one embodiment of the present invention. FIG. 5 illustrates a high-order LoG scale space generation device for feature point detection according to one embodiment of the present invention. FIG. 6 illustrates a feature point detection process using a higher-order LoG scale space according to an embodiment of the present invention. Specific details for implementing the invention

[0014] Hereinafter, embodiments of the present invention are described in detail with reference to the attached drawings so that those skilled in the art can easily implement the present invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein.

[0015] In describing the embodiments of the present invention, if it is determined that a detailed description of known configurations or functions may obscure the essence of the invention, such detailed description is omitted. Additionally, parts of the drawings unrelated to the description of the invention have been omitted, and similar parts are denoted by similar reference numerals.

[0016] In the present invention, distinct components are intended to clearly explain their respective features and do not necessarily imply that the components are separated. That is, multiple components may be integrated to form a single hardware or software unit, or a single component may be distributed to form multiple hardware or software units. Therefore, such integrated or distributed embodiments are included within the scope of the present invention, even without separate mention.

[0017] In the present invention, the components described in various embodiments do not necessarily mean essential components, and some may be optional components. Accordingly, embodiments composed of a subset of the components described in one embodiment are also included within the scope of the present invention. Furthermore, embodiments including additional components in addition to the components described in various embodiments are also included within the scope of the present invention.

[0018] Hereinafter, embodiments of the present invention will be described with reference to the attached drawings.

[0020] FIG. 2 is an image of a Gaussian scale space by octave and scale. FIG. 3 also illustrates a Gaussian scale space according to an embodiment of the present invention. More specifically, it illustrates a method for generating a Gaussian scale space for generating a higher-order LoG (Laplacian of Gaussian) scale space.

[0021] To ensure that feature point detection is not affected by the image scale (capture distance), multiple images are generated by applying resizing and a Gaussian filter; the set of these generated multiple images is called the Gaussian Scale Space.

[0022] Among the images in the Gaussian scale space, the set of images to which a Gaussian filter is applied at the original resolution of the input image is denoted as Octave 0, the set of images to which a Gaussian filter is applied when the horizontal and vertical resolutions are each reduced by half (size by 1 / 4) is denoted as Octave 1, the set of images to which a Gaussian filter is applied when the horizontal and vertical resolutions are each reduced by half compared to the images in Octave 1 is denoted as Octave 2, and the set of images to which a Gaussian filter is applied when the horizontal and vertical resolutions of the input image are each adjusted by 2 times (size by 4 times) is denoted as Octave -1. Images to which a Gaussian filter is applied within the Octave can be denoted in ascending order (scale 0, 1, 2, etc.) starting from the smallest smoothing intensity (depending on the standard deviation value when generating the Gaussian filter).

[0023] Figure 2 is for an image corresponding to the octave and scale numbers according to the above description. That is, as the octave number increases, the resolution decreases, and as the scale number increases, the smoothing intensity may also increase.

[0024] For example, an image of Octave 0, Scale 0 is an image smoothed with a Gaussian filter with a standard deviation of 1.6. Assuming that the input image has already been smoothed with a Gaussian filter having a standard deviation of 1, the image is the input image with a standard deviation This can be obtained by performing smoothing with the applied Gaussian filter.

[0025] Since this can be applied in the same way to the remaining scales and other octaves, a Gaussian filter can be generated and applied using the difference in standard deviations between scales. Within each octave, the standard deviation (to the image at scale 0) ) 1.6 is applied, and as the scale increases thereafter In order You can obtain an image smoothed with a Gaussian filter that applies a standard deviation increased by a factor of two.

[0026] for example, cast If set to this, a smoothed image is obtained from the scale 3 image with a standard deviation twice that of the scale 0 image applied. This image is resized to 1 / 4 of its size to generate the scale 0 image for the next octave. The image obtained through this downsampling is considered to be a smoothed image with a standard deviation of 1.6 already applied, and images corresponding to subsequent scales are generated using the same method as before. The above relationship can be extended and applied as is, even if the octave and scale increase.

[0027] In generating a Gaussian scale space for generating a higher-order LoG scale space according to an embodiment of the present invention described with reference to FIG. 3, the method described above may be followed. However, since an image generated by resizing an image corresponding to the upper octave scale 3 is required for generating a scale 0 image of the octaves excluding octave -1 to generate a higher-order LoG scale space, such an image must be generated and included. This is the same as the images G[0][0] and G[1][0] displayed within the dotted box (301). The image within the dotted box (301) is [x] is the Gaussian filter to be applied to the scale x image to generate the image corresponding to scale x+1. It can represent a value.

[0028] After generating a Gaussian scale space, a higher-order LoG scale space can be generated using it. By distributing the standard deviation σ of the Gaussian filter applied to the higher-order LoG scale space used for feature point detection across the Gaussian scale space and the higher-order LoG kernel, the utilization of the edge portions of the image can be increased. This will be explained in more detail with reference to Fig. 4.

[0030] FIG. 4 illustrates a higher-order LoG scale space according to an embodiment of the present invention. More specifically, it shows a method for generating a higher-order LoG scale space by utilizing a Gaussian scale space. At this time, the Gaussian scale space may be generated through the process of FIG. 3.

[0031] As an example, to generate a scale space by applying a high-order LoG kernel, the value of LoG obtained by summing a 2D Gaussian equation and its second partial derivatives with respect to x and y can be expressed as the following mathematical formula.

[0032] In the following mathematical formula, the Gaussian expression in the LoG calculation ...was omitted.

[0033] [Mathematical Formula]

[0034]

[0035] Higher-order LoGs can be obtained by differentiating the above Gaussian equation with respect to x and y to the 4th, 6th, 8th, etc. Using this formula, higher-order LoG kernels can be obtained, and You can create a scale space by changing the value. In this case, the size of the kernel is It must increase as the value increases.

[0036] In one embodiment, L[x][y] represents an image corresponding to octave x and scale y of a higher-order LoG scale space. A higher-order LoG scale space can be generated by applying a higher-order LoG kernel to an image in a Gaussian scale space. is a higher-order LoG filter to be applied to a scale x image in Gaussian scale space to generate a LoG scale space image corresponding to scale x. Represents the value. Octave -1, Scale 0 in Gaussian scale space of the Gaussian filter applied to the image Since it is different from the scale 0 images of other octaves There may be differences in values. The LoG kernels by order are It can be determined based on the value, and by applying this to the Gaussian scale space, a higher-order LoG scale space can be generated.

[0037] Although the image for Scale 0 (excluding Octave -1) was generated by resizing the image corresponding to Scale 3 of the higher octave, it can also be obtained by resizing the images corresponding to 0, 1, and 2 of the higher octave, and to be applied when generating a higher-order LoG scale space according to that image Only the value can be adjusted.

[0038] In detecting feature points of an image, a higher-order LoG scale space generated through the process of Fig. 4 can be used. For example, the location corresponding to the maximum / minimum value can be detected as the location of the feature point by comparing it with surrounding pixel values ​​at the same pixel location as an adjacent scale image.

[0040] FIG. 5 illustrates a high-order LoG scale space generation device for feature point detection according to one embodiment of the present invention.

[0041] More specifically, the above device (501) is intended to describe a device for generating a higher-order LoG scale space, comprising a receiving module (502) that receives an image as input from an external source (500), a memory (503) for storing the received image, a processor (504) that generates a Gaussian scale space through the image stored in the memory (503) and generates a higher-order LoG scale space using the image, and a transmitting module (505) that outputs the higher-order LoG scale space (506) for use in detecting feature points.

[0042] As an example, the receiving module (502) can receive an image from the outside (500) to be used to create a scale space.

[0043] The memory (503) stores the received image. It then transmits the received image to the processor (504) so ​​that the processor (504) can use it to create a scale space.

[0044] The processor (504) uses the image stored in the memory (503) to generate a plurality of images by applying resizing and a Gaussian filter so that the scale (shooting distance) of the image is not affected when detecting feature points, and uses this to generate a Gaussian scale space. The Gaussian scale space is intended to generate a higher-order LoG scale space, and the process of generating the Gaussian scale space may follow the method described above with reference to FIG. 3.

[0045] When the above Gaussian scale space is generated, a higher-order LoG scale space is generated using it. The method for generating the above higher-order LoG scale space can follow the method described above with reference to FIG. 4.

[0046] The higher-order LoG scale space generated through the processor (504) can be output (506) by the transmission module (505) and used for image feature point detection.

[0047] According to the present invention, by using a high-order LoG kernel, the utilization of the edge portions of an image is increased during feature point detection, and feature points can be detected by generating a scale space within an octave.

[0048] Therefore, more feature points can be detected than those obtained by directly using a higher-order LoG kernel, and superior results can be expected when performing feature-based image processing using this.

[0050] FIG. 6 illustrates a feature point detection process using a higher-order LoG scale space according to an embodiment of the present invention. First, an input image is received from an external source (S601), and a Gaussian scale space is generated (S602) by applying resizing and a Gaussian filter to the received input image. A higher-order LoG kernel is applied to the Gaussian scale space to generate a higher-order LoG scale space (S603), and feature points of the image can be detected (S604) using the higher-order LoG scale space.

[0051] As an example, the Gaussian scale space can be generated according to the process described with reference to FIG. 3, and the higher-order LoG scale space can also be generated according to the process described in FIG. 4.

[0053] The various embodiments of the present disclosure are not intended to list all possible combinations but to describe representative aspects of the present disclosure, and the matters described in the various embodiments may be applied independently or in combination of two or more.

[0054] In addition, various embodiments of the present disclosure may be implemented by hardware, firmware, software, or a combination thereof. In the case of implementation by hardware, it may be implemented by one or more ASICs (Application Specific Integrated Circuits), DSPs (Digital Signal Processors), DSPDs (Digital Signal Processing Devices), PLDs (Programmable Logic Devices), FPGAs (Field Programmable Gate Arrays), general processors, controllers, microcontrollers, microprocessors, etc.

[0055] The scope of the present disclosure includes software or machine-executable instructions (e.g., an operating system, an application, firmware, a program, etc.) that enable an operation according to a method of various embodiments to be executed on a device or computer, and a non-transitory computer-readable medium on which such software or instructions, etc. are stored and which is executable on a device or computer.

[0056] The present invention described above is not limited by the aforementioned embodiments and attached drawings, as various substitutions, modifications, and changes are possible within the scope of the technical concept of the present invention without departing from the technical spirit of the present invention to those skilled in the art to which the present invention belongs. Explanation of the symbols

[0057] 500: External 501: Higher-order LoG scale space generator for feature point detection 502: Receive module 503: Memory 504: Processor 505: Transmission Module 506: Output

Claims

Claim 1 A device for generating a high-order LoG (Laplacian of Gaussian) scale space for detecting feature points, comprising: a receiving module for receiving an input image from the outside; a memory for storing the input image; a processor for generating a Gaussian scale space using the stored input image and generating a high-order LoG scale space using the Gaussian scale space; and a transmitting module for outputting the high-order LoG scale space, wherein the high-order LoG scale space is generated by applying a high-order LoG kernel to the Gaussian scale space, and the standard deviation value of the Gaussian filter applied to the high-order LoG scale space is distributed and applied during the process of generating the Gaussian scale space and the process of applying the high-order LoG kernel so as to maintain the characteristics of the scale space formed by the Gaussian scale space. Claim 2 A high-order LoG scale space generation device for feature point detection according to claim 1, wherein the processor generates a Gaussian scale space by sequentially changing the resolution of the input image, applying a Gaussian filter to generate a resolution-specific image, and generating a scale-specific image by applying different smoothing intensities to each resolution-specific image. Claim 3 A high-order LoG scale space generation device for feature point detection, wherein, in paragraph 2, the processor generates a Gaussian filter based on the difference in standard deviation between scales and generates a scale-specific image using the generated Gaussian filter. Claim 4 delete Claim 5 A high-order LoG scale space generation device for feature point detection, wherein, in claim 1, the processor determines a LoG kernel of a specific order according to the value of the standard deviation and applies the determined LoG kernel to the Gaussian scale space. Claim 6 A method for generating a higher-order LoG scale space for feature point detection, comprising: a step of receiving an input image from an external source by a receiving module; a step of generating a Gaussian scale space using the input image by a processor; and a step of generating a higher-order LoG scale space using the Gaussian scale space by the processor, wherein the higher-order LoG scale space is generated by applying a higher-order LoG kernel to the Gaussian scale space, and the standard deviation value of the Gaussian filter applied to the higher-order LoG scale space is distributed and applied to the step of generating the Gaussian scale space and the step of applying the higher-order LoG kernel so as to maintain the characteristics of the scale space formed by the Gaussian scale space. Claim 7 A method for generating a high-order LoG scale space for feature point detection according to claim 6, wherein the step of generating the Gaussian scale space is performed by sequentially changing the resolution of the input image, applying a Gaussian filter to generate an image for each resolution, and applying different smoothing intensities to each of the images for each resolution to generate an image for each scale. Claim 8 A method for generating a high-order LoG scale space for feature point detection, wherein, in claim 7, the step of generating the Gaussian scale space comprises generating a Gaussian filter based on the difference in standard deviation between scales and generating a scale-specific image using the generated Gaussian filter. Claim 9 delete Claim 10 A method for generating a higher-order LoG scale space for feature point detection according to claim 6, wherein the step of generating the higher-order LoG scale space is performed by determining a LoG kernel of a certain order according to the value of the standard deviation and applying the determined LoG kernel of a certain order to the Gaussian scale space.

Images