WO2014002813A1 - 画像処理装置、画像処理方法および画像処理プログラム - Google Patents
画像処理装置、画像処理方法および画像処理プログラム Download PDFInfo
- Publication number
- WO2014002813A1 WO2014002813A1 PCT/JP2013/066586 JP2013066586W WO2014002813A1 WO 2014002813 A1 WO2014002813 A1 WO 2014002813A1 JP 2013066586 W JP2013066586 W JP 2013066586W WO 2014002813 A1 WO2014002813 A1 WO 2014002813A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- image
- template
- resolution
- reference image
- input
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/001—Industrial image inspection using an image reference approach
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/14—Measuring arrangements characterised by the use of optical techniques for measuring distance or clearance between spaced objects or spaced apertures
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/37—Determination of transform parameters for the alignment of images, i.e. image registration using transform domain methods
-
- 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
- G06T7/74—Determining position or orientation of objects or cameras using feature-based methods involving reference images or patches
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N23/00—Cameras or camera modules comprising electronic image sensors; Control thereof
- H04N23/60—Control of cameras or camera modules
- H04N23/667—Camera operation mode switching, e.g. between still and video, sport and normal or high- and low-resolution modes
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N25/00—Circuitry of solid-state image sensors [SSIS]; Control thereof
- H04N25/60—Noise processing, e.g. detecting, correcting, reducing or removing noise
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
Definitions
- the present invention relates to an image processing apparatus, an image processing method, and an image processing program for searching an area corresponding to a reference image from an input image.
- phase-only correlation (POC) method searches for corresponding points between images using phase difference information of spatial frequencies included in the images.
- Patent Document 1 discloses a pattern matching apparatus using the RIPOC method. This pattern collation apparatus collates an N-dimensional pattern (for example, fingerprint (two-dimensional) or three-dimensional (three-dimensional)) based on the spatial frequency characteristics. It is intended to be able to identify whether or not are the same.
- N-dimensional pattern for example, fingerprint (two-dimensional) or three-dimensional (three-dimensional)
- Non-Patent Document 1 shows an example of a POC method for size / rotation. More specifically, Non-Patent Document 1 discusses the POC method using Fourier Merin transform.
- the POC method and the size / rotation-compatible POC method as described above both use the spatial frequency phase difference information included in the image, so that there is an advantage that robust position detection can be performed, but the processing amount is very large. Also, there is a demerit that it takes calculation time. For example, when a matching process for searching for a position where a specific pattern (reference image / template image) included in an input image exists is performed using the size / rotation-compatible POC method, a raster for position search is used. Scanning is required, and the computation time required for the scanning can be enormous.
- an image processing apparatus that searches an input image for a region corresponding to a reference image.
- An image processing apparatus includes an input image acquisition unit that acquires first and second input images corresponding to an input image and having at least a first resolution and a second resolution higher than the first resolution, and a reference image Corresponding reference image acquisition unit for acquiring first and second reference images having resolutions respectively corresponding to the first and second resolutions, and a search between the first input image and the first reference image By processing, a relative shift amount of rotation and / or magnification between the first input image and the first reference image is calculated, and the first reference image based on the calculated relative shift amount is calculated.
- a first corresponding position determination unit that generates a first correction reference image by correction and determines a corresponding position at the first resolution by a search process between the first input image and the first correction reference image; To the first resolution And using the relative shift amount calculated by the first corresponding position determination unit using the position in the second input image corresponding to the corresponding position as a reference, and the second input image and the second reference image, And a second corresponding position determining unit that determines a corresponding position at the second resolution by performing a search process between the first and second resolutions.
- the second corresponding position determination unit restricts the search range according to the relative shift amount between the first input image and the first reference image, and then determines the second input image and the first input image.
- a relative shift amount of rotation and / or magnification between the second input image and the second reference image is calculated, and the calculated relative
- a second correction reference image is generated by correcting the second reference image based on the amount of deviation, and the search processing between the second input image and the second correction reference image is used to cope with the second resolution. Determine the position.
- the second corresponding position determination unit causes the reference image acquisition unit to generate the second reference image according to the relative shift amount between the first input image and the first reference image.
- the image processing apparatus calculates the similarity calculated in the search process between the first input image and the first reference image, and between the first input image and the first correction reference image.
- a control unit is further included that changes the method of determining the corresponding position at the second resolution by the second corresponding position determination unit based on at least one of the similarities calculated in the search process.
- an image processing method for searching a region corresponding to a reference image from an input image corresponds to an input image, the first and second input images having at least a first resolution and a second resolution higher than the first resolution, respectively, corresponding to the input image; Obtaining first and second reference images having resolutions respectively corresponding to the first and second resolutions; A search process between the first input image and the first reference image calculates a relative shift amount between rotation and / or magnification between the first input image and the first reference image, and A first correction reference image is generated by correcting the first reference image based on the calculated relative shift amount, and the first correction reference image is searched for between the first input image and the first correction reference image.
- Determining a corresponding position at a resolution of A search process between the second input image and the second reference image is performed using a relative shift amount with reference to the position in the second input image corresponding to the corresponding position in the first resolution.
- An image processing method comprising: determining a corresponding position at the second resolution.
- an image processing program for searching an area corresponding to a reference image from an input image.
- the image processing program obtains first and second input images corresponding to the input image and having at least a first resolution and a second resolution higher than the first resolution, respectively, and the reference image.
- the relative shift amount of rotation and / or magnification between one input image and the first reference image is calculated, and the first reference image is corrected based on the calculated relative shift amount, thereby correcting the first reference image.
- FIG. 1 It is a figure which shows an example of the input image used in the template matching process according to Embodiment 1 of this invention, a hierarchy image, and a template image. It is a figure which outlines the content of the template matching process according to Embodiment 1 of this invention. It is a block diagram which shows the function structure of the image processing apparatus according to Embodiment 1 of this invention. It is a figure which outlines the production
- search processing is efficiently performed by a coarse / fine strategy using multiple resolutions.
- the relative shift amount of the rotation and / or magnification between the image to be matched and the template image is evaluated, and after this is corrected, the position search is performed. Processing is executed.
- the relative shift amount refers to the relative shift amount between the image to be processed and the template image (reference image) and the relative magnification (size). This means at least one of the shift amounts due to the difference.
- information such as a relative shift amount and a matching position is propagated from the low resolution side to the high resolution side, thereby improving the efficiency of the search process.
- FIG. 1 is a schematic diagram showing an application example of template matching according to the embodiment of the present invention.
- system 1 according to the present embodiment is applied to a production line including belt conveyor 2 as an example.
- the object 3 (work) is continuously conveyed on the belt conveyor 2, and an image including the appearance of the object 3 is acquired by imaging the object 3 with the camera 4. (Hereinafter, this acquired image is also referred to as “input image”).
- the objects 3 conveyed on the belt conveyor 2 can be arranged in various directions, and therefore the rotation direction and / or size (enlargement / reduction ratio) of the region representing the object 3 included in the input image is determined. It is not always constant. Therefore, image processing apparatus 100 according to the present embodiment executes matching processing while adjusting the orientation and / or size of the template image. That is, the image processing apparatus 100 searches for a corresponding region in the input image after correcting the template image with some rotation amount and / or magnification.
- the image processing apparatus 100 may output information (rotation amount and / or magnification) indicating the degree of correction performed in the template matching process in addition to the position information indicating the corresponding region as a search result.
- FIG. 2 is a block diagram showing a configuration when template matching according to the embodiment of the present invention is realized by a personal computer.
- the CPU 102 controls the entire image processing apparatus 100 by executing various programs such as an operating system (OS: Operating System) and a template matching processing program 112 stored in the ROM 106, the auxiliary storage device 110, and the like.
- the RAM 104 functions as a working memory for executing a program by the CPU 102, and temporarily stores various data necessary for executing the program.
- the ROM 106 stores an initial program (boot program) that is executed when the image processing apparatus 100 is started.
- the network interface 108 exchanges data with other devices (such as server devices) via various communication media. More specifically, the network interface 108 is connected via a wired line such as Ethernet (registered trademark) (LAN (Local Area Network), WAN (Wide Area Network), etc.) and / or a wireless line such as a wireless LAN. Perform data communication.
- a wired line such as Ethernet (registered trademark) (LAN (Local Area Network), WAN (Wide Area Network), etc.) and / or a wireless line such as a wireless LAN. Perform data communication.
- the auxiliary storage device 110 typically includes a large-capacity magnetic recording medium such as a hard disk and the like, and includes an image processing program (template matching processing program 112) and a template generation image for realizing various processes according to the present embodiment. 114 and the like are stored. Further, the auxiliary storage device 110 may store a program such as an operating system.
- the template generation image 114 is used to generate a template image.
- search processing is efficiently performed by a coarse / fine strategy using multiple resolutions.
- this coarse / fine strategy requires a template image of each resolution
- the template generation image 114 is classified by resolution. Preferably stored. Details of the density strategy and the template image will be described later.
- the display unit 120 displays a GUI (Graphical User Interface) screen provided by the operating system, an image generated by executing the template matching processing program 112, and the like.
- GUI Graphic User Interface
- the memory card interface 124 reads and writes data with various memory cards (non-volatile recording media) 126 such as an SD (Secure Digital) card and a CF (Compact Flash (registered trademark)) card.
- non-volatile recording media such as an SD (Secure Digital) card and a CF (Compact Flash (registered trademark)) card.
- the camera interface 128 captures from the camera 4 a template generation image and / or an input image acquired by imaging a subject.
- the main body of the image processing apparatus 100 may not have a function of capturing an image of a subject.
- a necessary image is captured via a memory card 126 that stores a template generation image and / or an input image acquired by some device. That is, the memory card 126 is mounted on the memory card interface 124, and the template generation image and / or the input image read from the memory card 126 are stored (copied) in the auxiliary storage device 110 or the like.
- the template matching processing program 112 stored in the auxiliary storage device 110 is stored and distributed in a recording medium such as a CD-ROM (Compact Disk-Read Only Memory) or distributed from a server device or the like via a network. .
- the template matching processing program 112 implements processing by calling necessary modules among program modules provided as part of an operating system executed by the image processing apparatus 100 (personal computer) at a predetermined timing and order. It may be.
- the template matching processing program 112 itself does not include a module provided by the operating system, and image processing is realized in cooperation with the operating system.
- the template matching processing program 112 may be provided by being incorporated in a part of some program instead of a single program.
- the template matching processing program 112 itself does not include a module that is commonly used in the certain program, and image processing is realized in cooperation with the certain program. Even the template matching processing program 112 that does not include some of the modules does not depart from the spirit of the image processing apparatus 100 according to the present embodiment.
- templates matching processing program 112 may be realized by dedicated hardware.
- ⁇ b3: Realization Example with Other Configuration
- the personal computer described above may be mounted on a digital camera, a mobile phone, or a smartphone. Further, it may be in the form of a so-called cloud service in which at least one server device realizes processing according to the present embodiment.
- the user transmits at least two processing target images to the server device (cloud side) using his / her terminal (such as a personal computer or a smartphone), and the server device is transmitted to the transmitted processing target image.
- the server device side is assumed.
- ⁇ c1 Outline of processing >>
- a template that combines a search process that takes into account a relative shift between an input image / hierarchical image (partial image included) and a template image and a coarse / fine strategy using multiple resolutions is combined.
- a method of searching a region corresponding to an image from an input image is adopted.
- an image group having a plurality of resolutions corresponding to the input image is acquired, and a template matching process is performed in order from the image having the lowest resolution to search for a region corresponding to the template image.
- the calculation time is short, but the accuracy of the searched position is relatively low (accuracy of the corresponding resolution is limited).
- template matching processing for an image with higher resolution is executed. At this time, the processing is efficiently performed using the result of the template matching processing executed previously. Therefore, even if the resolution is high, the calculation time can be shortened. Thereafter, by performing the same processing, the position information searched in the image with the highest resolution is output as a result.
- an image obtained by reducing the first layer image 210 to an image size of 400 ⁇ 300 pixels is generated as an image of the second layer (hereinafter also referred to as “second layer image 220”). Furthermore, an image obtained by reducing the second layer image 220 to an image size of 200 ⁇ 150 pixels is generated as a third layer image (hereinafter also referred to as “third layer image 230”).
- FIGS. 4 and 5 are diagrams showing an example of input images and hierarchical images used in the template matching process according to the first embodiment of the present invention.
- hierarchical images with different resolutions are generated from the input image 200.
- each hierarchical image (first hierarchical image 210, second hierarchical image 220, and third hierarchical image 230) is obtained by reducing the input image 200 by a predetermined magnification.
- a search range 232 is set for the third hierarchy image 230 and corresponds to a region (hereinafter also referred to as “extraction region”) sequentially extracted from the search range 232.
- extraction region a region sequentially extracted from the search range 232.
- Template matching processing is executed with the template image.
- the extraction area has the same image size as the template image. It is assumed that the position 234 of the extraction region with the highest similarity is specified by this template matching process.
- a search range 222 is set for the second hierarchical image 220 with reference to the position 234 of the specified extraction region, and a template is extracted between the extraction region sequentially extracted from the search range 222 and the corresponding template image.
- a matching process is executed. It is assumed that the position 224 of the extraction region with the highest similarity is specified by this template matching process.
- a search range 212 is set for the first hierarchical image 210 with reference to the position 224 of the extraction region, and template matching is performed between the extraction region sequentially extracted from the search range 212 and the corresponding template image. Processing is executed. It is assumed that the position 214 of the extraction region with the highest similarity is specified by this template matching process.
- a search range 202 is set for the input image 200 with reference to the position 214 of the extraction region, and template matching between the extraction region sequentially extracted from the search range 202 and the corresponding template image is performed. Processing is executed. It is assumed that the position 204 of the extraction region with the highest similarity is specified by this template matching process. Then, the position 204 of this extraction area is output as a search result.
- FIG. 6 is a diagram showing an example of input images, hierarchical images, and template images used in the template matching process according to the first embodiment of the present invention.
- search processing is executed using a template image having a common size (for example, 128 ⁇ 128 pix) for the input image and each hierarchical image. That is, each template image is sequentially compared with an extraction area extracted from the target input image or any one of the hierarchical images.
- the matching process is executed after the template image is appropriately corrected in accordance with the relative shift between the input image / hierarchical image (partial image included) and the template image.
- the processing is made efficient by using the search result in the previous hierarchy.
- FIG. 7 is a diagram schematically showing the contents of the template matching process according to the first embodiment of the present invention.
- FIG. 7 illustrates the contents of the processing for the third layer image and the second layer image, but the same processing is executed for the first layer image and the input image by the same procedure.
- a search range is set for the third layer image, and an extraction region to be subjected to a matching process is extracted from the search range.
- the search process ((1) matching) for the rotation amount is executed between Note that the search range in the third layer image may be the entire third layer image.
- a relative rotation amount relative shift amount
- the template image is corrected using the calculated relative rotation amount ((3) rotation amount correction). With this correction, a correction template image is generated.
- a position search process ((4) matching) is performed between the extraction region and the correction template image.
- the similarity between the extraction region and the correction template image in the third hierarchy is calculated.
- the extraction region having the highest similarity with the correction template image is specified.
- a position indicating the extraction region having the highest similarity (a parallel movement amount indicating a relative position of the extraction region most matching the correction template image within the search range) is calculated ((5) position calculation). This completes the search process in the third hierarchy.
- the search process in the second layer is started.
- a search range is set for the second hierarchy image, and an extraction region to be subjected to matching processing is set from the search range, and between this extraction region and the template image for the second hierarchy is set.
- Search processing ((6) matching) for the rotation amount is executed.
- the search range is set based on the position indicating the extraction region with the highest similarity calculated in the third hierarchy.
- the search process for the rotation amount is only for a predetermined range based on the rotation amount for the extraction region having the highest similarity calculated in the third hierarchy. For example, if the rotation amount calculated in the third layer is ⁇ , in the search process for the rotation amount in the second layer, the search is performed only in the range of ⁇ ⁇ ⁇ ( ⁇ : a predetermined variation angle).
- the relative rotation amount (relative shift amount) between the extraction region and the template image in the second layer is calculated by the search process for the rotation amount ((7) rotation amount calculation). Then, the template image is corrected using the calculated relative rotation amount ((8) rotation amount correction). With this correction, a correction template image is generated.
- a position search process ((9) matching) is performed between the extraction region and the correction template image.
- the similarity between the extraction region and the correction template image in the second hierarchy is calculated.
- the extraction region having the highest similarity with the correction template image is specified.
- a position indicating the extraction region having the highest similarity (a parallel movement amount indicating the relative position of the extraction region that most closely matches the correction template image within the search range) is calculated ((10) position calculation). This completes the search process in the second hierarchy.
- the image processing apparatus 100 searches for an area corresponding to the template generation image (reference image) from the input image by performing the following functions. That is, the image processing apparatus 100 has a third layer resolution (first resolution) and a second layer resolution (second resolution higher than the first resolution) corresponding to the input image, respectively. It has a function (input image acquisition unit) for acquiring two-layer images (first and second input images). The image processing apparatus 100 further includes, from the template generation image (reference image), template images corresponding to the third and second layers (first and second references having resolutions respectively corresponding to the first and second resolutions). A function (reference image generation unit).
- the image processing apparatus 100 is extracted from the third hierarchy image by a search process between the third hierarchy image (first input image) and the template image (first reference image) corresponding to the third hierarchy. And calculating the relative shift amount of the rotation and / or magnification between the extracted region and the template image (first input image and first reference image) corresponding to the third hierarchy, and the calculated relative
- a corrected template image is generated, and the third layer A function (first corresponding position) for determining a corresponding position in the third layer (first resolution) by a search process between an image (first input image) and a correction template image (first correction reference image) Having a tough).
- the image processing apparatus 100 determines the above function (first corresponding position determination) with reference to the position in the second layer image (second input image) corresponding to the corresponding position in the third layer (first resolution). 2) by performing a search process between the second hierarchical image (second input image) and the template image (second reference image) using the relative shift amount calculated by A function (second corresponding position determining unit) for determining a corresponding position in the hierarchy (second resolution); That is, the second corresponding position determination unit performs a search process between the second input image and the second reference image, thereby rotating between the second input image and the second reference image.
- the second correction reference image is generated by correcting the second reference image based on the calculated relative displacement amount, and the second input image and the second input image The corresponding position at the second resolution is determined by the search process with the correction reference image.
- the image processing apparatus 100 sets the relative shift amount between the extraction region extracted from the third layer image and the template image (first input image and first reference image) corresponding to the third layer.
- the search range is limited accordingly, and a search process between the second hierarchical image (second input image) and the template image (second reference image) is performed, so that the second hierarchical image (second A relative shift amount of rotation and / or magnification between the input image) and the template image (second reference image) is calculated, and a template image (second reference) based on the calculated relative shift amount is calculated.
- a correction template image (second correction reference image) is generated by correcting the second reference image, and the second layer image (second input image) and the correction template image (second correction reference image) are Through the search process between Ability to determine the corresponding position in the serial second resolution having (second correcting unit).
- the second corresponding position determination unit causes the reference image acquisition unit to generate the second reference image according to the relative shift amount between the first input image and the first reference image.
- the image processing apparatus 100 uses the result of the template matching process performed in the previous layer to improve the processing efficiency.
- FIG. 8 is a block diagram showing a functional configuration of image processing apparatus 100 according to the first embodiment of the present invention.
- image processing apparatus 100 according to the present embodiment has an image acquisition unit 150, a multi-resolution image generation unit 152, a template generation unit 154, a search control unit 156, as its main functional configuration.
- a position detector 158 are realized by the CPU 102 executing the template matching processing program 112 in the image processing apparatus 100 shown in FIG. Details of each functional configuration will be described below.
- the image acquisition unit 150 acquires an input image to be processed. Typically, an image generated by the connected camera 4 capturing a subject is acquired as an input image. Alternatively, an input image generated by some external device may be acquired via various recording media or communication media.
- Multi-resolution image generation unit 152 acquires an image (hierarchical image) having a plurality of resolutions corresponding to the input image. Typically, the multi-resolution image generation unit 152 generates a plurality of images (hierarchical images) having different resolutions corresponding to the input image acquired by the image acquisition unit 150. As shown in FIGS. 4 and 5, when the image size of the input image is 1600 ⁇ 1200 pix, the conversion magnification is 1 ⁇ 2, and the number of hierarchies is 4, 800 ⁇ 600 pix and 400 ⁇ corresponding to the input image, respectively. An image group (first hierarchical image, second hierarchical image, and third hierarchical image) having an image size of 300 pix and 200 ⁇ 150 pix is generated.
- the point (x, y) of the third layer image is the point (2x, 2y), (2x + 1, 2y), (2x, 2) of the second layer image.
- 2y + 1) and (2x + 1, 2y + 1) is obtained as an average value of 2 ⁇ 2 pixels.
- the conversion magnification may be determined flexibly as necessary. For example, a smaller value such as 1/3 instead of 1/2 may be adopted as the conversion magnification. In this case, since the number of hierarchies can be reduced, the processing can be further speeded up. Conversely, a larger value such as 1 / 1.5 may be adopted as the conversion magnification. In this case, the number of hierarchies can be increased and the calculation load can be increased. However, since the search can be performed in more detail, the robustness can be improved.
- Template generation unit 154 sets a template image for each hierarchical image. Typically, the template generation unit 154 generates a template image corresponding to each hierarchical image (resolution) from the input template generation image 114.
- FIG. 9 is a diagram schematically illustrating a template image generation procedure according to the first embodiment of the present invention. As shown in FIG. 9, template images of the same size corresponding to a plurality of template generation images corresponding to the respective hierarchical images are generated.
- the user captures a reference subject to set a template image and acquires a template generation image. Then, the user designates a position to be searched for the acquired template generation image.
- the template generation unit 154 sets each template image so that the position to be searched becomes the barycentric position. In the example illustrated in FIG. 9, the template generation unit 154 sets template images each having an image size of 128 ⁇ 128 pix. Similarly, corresponding template images are set for other hierarchical images. In the example shown in FIG. 9, the template images corresponding to the other hierarchical images are also set to 128 ⁇ 128 pix having the same image size as the template image set first. In addition, the template image corresponding to each hierarchical image is also set so that the searched position becomes the center of gravity position.
- FIG. 9 shows an example in which template images having the same image size are set between hierarchical images, but at least as the resolution of the hierarchical images becomes lower (rougher), the regions included in the template images are the same or wider. What is necessary is just to set.
- the template image may be set at a position away from the searched position by a certain amount ( ⁇ x, ⁇ y).
- the position to be searched can be specified by shifting the result by a certain amount ( ⁇ x, ⁇ y) with respect to the position actually searched using the template image.
- ⁇ c7 Search Control Unit 156
- the search control unit 156 efficiently performs a search process using a coarse / fine strategy using multiple resolutions.
- a procedure for performing a position search on the input image will be described.
- the procedure of multi-resolution analysis when the image size of the input image is 1600 ⁇ 1200 pix, the conversion magnification is 1/2, and the number of layers is 4 will be described.
- a position search is executed using a template image corresponding to the third layer image.
- the corresponding position (x3, y3) is specified.
- an area having the same image size as the template image corresponding to the third hierarchy image is extracted from the third hierarchy image, and the contents of the extracted area (extraction area) are input to the position detection unit 158.
- the position is searched.
- a region to be subjected to template matching processing is extracted from the third layer image (200 ⁇ 150 pix).
- a region having the same image size as the template image (128 ⁇ 128 pix) corresponding to the third layer image is cut out so that the centroid position of the extraction region matches the centroid position of the third layer image.
- a template matching process is performed on the extraction region and the template image, and the rotation amount and position are specified.
- the position (x3, y3) on the third layer image is estimated in the second layer image.
- the approximate position on the second layer image is next determined using the result obtained by the position search on the third layer image, and then the position is determined. Perform a search.
- the coordinates obtained by doubling the position specified in the third layer image are set as the initial position in the second layer image.
- region used as the object of a template matching process between the template images corresponding to a 2nd hierarchy image is extracted from a 2nd hierarchy image.
- This extraction procedure is the same as the extraction procedure for the third layer image.
- a template matching process is performed on the extraction region and the template image extracted from the second resolution image, and the rotation amount and position are specified.
- the approximate position on the image in the first layer image is obtained next using the result of the position search in the second layer image.
- the result on the low resolution side is gradually brought closer to the correct answer so as to be the initial position on the high resolution side.
- the final position of the first layer image and the input image is specified by specifying the position in the same manner.
- the position detection unit 158 searches for the rotation amount and the position using the template image corresponding to the extraction area extracted from the input image and / or the hierarchical image.
- RIPOC rotation invariant phase only correlation
- FIG. 10 is a schematic diagram for illustrating processing contents in position detection unit 158 according to the first embodiment of the present invention.
- FIG. 11 is a schematic diagram for explaining the processing contents executed in the POC shown in FIG.
- the RIPOC method calculates how much the extraction area is rotated with respect to the template image, corrects the rotation angle of the template image (reference numeral 167), and corrects the rotation angle. And a process (reference numeral 168) for calculating how much similarity and parallel movement amount exist between the correction template image acquired by the above and the extraction region.
- the position detection unit 158 includes Fourier transform processes 161 and 171, logarithmic processes 162 and 172, polar coordinate transform processes 163 and 173, POC process 164, rotation amount correction process 165, and POC process. 166.
- Fourier transform processing 161 and 171 calculate frequency components (amplitude component and phase component) included in the template image and the extraction region, respectively.
- the phase component is not necessarily required, and thus may not be calculated.
- Logarithmic processing 162 and polar coordinate conversion processing 163 logarithmize the amplitude component of the template image and convert it to polar coordinates.
- the logarithmic process 172 and the polar coordinate conversion process 173 logarithmize the amplitude component for the extraction region and convert it to polar coordinates.
- the rotation amount is expressed as a coordinate point on two-dimensional coordinates.
- the POC process 164 calculates the similarity and the parallel movement amount (corresponding to the rotation amount) for the results of the polar coordinate conversion output from the polar coordinate conversion processes 163 and 173, respectively.
- FIG. 11 schematically shows a processing result in the POC processing 164.
- the amount of parallel movement between two images is represented as a peak position.
- the images are sequentially shifted, and the correlation value is calculated between the components of the spatial frequency included in each image, thereby searching for the one with the highest similarity.
- the amount of translation between images can be calculated by searching a range of about ⁇ 1/4 of the image size of the template image. That is, in the POC process 164, the position where the similarity is the highest among the results of the polar coordinate conversion is specified, and the corresponding rotation amount is output to the rotation amount correction process 165.
- the rotation amount correction processing 165 corrects the rotation of the template image according to the rotation amount calculated in the POC processing 164. That is, in the rotation amount correction process 165, the template image is rotationally corrected to generate a corrected template image.
- a method of rotating the template image itself in real space can be employed.
- the frequency space is handled in the POC processing 166, it is not necessary to rotate in the real space, and the data (amplitude information and phase information) after Fourier transform, which is an internal representation of the POC processing, is used. You may employ
- the POC process 166 calculates the similarity and the amount of parallel movement between the corrected template image and the extraction area. The position with the highest similarity indicates a region that matches the corrected template image included in the input image.
- the efficiency of the rotation amount correction process 165 by the coarse / dense strategy using multi-resolution will be described.
- the amount of rotation of the object 3 included in the input image is indefinite, so it is necessary to search for the amount of rotation within a maximum range of 360 °.
- the angle that can be searched by the RIPOC method is ⁇ 90 °
- the rotation amount ( ⁇ + ⁇ ) is also a calculation candidate. Therefore, the POC process 164 and the rotation amount correction process 165 shown in FIG.
- the POC process 166 performs two rotations.
- POC processing position search
- the height of the peak position by POC process about each template image is compared, and the one where the height of a peak position is higher is output as a translation amount.
- the rotation correction amount of the template used when calculating the selected peak position is output as the rotation amount in the third layer.
- FIG. 12 is a flowchart showing an overall procedure of template matching processing according to the first embodiment of the present invention. Each step shown in FIG. 12 is typically realized by the CPU 102 (FIG. 2) executing the template matching processing program 112.
- the CPU 102 acquires an input image from the camera 4 or the like (step S100). Subsequently, the CPU 102 multi-resolutions the input image to generate a hierarchical image (step S102).
- the first layer image, the second layer image, and the third layer image corresponding to the input image have been generated. It is assumed that a template image corresponding to the input image and each hierarchical image is prepared in advance.
- the CPU 102 sets an initial position for performing a position search for the third layer image (step S104).
- the CPU 102 extracts an area having the same image size as the template image from the third hierarchy image (step S106). This extraction area is set so that the initial position set in step S104 is the center of gravity position.
- the CPU 102 receives the extraction area and the template image as input and executes a rotation amount search process (step S108), and generates a correction template image based on the rotation amount obtained by the rotation amount search process (step S108).
- Step S110 Typically, a corrected template image is generated by correcting the template image.
- the CPU 102 executes position search processing using the correction template image generated in step S110 and the extraction region as inputs (step S112).
- processing for a higher hierarchy is executed.
- the CPU 102 sets an initial position for performing a position search on the target hierarchical image or input image based on the position calculated in the lower hierarchy (step S114). Subsequently, the CPU 102 extracts an area having the same image size as the corresponding template image from the target hierarchical image or input image (step S116).
- the CPU 102 sets a search range for the rotation amount based on the rotation amount for template image correction calculated in the lower hierarchy (step S118), and inputs the extraction area and the template image to input the rotation amount.
- Search processing is executed (step S120), and a correction template image is generated based on the rotation amount obtained by the rotation amount search processing (step S122).
- a corrected template image is generated by correcting the template image.
- the CPU 102 executes position search processing using the correction template image generated in step S122 and the extraction region as inputs (step S124).
- step S126 determines whether or not the current target image is an input image. If the current target image is not an input image (NO in step S126), the next higher-level image is set as the target (step S128), and the processes in and after step S114 are repeated.
- step S126 if the current target image is an input image (YES in step S126), the result of the search process of step S124 executed immediately before is output (step S130). Then, the process ends.
- the search is performed while narrowing down both the relative shift amount and the position from the low resolution side, so that the amount of calculation can be reduced and the accuracy can be reduced. Robustness can also be improved.
- ⁇ d2 Efficiency of search processing
- search processing is performed in which a relative shift is corrected between the hierarchical image (partial image included) and the template image on the low resolution side.
- the rotation amount used when the position with the highest similarity is calculated is used to generate a template image in the next layer.
- FIG. 13 is a diagram schematically showing the contents of the template matching process according to the second embodiment of the present invention.
- FIG. 13 illustrates the contents of the processing for the third layer image and the second layer image, but the same processing is executed for the first layer image and the input image by the same procedure.
- a search range is set for the third hierarchy image, and an extraction region to be subjected to matching processing is extracted from the search range, and this extraction region and the template image for the third hierarchy are extracted.
- the search process ((1) matching) for the rotation amount is executed between Note that the search range in the third layer image may be the entire third layer image.
- a relative rotation amount relative shift amount
- the template image is corrected using the calculated relative rotation amount ((3) rotation amount correction). With this correction, a correction template image is generated.
- a position search process ((4) matching) is performed between the extraction region and the correction template image.
- the similarity between the extraction region and the correction template image in the third hierarchy is calculated.
- the extraction region having the highest similarity with the correction template image is specified.
- a position indicating the extraction region having the highest similarity (a parallel movement amount indicating a relative position of the extraction region most matching the correction template image within the search range) is calculated ((5) position calculation). This completes the search process in the third hierarchy.
- the search process in the second layer is started.
- the template image is dynamically generated (or recreated) in consideration of the rotation amount when the highest similarity is shown in the third hierarchy ((6) rotation amount correction). Thereby, the degree of matching between the template image and the extraction region can be increased in advance.
- a template image corresponding to the second layer is generated according to the correction amount (rotation amount) used for generating the correction template image corresponding to the third layer.
- an initial position is set for the second hierarchy image based on the calculated parallel movement amount, and an extraction area to be subjected to matching processing is set based on the initial position, and the extraction area and the second hierarchy are set.
- a search process ((7) matching) for the rotation amount is executed with the template image for use.
- the search range of the search process for the rotation amount is set to ⁇ ⁇ ( ⁇ : a predetermined fluctuation angle), for example. In the present embodiment, since the template image considering the approximate amount of rotation is generated first, the search range can be limited.
- the relative rotation amount (relative shift amount) between the extraction region and the template image in the second hierarchy is calculated by the search process for the rotation amount ((8) rotation amount calculation). Then, the template image is corrected using the calculated relative rotation amount ((9) rotation amount correction). With this correction, a correction template image is generated.
- a position search process ((10) matching) is executed between the extraction region and the correction template image.
- the similarity between the extraction region and the correction template image in the second hierarchy is calculated.
- the extraction region having the highest similarity with the correction template image is specified.
- a position indicating the extraction region having the highest similarity (a parallel movement amount indicating the relative position of the extraction region most matching the correction template image in the search range) is calculated ((11) position calculation). This completes the search process in the second hierarchy.
- FIG. 14 is a diagram schematically illustrating a template image generation procedure according to the second embodiment of the present invention.
- a template image corresponding to each layer is generated by cutting out from a template generation image.
- a template image corresponding to the lowest hierarchy is generated by cutting out from a template generation image ((1) cutting out). This clipping is arbitrarily set by the user. Corresponding positions are searched from matching target images (hierarchical images) using the template image generated in the lowest hierarchy ((2) matching). At this time, a relative shift amount (in this example, a rotation amount) from the template image is calculated.
- the template image is cut out from the template generation image using the searched matching information (rotation amount) ((3) matching information (rotation amount)) ((4) cut out).
- a template image corresponding to a state that is substantially close to the amount of rotation of the object shown in the matching target image is generated.
- a corresponding position is searched from the matching target image (hierarchical image) ((5) matching). Thereafter, the corresponding position is searched similarly.
- a template image on the high resolution side is generated using the rotation amount on the low resolution side.
- the image processing apparatus 100 searches the area corresponding to the template generation image (reference image) from the input image by performing the following functions. That is, the image processing apparatus 100 has a third layer resolution (first resolution) and a second layer resolution (second resolution higher than the first resolution) corresponding to the input image, respectively. It has a function (input image acquisition unit) for acquiring two-layer images (first and second input images). The image processing apparatus 100 further includes, from the template generation image (reference image), template images corresponding to the third and second layers (first and second references having resolutions respectively corresponding to the first and second resolutions). A function (reference image generation unit).
- the image processing apparatus 100 is extracted from the third hierarchy image by a search process between the third hierarchy image (first input image) and the template image (first reference image) corresponding to the third hierarchy. And calculating a relative shift amount between the extracted region and the template image corresponding to the third hierarchy (the first input image and the first reference image), and based on the calculated relative shift amount. It has a function (first correction unit) for generating a correction template image (first correction reference image) by correcting the template image (first reference image). The image processing apparatus 100 further performs corresponding processing in the third hierarchy (first resolution) by a search process between the third hierarchy image (first input image) and the correction template image (first correction reference image). (First corresponding position determining unit).
- the image processing apparatus 100 uses the position in the second hierarchy (second resolution) corresponding to the corresponding position in the third hierarchy (first resolution) as a reference for the second hierarchy image (second input image).
- second resolution position in the second hierarchy
- first resolution position in the third hierarchy
- second input image second input image
- the image processing apparatus 100 generates a second template image (second reference image) according to the rotation amount (correction amount) used for generating the first correction template image (first correction reference image). Generate.
- the image processing apparatus 100 uses the rotation amount (correction amount) used for generating the second layer image (second input image) and the correction template image (first correction reference image) corresponding to the third layer.
- a correction template image (second correction reference image) is generated by performing a search process with the template image (second reference image) corresponding to the second hierarchy generated accordingly.
- the image processing apparatus 100 uses the result of the template matching process performed in the previous layer to improve the processing efficiency.
- FIG. 15 is a flowchart showing an overall procedure of template matching processing according to the second embodiment of the present invention. Each step shown in FIG. 15 is typically realized by the CPU 102 (FIG. 2) executing the template matching processing program 112.
- step S119 is executed instead of the processing in step S118 as compared with the processing procedure shown in FIG. That is, after setting an initial position for performing a position search for the target hierarchical image or input image based on the position calculated in the lower hierarchy (step S114), the CPU 102 performs the target hierarchical image or input image. A region corresponding to the image and having the same image size as the template image is extracted (step S116).
- the CPU 102 generates a template image corresponding to the template generation image based on the rotation amount for template image correction calculated in the lower hierarchy. (Step S119). Further, the CPU 102 receives the extracted area and the generated template image as input and executes a rotation amount search process (step S120), and generates a correction template image based on the rotation amount obtained by the rotation amount search process. (Step S122). Typically, a corrected template image is generated by correcting the template image. Further, the CPU 102 executes position search processing using the correction template image generated in step S122 and the extraction region as inputs (step S124).
- a fine template image such as 1 °
- a rougher template image such as 10 ° is created, and the closest one is selected. Also good. Even with such processing, a certain accuracy improvement effect can be obtained.
- a template image having almost no relative shift amount (relative rotation amount and relative magnification) can be used in the high-resolution hierarchical image, so that the search accuracy can be improved.
- FIG. 16 is a diagram schematically showing the contents of the template matching process according to the third embodiment of the present invention.
- a search range is set for the third hierarchy image, and an extraction region to be subjected to matching processing is extracted from the search range, and this extraction region and the template image for the third hierarchy are extracted.
- the search process ((1) matching) for the rotation amount is executed between By the search process for the rotation amount, a relative rotation amount (relative shift amount) between the extraction region and the template image in the third hierarchy is calculated ((2) rotation amount calculation).
- the template image is corrected using the calculated relative rotation amount ((3) rotation amount correction). With this correction, a correction template image is generated.
- a position search process ((4) matching) is performed between the extraction region and the correction template image.
- the similarity between the extraction region and the correction template image in the third hierarchy is calculated.
- the extraction region having the highest similarity with the correction template image is specified.
- a position indicating the extraction region having the highest similarity (a parallel movement amount indicating a relative position of the extraction region most matching the correction template image within the search range) is calculated ((5) position calculation). This completes the search process in the third hierarchy.
- the search process in the second layer is started.
- the template image is dynamically generated (or recreated) in consideration of the rotation amount when the highest similarity is shown in the third hierarchy ((6) rotation amount correction). That is, in the template matching process in the second hierarchy, a template image corresponding to the second hierarchy is generated according to the correction amount (rotation amount) used for generating the correction template image corresponding to the third hierarchy.
- an initial position is set for the second layer image based on the calculated parallel movement amount, and an extraction region to be subjected to matching processing is set based on the initial position.
- the search process ((7) matching) about a position is performed between an extraction area
- the search process for this position the similarity between the extraction region and the template image in the second hierarchy is calculated.
- the extraction region having the highest similarity with the template image is specified.
- a position indicating the extraction region having the highest similarity (a parallel movement amount indicating a relative position of the extraction region most matching the template image in the search range) is calculated ((8) position calculation). This completes the search process in the second hierarchy.
- the search process in the first hierarchy is started.
- the template image is dynamically generated (or recreated) in consideration of the rotation amount when the highest similarity is shown in the third hierarchy ((9) rotation amount correction).
- an initial position is set for the first layer image based on the parallel movement amount calculated in the second layer, and an extraction region to be subjected to matching processing is set based on the initial position.
- the search process ((10) matching) about a position is performed between an extraction area
- the extraction region having the highest similarity with the template image is specified.
- a position indicating the extraction region having the highest similarity is calculated ((11) position calculation). This completes the search process in the first hierarchy.
- the same search process is executed for the input image.
- the image processing apparatus 100 is generated according to the rotation amount (correction amount) used for generating the correction template image (first correction reference image) corresponding to the third hierarchy.
- the rotation amount corrected amount
- the image processing apparatus 100 is generated according to the rotation amount (correction amount) used for generating the correction template image (first correction reference image) corresponding to the third hierarchy.
- ⁇ E3 Functional configuration>
- the functional configuration of the image processing apparatus and / or image processing program according to the present embodiment is the same as that of the block diagram shown in FIG.
- the template image generation processing by the template generation unit 154 is different from the processing in the first embodiment as described above.
- FIG. 17 is a flowchart showing an overall procedure of template matching processing according to the third embodiment of the present invention.
- the processes in steps S ⁇ b> 120 and S ⁇ b> 122 are omitted as compared with the processing procedure illustrated in FIG. 15.
- the other processing is as described with reference to FIG. 15, and therefore detailed description will not be repeated.
- the “matching degree” is an index indicating how similar the input image or the hierarchical image is to the template image, and a correlation value or the like is used. As shown in FIG. 11, in POC, the magnitude (peak value) of a correlation peak obtained after inverse Fourier transform can be used.
- the degree of parallel movement is performed by searching for the similarity (processing result by the POC process 164 in FIG. 10) and the position acquired when calculating the rotation amount for correcting the rotation of the template image.
- Similarity processing result by the POC processing 166 in FIG. 10 acquired when calculating.
- rotational similarity rotational similarity
- positional deviation similarity positional deviation similarity
- Pattern example Examples of patterns that optimize the matching method based on the similarity on the low resolution side include the following.
- ⁇ Pattern 3> Low resolution side: rotation search + rotation amount correction + position search ⁇ positional deviation similarity: high ⁇ high resolution side: template image generation based on rotation amount on low resolution side + rotation search (rotation amount on low resolution side) (2) Low resolution side: rotation search + rotation amount correction + position search ⁇ positional deviation similarity: low ⁇ high resolution side: rotation amount on low resolution side Template image generation based on + rotation search (no search range restriction based on low-resolution rotation amount) + rotation amount correction + position search ⁇ Pattern 4> (1) Low resolution side: rotation search + rotation amount correction + position search ⁇ rotation similarity: high ⁇ high resolution side: template image generation based on the rotation amount on the low resolution side + rotation search (to the rotation amount on the low resolution side) (2) Low resolution side: rotation search + rotation amount correction + position search ⁇ rotation similarity: low ⁇ high resolution side: template based on rotation amount on low resolution side Image generation + rotation search (no limitation of search range based on rotation amount on the
- image processing apparatus 100 performs a search for the similarity (rotation amount similarity) calculated in the template image rotation amount correction on the low resolution side and the position on the low resolution side. Based on at least one of the calculated similarities (positional deviation similarity), it includes a function (control unit) that varies the procedure for generating a template image or a correction template image used for search processing on the high resolution side. That is, the control unit of the image processing apparatus 100 calculates the similarity calculated in the search process between the first input image and the first reference image, and the first input image and the first correction reference image. Based on at least one of the similarities calculated in the search process between the two, the second corresponding position determining unit changes the method for determining the corresponding position at the second resolution. More specifically, the image processing apparatus 100 changes the generation procedure based on the similarity calculated when determining at least one of the relative shift amount and the positional shift amount.
- the RIPOC method is used as a method for calculating and correcting a relative shift (geometric shift) between the input image / hierarchical image (partial image included) and the template image.
- a relative shift geometric shift
- various algorithm methods for calculating and correcting a geometric shift including a magnification (size) can also be adopted.
- a matching method using Fourier Merin transform may be adopted.
- the magnification (size) search process can be performed in the same manner as the rotation amount search process described above.
- reference 1 Hidet al., "Fusion of image signal processing and image pattern recognition-DCT code limited correlation and its application", Tokyo Metropolitan University, Faculty of System Design The method described in Dynamic Image Processing Actual Utilization Workshop 2007 (2007.3.8)
- Hishi Hidet al.
- Embodiments of the present invention include the following aspects.
- An image processing apparatus is directed to a template matching method that takes into account the relative rotation amount and / or relative magnification between a template image and an acquired image.
- the image processing apparatus includes: an image acquisition unit that acquires an image; a unit that acquires a template image; a unit that searches for a corresponding position from the acquired image using a correlation between a partial region of the acquired image and the template image; At the time of search, a relative deviation amount calculation unit that calculates a relative rotation amount and / or a magnification with respect to the template image, and a value calculated by the relative deviation amount calculation unit are used to obtain the acquired image and the template image. And a unit for calculating a corresponding position after correcting the relative shift amount.
- the image processing apparatus 100 further includes a resolution conversion unit that converts the image acquired by the image acquisition unit into a plurality of resolutions, a unit that acquires a template image for each resolution, and a low-resolution image using a template image. Correspondence in the acquired image by specifying the provisional position in the resolution image, the part that sets the provisional position in the high resolution from the provisional position obtained at each resolution, and the provisional position in order from the low resolution to the high resolution A part for determining the position and a part for limiting a search range of the relative shift amount at the high resolution based on the relative shift amount obtained at the low resolution are included.
- the search is performed by narrowing down both the relative shift amount and the position from the low resolution side by multiplexing the images, the calculation speed can be reduced and the robustness of the accuracy can be improved. .
- An image processing apparatus is directed to a template matching method that takes into account the relative rotation amount and / or relative magnification between a template image and an acquired image.
- the image processing apparatus includes: an image acquisition unit that acquires an image; a unit that acquires a template image; a unit that searches for a corresponding position from the acquired image using a correlation between a partial region of the acquired image and the template image; At the time of search, a relative deviation amount calculation unit that calculates a relative rotation amount and / or a magnification with respect to the template image, and a value calculated by the relative deviation amount calculation unit are used to obtain the acquired image and the template image. And a unit for calculating a corresponding position after correcting the relative shift amount.
- the image processing apparatus 100 further includes a resolution conversion unit that converts the image acquired by the image acquisition unit into a plurality of resolutions, a unit that acquires a template image for each resolution, and a low-resolution image using a template image. Correspondence in the acquired image by specifying the provisional position in the resolution image, the part that sets the provisional position in the high resolution from the provisional position obtained at each resolution, and the provisional position in order from the low resolution to the high resolution A part for determining a position, and a part for using a template image used for template matching at a high resolution as an image area that is shifted so as to be comparable to the relative shift amount obtained at a low resolution. Including.
- the template with the relative displacement amount of the rotation and the magnification is used in the high-resolution hierarchical image, so that the accuracy can be improved.
- the image processing apparatus calculates only a positional shift amount after correcting a relative shift amount between an image and a template at a high resolution based on a relative shift amount obtained at a low resolution. calculate.
- the high-resolution hierarchical image does not perform the calculation for obtaining the relative shift amount of the rotation and the magnification, so that high-speed processing can be realized.
- the image processing device limits the calculation range when calculating the relative rotation amount and / or the relative magnification at the time of matching at the high resolution based on the similarity calculated in the template matching processing at the low resolution. Control the amount to do.
- the similarity is a similarity obtained when calculating any one of a relative rotation amount, a relative magnification, and a positional deviation amount.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Data Mining & Analysis (AREA)
- Signal Processing (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
Description
第1の入力画像と第1の基準画像との間の探索処理によって、第1の入力画像と第1の基準画像との間の回転および/または倍率の相対的なずれ量を算出するとともに、算出された相対的なずれ量に基づく第1の基準画像の補正によって第1の補正基準画像を生成し、第1の入力画像と第1の補正基準画像との間の探索処理によって、第1の解像度における対応位置を決定するステップと、
第1の解像度における対応位置に相当する第2の入力画像における位置を基準として、相対的なずれ量を利用して、第2の入力画像と第2の基準画像との間の探索処理を行なうことで、第2の解像度における対応位置を決定するステップとを備えた画像処理方法。
本実施の形態は、基準画像に対応する領域を入力画像から探索する画像処理装置および画像処理方法に向けられている。このような基準画像に対応する領域を入力画像から探索する画像処理方法の典型例として、以下では、テンプレートマッチング処理について説明する。このテンプレートマッチング処理では、マッチング処理における基準画像としてテンプレート画像が用いられる。
まず、本発明の実施の形態に従うテンプレートマッチング処理を実現する画像処理装置の実装例について説明する。
図1は、本発明の実施の形態に従うテンプレートマッチングの適用例を示す模式図である。図1を参照して、本実施の形態に従うシステム1は、一例として、ベルトコンベア2を含む生産ラインに適用される。このシステム1において、ベルトコンベア2上を対象物3(ワーク)が連続的に搬送されるとともに、この対象物3をカメラ4により撮像することで、対象物3の外観を含む画像が取得される(以下、この取得される画像を「入力画像」とも称す。)。
図2は、本発明の実施の形態に従うテンプレートマッチングをパーソナルコンピューターにより実現した場合の構成を示すブロック図である。
上述したパーソナルコンピューターにより実現する例に加えて、例えば、デジタルカメラ、携帯電話、あるいはスマートフォン上に実装してもよい。さらに、少なくとも1つのサーバー装置が本実施の形態に従う処理を実現する、いわゆるクラウドサービスのような形態であってもよい。この場合、ユーザーは、自身の端末(パーソナルコンピューターやスマートフォンなど)を用いて、少なくとも2つの処理対象画像をサーバー装置(クラウド側)へ送信し、当該送信された処理対象画像に対して、サーバー装置側が本実施の形態に従う画像処理を行なうような構成が想定される。さらに、サーバー装置側がすべての機能(処理)を行なう必要はなく、ユーザー側の端末とサーバー装置とが協働して、本実施の形態に従う画像処理を実現するようにしてもよい。
次に、本実施の形態1に従うテンプレートマッチング処理の詳細について説明する。
本実施の形態においては、入力画像/階層画像(に含まれる部分画像)とテンプレート画像との間の相対的なずれを考慮した探索処理と、多重解像度を用いた粗密戦略とを組み合わせて、テンプレート画像に対応する領域を入力画像から探索する方法を採用する。
次に、本実施の形態に従うテンプレートマッチング処理における処理の効率化手法について説明する。本実施の形態においては、入力画像/階層画像(に含まれる部分画像)とテンプレート画像との間に相対的なずれを補正した探索処理が実行される。この相対的なずれ量は、入力画像/階層画像(に含まれる部分画像)とテンプレート画像の間の、相対的な回転方向のずれ量および/または相対的な倍率(サイズ)の違いによるずれ量である。以下では、相対的なずれ量として回転量を考慮する例について説明するが、これに限られず、相対的な倍率を考慮する構成についても同様の構成が適用可能である。相対的な回転量を考慮した探索処理の一例として、回転不変位相限定相関(RIPOC)法を用いた構成例について説明する。
このように、本実施の形態に従う画像処理装置100は、以下のような機能を発揮することで、テンプレート生成用画像(基準画像)に対応する領域を入力画像から探索する。すなわち、画像処理装置100は、入力画像に対応する、第3階層の解像度(第1の解像度)および第2階層の解像度(第1の解像度より高い第2の解像度)をそれぞれ有する第3および第2階層画像(第1および第2の入力画像)を取得する機能(入力画像取得部)を有する。画像処理装置100は、さらにテンプレート生成用画像(基準画像)から、第3および第2階層に対応するテンプレート画像(第1および第2の解像度にそれぞれ対応する解像度を有する第1および第2の基準画像)を生成する機能(基準画像生成部)を有する。
次に、本実施の形態に従う画像処理装置および/または画像処理プログラムの機能構成について説明する。
画像取得部150は、処理対象となる入力画像を取得する。典型的には、接続されたカメラ4が被写体を撮像することで生成される画像を入力画像として取得する。これに代えて、何らかの外部装置で生成された入力画像を各種の記録媒体または通信媒体を介して取得するようにしてもよい。
多重解像度画像生成部152は、入力画像に対応する、複数の解像度を有する画像(階層画像)を取得する。典型的には、多重解像度画像生成部152は、画像取得部150によって取得された入力画像に対応する、解像度が異なる複数の画像(階層画像)を生成する。図4および図5に示すように、入力画像の画像サイズを1600×1200pixとし、変換倍率を1/2とし、階層数を4としたときには、入力画像に対応する、それぞれ800×600pix、400×300pix、200×150pixの画像サイズを有する画像群(第1階層画像、第2階層画像、第3階層画像)が生成される。
テンプレート生成部154は、階層画像毎にテンプレート画像を設定する。典型的には、テンプレート生成部154は、入力されたテンプレート生成用画像114から、それぞれの階層画像(解像度)に応じたテンプレート画像を生成する。
探索制御部156は、図3を参照して説明したように、多重解像度を用いた粗密戦略によって効率的に探索処理を行なう。以下、入力画像に対する位置探索を行なうための手順について説明する。ここでは、上述したように、入力画像の画像サイズを1600×1200pixとし、変換倍率を1/2とし、階層数を4としたときの多重解像度解析の手順について説明する。
位置検出部158は、入力画像および/または階層画像から抽出された抽出領域と対応するテンプレート画像とを用いて、回転量および位置が探索される。本実施の形態においては、一例として、回転対応の位置検出技術である回転不変位相限定相関(RIPOC:Rotation Invariant Phase Only Correlation)法を用いる方法について説明する。
次に、本実施の形態に従うテンプレートマッチング処理の全体手順について説明する。
CPU102は、対象の階層画像または入力画像に対して、低階層において算出された位置に基づいて、位置探索を行なうための初期位置を設定する(ステップS114)。続いて、CPU102は、対象の階層画像または入力画像から対応するテンプレート画像と同じ画像サイズの領域を抽出する(ステップS116)。
本実施の形態によれば、入力画像を多重解像度化することで、低解像度側から相対的なずれ量および位置の両方を絞り込みながら探索するので、演算量を低減することがでるとともに、精度のロバスト性も向上できる。
上述の実施の形態1では、低解像度側から高解像度側へ位置および回転量の情報を伝播させていく際、回転量の探索範囲を制限した上で、予め準備されているテンプレート画像の回転補正を行なう構成について例示した。これに対して、実施の形態2においては、低解像度側で検出された回転量の情報に基づいて、テンプレート画像を動的に生成する構成について例示する。
本実施の形態に従うテンプレートマッチング処理の概要は、上述の図3~図5を参照して説明した内容と実質的に同一である。そのため、詳細な説明は繰り返さない。
次に、本実施の形態に従うテンプレートマッチング処理における処理の効率化手法について説明する。本実施の形態においては、低解像度側において階層画像(に含まれる部分画像)とテンプレート画像との間に相対的なずれを補正した探索処理が実行される。そして、この探索処理において、最も類似度が高い位置(上述の図11に示すピーク位置)が算出された際に利用された回転量が次の階層におけるテンプレート画像の生成に利用される。
次に、テンプレート画像の生成手順について説明する。
本実施の形態に従う画像処理装置および/または画像処理プログラムの機能構成については、図8に示すブロック図と同様である。但し、テンプレート生成部154によるテンプレート画像の生成処理については、上述したように、実施の形態1における処理とは異なっている。
次に、本実施の形態に従うテンプレートマッチング処理の全体手順について説明する。
《d5.変形例》
上述の説明では、予め作成されたテンプレート画像を作り直す処理について説明したが、テンプレート生成用画像から回転量を考慮して複数のテンプレート画像を作成しておき、メモリ上に保持しておいてもよい。このような方法を採用することで、低階層側で取得された回転量に応じて、最適なテンプレート画像を選択するだけでよくなる。その結果、演算時間を短縮化できる。
本実施の形態によれば、高解像度側の階層画像では相対的なずれ量(相対的な回転量、および、相対的な倍率)がほぼ存在しないテンプレート画像を使用できるので、探索精度を向上できる。
上述の実施の形態2では、低解像度側で検出された回転量の情報に基づいて、テンプレート画像を動的に生成する構成について例示した。この実施の形態2においては、さらに回転量についての探索処理が実行され、テンプレート画像が補正された上で、位置についての探索処理が実行される。これに対して、実施の形態3においては、生成されたテンプレート画像に対する補正処理を省略する構成について説明する。低解像度側で算出された回転量に基づいて動的に生成されるテンプレート画像は、入力画像に含まれる対象物3の向きに適合していると考えられるので、このようなテンプレート画像の回転補正を省略したとしても、比較的高い精度を保つことができる。
本実施の形態に従うテンプレートマッチング処理の概要は、上述の図3~図5を参照して説明した内容と実質的に同一である。そのため、詳細な説明は繰り返さない。
次に、本実施の形態に従うテンプレートマッチング処理における処理の効率化手法について説明する。
このように、本実施の形態に従う画像処理装置100は、第3階層に対応する補正テンプレート画像(第1の補正基準画像)の生成に用いた回転量(補正量)に応じて生成された第2階層に対応するテンプレート画像(第2の基準画像)に対応する領域を第2階層画像から探索することで、第2の階層画像にける対応位置を決定する。
本実施の形態に従う画像処理装置および/または画像処理プログラムの機能構成については、図8に示すブロック図と同様である。但し、テンプレート生成部154によるテンプレート画像の生成処理については、上述したように、実施の形態1における処理とは異なっている。
次に、本実施の形態に従うテンプレートマッチング処理の全体手順について説明する。
上述の説明では、第1番目の階層(最低解像度)においてのみ回転量についての探索が実行される例を示すが、他の階層においても必要に応じて回転量についての探索を行なってもよい。例えば、最高解像度(すなわち、入力画像)においても、回転量についての探索およびテンプレート画像の回転量補正を行なった上で、位置についての探索処理を行なって平行移動量を算出するようにしてもよい。これにより、よりロバストな回転量算出および平行移動量算出が可能になる。
本実施の形態によれば、高解像度側の階層画像では相対的なずれ量(相対的な回転量、および、相対的な倍率)を算出するための演算を省略できるので、処理を高速化できる。
上述の実施の形態1~3では、予め定められた手順に従って、テンプレートマッチング処理が実行される例について説明した。これに対して、実施の形態4においては、低解像度側でのテンプレートマッチング処理のマッチング度合いに応じて、高解像度側でのマッチング方法を最適化する方法について説明する。
「マッチング度合い」は、入力画像または階層画像がテンプレート画像に対してどのくらい類似しているかを示す指標であり、相関値などが用いられる。図11に示すように、POCでは、逆フーリエ変換後に得られる相関ピークの大きさ(ピーク値)を用いることができる。
このような、低解像度側における類似度に基づいてマッチング方法を最適化するパターン例としては、以下のようなものが挙げられる。
(1)低解像度側:回転探索+回転量補正+位置探索→位置ずれ類似度:高→高解像度側:低解像度側の回転量に基づくテンプレート画像の生成+位置探索
(2)低解像度側:回転探索+回転量補正+位置探索→位置ずれ類似度:低→高解像度側:低解像度側の回転量に基づくテンプレート画像の生成+回転探索+回転量補正+位置探索
<パターン2>
(1)低解像度側:回転探索+回転量補正+位置探索→回転類似度:高→高解像度側:低解像度側の回転量に基づくテンプレート画像の生成+位置探索
(2)低解像度側:回転探索+回転量補正+位置探索→回転類似度:低→高解像度側:低解像度側の回転量に基づくテンプレート画像の生成+回転探索+回転量補正+位置探索
すなわち、上述のパターン1および2においては、類似度が高い場合には、高解像度側においては、テンプレート画像に対する回転量についての探索および回転量補正を省略する。
<パターン3>
(1)低解像度側:回転探索+回転量補正+位置探索→位置ずれ類似度:高→高解像度側:低解像度側の回転量に基づくテンプレート画像の生成+回転探索(低解像度側の回転量に基づく探索範囲の制限有)+回転量補正+位置探索
(2)低解像度側:回転探索+回転量補正+位置探索→位置ずれ類似度:低→高解像度側:低解像度側の回転量に基づくテンプレート画像の生成+回転探索(低解像度側の回転量に基づく探索範囲の制限無)+回転量補正+位置探索
<パターン4>
(1)低解像度側:回転探索+回転量補正+位置探索→回転類似度:高→高解像度側:低解像度側の回転量に基づくテンプレート画像の生成+回転探索(低解像度側の回転量に基づく探索範囲の制限有)+回転量補正+位置探索
(2)低解像度側:回転探索+回転量補正+位置探索→回転類似度:低→高解像度側:低解像度側の回転量に基づくテンプレート画像の生成+回転探索(低解像度側の回転量に基づく探索範囲の制限無)+回転量補正+位置探索
すなわち、上述のパターン3および4においては、類似度が高い場合には、高解像度側においては、テンプレート画像に対する回転量についての探索において探索範囲の制限を有効化し、そうでなければ、探索範囲を制限しない。
上述の説明においては、回転量類似度および位置ずれ類似度について言及したが、それ以外の類似度、例えば倍率類似度などについても同様に採用することができる。この場合には、例えば、3つの類似度の重み付け平均値などに依存して、高解像度側でのマッチング方法を決定してもよい。
低解像度側におけるマッチング度合いに応じて高解像度側でのマッチング方法を最適化することで、精度の安定性および演算時間の短縮化を両立できる。例えば、低解像度側で確からしいマッチングができている場合には、高解像度側での回転量についての探索を行なわなくとも、高い精度で平行移動量を算出できると考えられるので、回転量の探索処理を省略できる。これによって、トータルの演算量を低減できる。
上述の実施の形態においては、入力画像/階層画像(に含まれる部分画像)とテンプレート画像との間に相対的なずれ(幾何学的なずれ)を算出および補正する方法として、RIPOC法を用いた例について説明した。これに加えて、あるいは、これに代えて、倍率(サイズ)も含めた幾何学的なずれを算出および補正する各種のアルゴリズム方法を採用することもでき、例えば、上述の非特許文献1に開示されるフーリエメリン変換を用いたマッチング方法を採用してもよい。このフーリエメリン変換を用いたマッチング方法によれば、上述した回転量の探索処理と同様に、倍率(サイズ)の探索処理を行なうことができる。
本発明の実施の形態としては、以下のような態様を含む。
Claims (7)
- 基準画像に対応する領域を入力画像から探索する画像処理装置であって、
前記入力画像に対応する、少なくとも第1の解像度および前記第1の解像度より高い第2の解像度をそれぞれ有する第1および第2の入力画像を取得する入力画像取得部と、
前記基準画像に対応する、前記第1および第2の解像度にそれぞれ対応する解像度を有する第1および第2の基準画像を取得する基準画像取得部と、
前記第1の入力画像と前記第1の基準画像との間の探索処理によって、前記第1の入力画像と前記第1の基準画像との間の回転および/または倍率の相対的なずれ量を算出するとともに、算出された相対的なずれ量に基づく前記第1の基準画像の補正によって第1の補正基準画像を生成し、前記第1の入力画像と前記第1の補正基準画像との間の探索処理によって、前記第1の解像度における対応位置を決定する第1の対応位置決定部と、
前記第1の解像度における対応位置に相当する前記第2の入力画像における位置を基準として、前記第1の対応位置決定部によって算出された相対的なずれ量を利用して、前記第2の入力画像と前記第2の基準画像との間の探索処理を行なうことによって、前記第2の解像度における対応位置を決定する第2の対応位置決定部とを備えた画像処理装置。 - 前記第2の対応位置決定部は、前記第1の入力画像と前記第1の基準画像との間の相対的なずれ量に応じて探索範囲を制限した上で、前記第2の入力画像と前記第2の基準画像との間の探索処理を行なうことで、前記第2の入力画像と前記第2の基準画像との間の回転および/または倍率の相対的なずれ量を算出するとともに、算出された相対的なずれ量に基づく前記第2の基準画像の補正によって第2の補正基準画像を生成し、前記第2の入力画像と前記第2の補正基準画像との間の探索処理によって、前記第2の解像度における対応位置を決定する、請求項1に記載の画像処理装置。
- 前記第2の対応位置決定部は、前記基準画像取得部に、前記第1の入力画像と前記第1の基準画像との間の相対的なずれ量に応じて前記第2の基準画像を生成させる、請求項2に記載の画像処理装置。
- 前記第2の対応位置決定部は、前記第2の入力画像と前記第2の基準画像との間の探索処理を行なうことで、前記第2の入力画像と前記第2の基準画像との間の回転および/または倍率の相対的なずれ量を算出するとともに、算出された相対的なずれ量に基づく前記第2の基準画像の補正によって第2の補正基準画像を生成し、前記第2の入力画像と前記第2の補正基準画像との間の探索処理によって、前記第2の解像度における対応位置を決定する、請求項3に記載の画像処理装置。
- 前記第1の入力画像と前記第1の基準画像との間の探索処理において算出される類似度、および、前記第1の入力画像と前記第1の補正基準画像との間の探索処理において算出される類似度の少なくとも一方に基づいて、前記第2の対応位置決定部による前記第2の解像度における対応位置の決定方法を変化させる制御部をさらに備える、請求項1~4のいずれか1項に記載の画像処理装置。
- 基準画像に対応する領域を入力画像から探索する画像処理方法であって、
前記入力画像に対応する、少なくとも第1の解像度および前記第1の解像度より高い第2の解像度をそれぞれ有する第1および第2の入力画像を取得するステップと、
前記基準画像に対応する、前記第1および第2の解像度にそれぞれ対応する解像度を有する第1および第2の基準画像を取得するステップと、
前記第1の入力画像と前記第1の基準画像との間の探索処理によって、前記第1の入力画像と前記第1の基準画像との間の回転および/または倍率の相対的なずれ量を算出するとともに、算出された相対的なずれ量に基づく前記第1の基準画像の補正によって第1の補正基準画像を生成し、前記第1の入力画像と前記第1の補正基準画像との間の探索処理によって、前記第1の解像度における対応位置を決定するステップと、
前記第1の解像度における対応位置に相当する前記第2の入力画像における位置を基準として、前記相対的なずれ量を利用して、前記第2の入力画像と前記第2の基準画像との間の探索処理を行なうことで、前記第2の解像度における対応位置を決定するステップとを備えた画像処理方法。 - 基準画像に対応する領域を入力画像から探索する画像処理プログラムであって、前記画像処理プログラムはコンピューターに、
前記入力画像に対応する、少なくとも第1の解像度および前記第1の解像度より高い第2の解像度をそれぞれ有する第1および第2の入力画像を取得するステップと、
前記基準画像に対応する、前記第1および第2の解像度にそれぞれ対応する解像度を有する第1および第2の基準画像を取得するステップと、
前記第1の入力画像と前記第1の基準画像との間の探索処理によって、前記第1の入力画像と前記第1の基準画像との間の回転および/または倍率の相対的なずれ量を算出するとともに、算出された相対的なずれ量に基づく前記第1の基準画像の補正によって第1の補正基準画像を生成し、前記第1の入力画像と前記第1の補正基準画像との間の探索処理によって、前記第1の解像度における対応位置を決定するステップと、
前記第1の解像度における対応位置に相当する前記第2の入力画像における位置を基準として、前記相対的なずれ量を利用して、前記第2の入力画像と前記第2の基準画像との間の探索処理を行なうことで、前記第2の解像度における対応位置を決定するステップとを実行させる画像処理プログラム。
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US14/411,295 US20150345936A1 (en) | 2012-06-26 | 2013-06-17 | Image Processing Apparatus, Image Processing Method And Image Processing Program |
EP13810312.2A EP2866195A4 (en) | 2012-06-26 | 2013-06-17 | IMAGE PROCESSING DEVICE, IMAGE PROCESSING METHOD, AND IMAGE PROCESSING PROGRAM |
JP2013544957A JP5477518B1 (ja) | 2012-06-26 | 2013-06-17 | 画像処理装置、画像処理方法および画像処理プログラム |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2012142969 | 2012-06-26 | ||
JP2012-142969 | 2012-06-26 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2014002813A1 true WO2014002813A1 (ja) | 2014-01-03 |
Family
ID=49782974
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2013/066586 WO2014002813A1 (ja) | 2012-06-26 | 2013-06-17 | 画像処理装置、画像処理方法および画像処理プログラム |
Country Status (4)
Country | Link |
---|---|
US (1) | US20150345936A1 (ja) |
EP (1) | EP2866195A4 (ja) |
JP (1) | JP5477518B1 (ja) |
WO (1) | WO2014002813A1 (ja) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018037891A1 (ja) * | 2016-08-25 | 2018-03-01 | 日本電気株式会社 | 画像処理装置、画像処理システム、画像処理方法およびプログラム記録媒体 |
JP2019070974A (ja) * | 2017-10-10 | 2019-05-09 | 株式会社アクセル | 画像処理装置、画像処理方法、および画像処理プログラム |
JP2019139640A (ja) * | 2018-02-14 | 2019-08-22 | シヤチハタ株式会社 | 認証システムおよび認証方法 |
JP7496546B2 (ja) | 2020-05-28 | 2024-06-07 | パナソニックIpマネジメント株式会社 | 画像処理方法、プログラム及び画像処理システム |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP6278108B2 (ja) * | 2014-03-14 | 2018-02-14 | オムロン株式会社 | 画像処理装置、画像センサ、画像処理方法 |
JP6333871B2 (ja) * | 2016-02-25 | 2018-05-30 | ファナック株式会社 | 入力画像から検出した対象物を表示する画像処理装置 |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH10124667A (ja) | 1996-08-26 | 1998-05-15 | Yamatake Honeywell Co Ltd | パターン照合装置 |
JP2001223874A (ja) * | 2000-02-04 | 2001-08-17 | Kiyoharu Aizawa | 任意焦点画像合成装置及びこれに用いる複数画像同時撮像型カメラ |
JP2005235089A (ja) | 2004-02-23 | 2005-09-02 | Japan Science & Technology Agency | 画像認識装置、画像認識方法及びプログラム |
JP2011198148A (ja) * | 2010-03-19 | 2011-10-06 | Fujitsu Ltd | 画像処理装置、画像処理方法および画像処理プログラム |
JP2012069003A (ja) * | 2010-09-24 | 2012-04-05 | Juki Corp | 形状ベースマッチングパラメータの調整装置、形状ベースマッチングパラメータの調整方法および部品実装装置 |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6219452B1 (en) * | 1999-01-06 | 2001-04-17 | National Instruments Corporation | Pattern matching system and method which performs local stability analysis for improved efficiency |
US7113652B2 (en) * | 2003-01-09 | 2006-09-26 | Banner Engineering Corp. | System and method for using normalized gray scale pattern find |
US8483489B2 (en) * | 2011-09-02 | 2013-07-09 | Sharp Laboratories Of America, Inc. | Edge based template matching |
-
2013
- 2013-06-17 JP JP2013544957A patent/JP5477518B1/ja not_active Expired - Fee Related
- 2013-06-17 EP EP13810312.2A patent/EP2866195A4/en not_active Withdrawn
- 2013-06-17 WO PCT/JP2013/066586 patent/WO2014002813A1/ja active Application Filing
- 2013-06-17 US US14/411,295 patent/US20150345936A1/en not_active Abandoned
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH10124667A (ja) | 1996-08-26 | 1998-05-15 | Yamatake Honeywell Co Ltd | パターン照合装置 |
JP2001223874A (ja) * | 2000-02-04 | 2001-08-17 | Kiyoharu Aizawa | 任意焦点画像合成装置及びこれに用いる複数画像同時撮像型カメラ |
JP2005235089A (ja) | 2004-02-23 | 2005-09-02 | Japan Science & Technology Agency | 画像認識装置、画像認識方法及びプログラム |
JP2011198148A (ja) * | 2010-03-19 | 2011-10-06 | Fujitsu Ltd | 画像処理装置、画像処理方法および画像処理プログラム |
JP2012069003A (ja) * | 2010-09-24 | 2012-04-05 | Juki Corp | 形状ベースマッチングパラメータの調整装置、形状ベースマッチングパラメータの調整方法および部品実装装置 |
Non-Patent Citations (3)
Title |
---|
HITOSHI KIYA: "Integration of Image Signal Processing and Image Pattern Recognition - DCT Sign Only Correlation and Its Application", TOKYO METROPOLITAN UNIVERSITY, FACULTY OF SYSTEM DESIGN, JSPE WORKSHOP ON DYNAMIC IMAGE PROCESSING FOR REAL APPLICATION, 2007 |
QIN-SHENG CHEN: "Symmetric Phase-Only Matched Filtering of Fourier-Mellin Transforms for Image Registration and Recognition", IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, vol. 16, no. 12, December 1994 (1994-12-01), XP000486818, DOI: doi:10.1109/34.387491 |
See also references of EP2866195A4 |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018037891A1 (ja) * | 2016-08-25 | 2018-03-01 | 日本電気株式会社 | 画像処理装置、画像処理システム、画像処理方法およびプログラム記録媒体 |
JPWO2018037891A1 (ja) * | 2016-08-25 | 2019-06-20 | 日本電気株式会社 | 画像処理装置、画像処理システム、画像処理方法およびプログラム記録媒体 |
US10922823B2 (en) | 2016-08-25 | 2021-02-16 | Nec Corporation | Motion analyis device, motion analysis method, and program recording medium |
JP7063267B2 (ja) | 2016-08-25 | 2022-05-09 | 日本電気株式会社 | 画像処理装置、画像処理システム、画像処理方法およびコンピュータプログラム |
JP2019070974A (ja) * | 2017-10-10 | 2019-05-09 | 株式会社アクセル | 画像処理装置、画像処理方法、および画像処理プログラム |
JP2019139640A (ja) * | 2018-02-14 | 2019-08-22 | シヤチハタ株式会社 | 認証システムおよび認証方法 |
JP7076772B2 (ja) | 2018-02-14 | 2022-05-30 | シヤチハタ株式会社 | 認証システムおよび認証方法 |
JP7496546B2 (ja) | 2020-05-28 | 2024-06-07 | パナソニックIpマネジメント株式会社 | 画像処理方法、プログラム及び画像処理システム |
Also Published As
Publication number | Publication date |
---|---|
JPWO2014002813A1 (ja) | 2016-05-30 |
JP5477518B1 (ja) | 2014-04-23 |
US20150345936A1 (en) | 2015-12-03 |
EP2866195A1 (en) | 2015-04-29 |
EP2866195A4 (en) | 2016-06-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP5477518B1 (ja) | 画像処理装置、画像処理方法および画像処理プログラム | |
JP4284288B2 (ja) | パターン認識装置及びその方法 | |
US20150262346A1 (en) | Image processing apparatus, image processing method, and image processing program | |
CN104751465A (zh) | 一种基于lk光流约束的orb图像特征配准方法 | |
US20130051626A1 (en) | Method And Apparatus For Object Pose Estimation | |
US9390330B2 (en) | Apparatus and method for extracting correspondences between aerial images | |
WO2017099251A1 (en) | Method and system for registering data | |
JP6512100B2 (ja) | 画像処理を実行する情報処理装置及び画像処理方法 | |
JP2013025799A (ja) | 画像検索方法、システム、及びプログラム | |
JP4521235B2 (ja) | 撮影画像の変化抽出装置及び方法 | |
KR20170022736A (ko) | 딕셔너리 학습 기반 해상도 향상 장치 및 방법 | |
JP6052186B2 (ja) | 画像処理装置 | |
US20200005078A1 (en) | Content aware forensic detection of image manipulations | |
Al-Hammadi et al. | Improving SURF based copy-move forgery detection using super resolution | |
CN114511865A (zh) | 一种结构化信息的生成方法、装置和计算机可读存储介质 | |
JP6086491B2 (ja) | 画像処理装置およびそのデータベース構築装置 | |
JP6163868B2 (ja) | 画像処理方法、画像処理装置および画像処理プログラム | |
JP5051671B2 (ja) | 情報処理装置、情報処理方法およびプログラム | |
CN106056599B (zh) | 一种基于物体深度数据的物体识别算法及装置 | |
US11232323B2 (en) | Method of merging images and data processing device | |
JP6056354B2 (ja) | 画像処理装置、画像処理方法および画像処理プログラム | |
WO2018076137A1 (zh) | 一种高光谱图像特征描述子的获取方法及装置 | |
JP2014029677A (ja) | 画像処理装置、画像処理方法および画像処理プログラム | |
JP2005173128A (ja) | 輪郭形状抽出装置 | |
JP3706603B2 (ja) | データ特徴抽出装置及びデータ照合装置 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
ENP | Entry into the national phase |
Ref document number: 2013544957 Country of ref document: JP Kind code of ref document: A |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 13810312 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 14411295 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
REEP | Request for entry into the european phase |
Ref document number: 2013810312 Country of ref document: EP |