JP2006013651A - Characteristic point detection apparatus, characteristic point detecting method, and characteristic point detection program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a characteristic point detection apparatus for utilizing the data structure of the JPEG 2000 so as to detect a characteristic point with high accuracy without the need for noise elimination processing wherein a computing time can be shortened through parallel processing. <P>SOLUTION: Upon the receipt of image data of 2LL components by a transmission reception section 11, a characteristic point candidate detection processing section 121 carries out detection processing of characteristic point candidates from the image data of the 2LL components (S2). A decoding processing section 15 generates image data of one LL component (S3). The characteristic point candidate detection processing section 121 carries out detection processing of characteristic point candidates from the image data of the LL components (S4). A characteristic point determining section 122 determines a characteristic point on the basis of the characteristic point candidates detected in S4 and the characteristic point candidates detected in S2 (S5). A decode processing section 15 generates original image data (S6). An image region segmentation processing section 13 applies segmentation processing of a watermark embedded image region to the original image data (S7). A watermark detection processing section 14 carries out processing of detecting watermarks. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a feature point detection device that executes feature point detection processing in parallel with server-side data reception processing, and a watermark detection device and watermark detection system including such a feature point detection device.

  As shown in FIG. 1, there is a watermark detection system 100 that uses a digital camera (or a mobile phone 2 with a camera) to photograph a printed matter 3 on which an image 31 with a digital watermark is printed, and detects a watermark in the server 1 ( For example, Patent Document 1).

  In such a system, various processes are required on the detection side (server 1) in order to improve the watermark detection rate. As one of them, there is an “image region extraction process” for extracting an area in which a digital watermark is embedded (for example, Non-Patent Document 1).

  The image area is usually cut out based on feature points (vertices of the image area, edges in the image, etc.) detected from the captured image. In general, as shown in FIG. 19, a method of detecting four corners of a watermarked image and cutting out an image area of the watermarked image based on the detected corner position is often used.

  FIG. 20 shows a processing timing chart in such a watermark detection system 100. When the printed matter 3 on which the watermarked image 31 is printed is photographed by the camera-equipped mobile phone 2, the photographed image data is once transmitted to the server 1. After receiving the captured image data, the server 1 performs a decoding process of the captured image data, a feature point detection process, an image region extraction process in which a digital watermark using the detected feature point is embedded, and a watermark detection process. Do. The watermark information detected by the server 1 is transmitted again to the camera-equipped mobile phone 2. Alternatively, service data (for example, video content or music content) based on the detected watermark information is transmitted to the camera-equipped mobile phone 2.

  Incidentally, there is a standard for a high-quality image compression method using wavelet transform called JPEG2000. In the JPEG2000 encoding method, a code is configured with a pass (sub-bit plane) as a minimum unit. That is, it is possible to give priority in pass units after encoding and control temporal priority during decoding. There are four priorities: L: layer, R: spatial resolution level, P: position, and C: color component (Non-Patent Document 2).

  FIG. 21 and FIG. 22 show a state when the captured image used for the explanation in FIG. 19 is stage-decomposed by wavelet transformation. FIG. 21 shows an example of processing when an analysis bank is applied to an image signal, and the image is decomposed into four bands (LL, LH, HL, HH) (this is referred to as “one-stage decomposition”). An analysis bank has a function of decomposing a signal into two bands and is used in an encoder. According to this processing example, the image in the lowest range (LL) has more energy than other bands, and maintains the image quality closest to the original image. In JPEG2000, this lowest range (LL) is further decomposed. FIG. 22 is an example of two-stage decomposition (in JPEG 2000, five-stage decomposition is the default).

  FIG. 23 shows the data structure of an image file in the JPEG 2000 format when it is decomposed in two stages. In this case, the data portion of the image file has a hierarchical structure for each resolution level. That is, among the plurality of image data having different frequency bands from the low frequency side to the high frequency side constituting the hierarchically encoded image data based on the resolution, the low frequency side image data precedes the high frequency image data. The side image data is behind.

When a JPEG2000 image having such a data structure is transmitted, first, low-frequency image data is transmitted, and then high-frequency image data is transmitted.
Japanese translation of PCT publication No. 2002-544637 Kenichi Kanaya “Image Understanding -Mathematical Mathematical 3D Recognition” Chapter 3 "Interface", January 2002 issue, CQ Publisher, pp 46-58

  However, in such a watermark detection system 100, each processing is performed after transmission of the captured image data to the server 1 is completed. Therefore, there is a problem that the client has to wait for a long time from the start of transmission of the watermarked image to the acquisition of the watermark extraction result.

  The first object of the present invention is to reduce the processing time in the system by executing the data reception process and the feature point detection process on the server side in parallel using the data structure of JPEG2000.

  By the way, it is important to detect feature points with high accuracy in the clipping process of an image region in which a digital watermark is embedded. This is because the watermark information cannot be correctly read out from the watermarked image unless the feature point (for example, the corner of the watermarked image region) can be detected correctly.

  However, images taken with a digital camera or the like contain noise due to camera shake and illumination, which causes a reduction in feature point detection accuracy. Therefore, in order to detect feature points with high accuracy from captured image data, noise removal processing is one of the necessary processing.

  In general, a smoothing filter represented by a Gaussian filter is used for noise removal processing. The smoothing filter takes a weighted average in which a large weight is given to pixels close to the target pixel and a small weight is given to pixels far from the target pixel. Therefore, after the noise removal processing, an edge is blurred, in other words, an image in which a high frequency component is cut is generated.

  From this, it is considered that the low-frequency image in JPEG 2000 and the image after noise removal processing have similar characteristics. Therefore, a second object of the present invention is to improve the watermark detection rate by using the data structure of JPEG2000.

  The feature point detection apparatus according to claim 1 is a feature point detection apparatus for detecting a feature point from hierarchically encoded image data, from a low frequency side to a high frequency side constituting the hierarchically encoded image data. The feature point is detected based only on the low-frequency side image data among the plurality of image data having different frequency bands.

  Here, the feature point means, for example, a vertex of an image area or an edge in an image. Hierarchically encoded image data refers to image data having a structure in which image data is decomposed according to resolution, such as JPEG2000.

  According to the feature point detecting apparatus of the first aspect, since the feature point is detected based on the image data of the low frequency component from which the high frequency component is cut, the feature point can be detected more accurately.

  The feature point detection apparatus according to claim 2 is a feature point detection apparatus that detects a feature point from hierarchically encoded image data, and includes a first feature point candidate detection unit, a second feature point candidate detection unit, And feature point determination means. The first feature point candidate detection means includes a low frequency side among n (n: natural number) image data having different frequency bands from the low frequency side to the high frequency side constituting the hierarchically encoded image data. First feature point candidates obtained from l (l <n, l: natural number) image data are detected. The second feature point candidate detection means is a second feature point obtained from m (l <m <n, m: natural number) image data on the low frequency side of the n image data. Detect candidates. The feature point determination means includes the coordinates of the first feature point candidate detected by the first feature point candidate detection means, and the second feature point candidate detected by the second feature point candidate detection means. When the “coordinates” match, the feature point candidate is determined to be a feature point of the hierarchically encoded image data.

  The feature point detection apparatus according to claim 3 is a feature point detection apparatus that detects a feature point from hierarchically encoded image data, and includes a first feature point candidate detection unit, a second feature point candidate detection unit, And feature point determination means. The first feature point candidate detection means includes a low frequency side among n (n: natural number) image data having different frequency bands from the low frequency side to the high frequency side constituting the hierarchically encoded image data. First feature point candidates obtained from l (l <n, l: natural number) image data are detected. The second feature point candidate detection means is a second feature point obtained from m (l <m <n, m: natural number) image data on the low frequency side of the n image data. Detect candidates. The coordinate conversion means includes coordinate values of the first feature point candidates detected by the first feature point candidate detection means, and the second feature points detected by the second feature point candidate detection means. Candidate coordinate values are converted into coordinate values in the original image size. The feature point determining means is a coordinate value in the original image size of the first feature point candidate converted by the coordinate converting means, and an original image of the second feature point candidate converted by the coordinate converting means. When the coordinate value in size matches, the coordinate value is determined as the coordinate value of the feature point of the hierarchically encoded image data.

  According to a fourth aspect of the present invention, there is provided an image processing apparatus including a reception unit that receives hierarchically encoded image data and a feature point detection unit that detects a feature point from the image data received by the reception unit. When the receiving unit receives image data on the low frequency side among a plurality of image data having different frequency bands from the low frequency side to the high frequency side constituting the hierarchically encoded image data, The feature point detection unit starts the feature point detection process from only the low-frequency image data without waiting for reception of high-frequency image data by the reception unit.

  The feature point detection method according to claim 5 is a feature point detection method for detecting a feature point from hierarchically encoded image data, from a low frequency side to a high frequency side constituting the hierarchically encoded image data. The feature point is detected based only on the low-frequency side image data among the plurality of image data having different frequency bands.

  The feature point detection method according to claim 6 receives image data on the low frequency side among a plurality of image data having different frequency bands from the low frequency side to the high frequency side constituting the hierarchically encoded image data. A receiving step, a feature point detecting step for detecting a feature point from the image data received in the receiving step, and a feature point detecting step comprising: image data on a high frequency side of the plurality of image data. Before receiving the feature point, the feature point detection process is started only from the low-frequency image data received in the reception step.

  The feature point detection program according to claim 7 is a feature point detection program for causing a computer to realize a function of detecting a feature point from hierarchically encoded image data. To allow a computer to realize a feature point detection function based on only low-frequency side image data among a plurality of image data having different frequency bands from the low frequency side to the high frequency side constituting the image data. It is characterized by.

  The image processing program according to claim 8 receives image data on the low frequency side among a plurality of image data having different frequency bands from the low frequency side to the high frequency side constituting the hierarchically encoded image data. A receiving step; a feature point detecting step for detecting a feature point from the image data received in the receiving step; and the feature point detecting step comprising: extracting image data on a high frequency side from the plurality of image data. Before reception, feature point detection processing is started only from the image data on the low frequency side received in the reception step.

  According to the watermark detection system of the present invention, the waiting time from the start of transmission of a watermarked image to the acquisition of a watermark extraction result on the client side can be reduced. In addition, since the feature points can be detected with high accuracy, the watermark detection rate can be improved.

  Referring again to FIG. 1, the configuration of a watermark detection system 100 to which the present invention is applied is shown. The watermark detection system 100 includes a server 1 and a mobile phone 2 with a camera. The camera-equipped mobile phone 2 generates image data (JPEG2000 format) of the printed material 3 and transmits the image data to the server 1.

  The server 1 performs watermark detection from the image data transmitted from the camera-equipped mobile phone 2. The server 1 returns a watermark detection result to the camera-equipped mobile phone 2. Alternatively, the server 1 transmits service data corresponding to the watermark detection result to the camera-equipped mobile phone 2.

  Normally, the 5-stage decomposition is standard in the JPEG2000 format. However, in the present embodiment, for simplicity of explanation, the JPEG2000 encoder incorporated in the camera-equipped mobile phone 2 has a 2-stage decomposition as shown in FIG. The following description will be made on the assumption that the data is encoded in the above format. That is, it is assumed that photographed image data having a structure as shown in FIG. 23 is transmitted from the camera-equipped mobile phone 2 to the server 1.

  With reference to FIG. 2, the configuration of the server 1 of the watermark detection system 100 to which the present invention is applied will be described.

  The server 1 includes a transmission / reception unit 11, a feature point detection processing unit 12, an image region cutout processing unit 13, a watermark detection processing unit 14, a decoding processing unit 15, an image composition unit 16, and a control unit 17.

  The transmission / reception unit 11 receives photographed image data transmitted from the camera-equipped mobile phone 2 or transmits watermark detection result data or service data corresponding to the watermark detection result to the camera-equipped cellular phone 2.

  The decoding processing unit 15 decodes the encoded image data received by the transmission / reception unit 11. Decoding as used herein mainly refers to reconstructing image data by synthesizing image data subjected to stage decomposition. For example, what is called a synthesis bank is used for the synthesis of the image data. The synthesis bank has a function of combining two band signals into one, and is used in a decoder.

  The feature point detection processing unit 12 performs processing for detecting feature points (for example, four corners of the watermarked image region) used to cut out the watermarked image region from the image data decoded by the decoding processing unit 15.

  The feature point detection processing unit 12 includes a feature point candidate detection processing unit 121 and a feature point determination unit 122. The feature point candidate detection processing unit 121 extracts a feature point candidate from the stage-decomposed image data or the image data decoded by the decoding processing unit 15. If the image data transmitted from the camera-equipped mobile phone 2 to the server 1 is encoded image data that has been decomposed into two stages, the stage-decomposed image data that is the target of feature point candidate extraction is: This is the image data of the 2LL component as referred to in FIG. The image data decoded by the decoding processing unit 15 is reconstructed by the decoding processing unit 15 based on 2LL component image data, 2LH component image data, 2HL component image data, and 2HH component image data. LL component image data.

  The feature point determination unit 122 determines a feature point from the feature point candidates detected by the feature point candidate detection processing unit 121. The control unit 17 performs various settings and controls each configuration of the server 1.

  The image region cutout processing unit 13 performs a watermarked image region cutout process from the captured image data decoded by the decoding processing unit 15. In this cutout process, the feature points detected by the feature point detection processing unit 12 are used.

  The watermark detection processing unit 14 performs processing for detecting a watermark from the image region cut out by the image region cut-out processing unit 13.

  These configurations can be realized in hardware by a CPU, memory, or other LSI of any computer, and in software, a program having an image processing function and a digital watermark embedding function loaded in the memory, etc. Here, functional blocks realized by their cooperation are depicted. Therefore, those skilled in the art will understand that these functional blocks can be realized in various forms by hardware only, software only, or a combination thereof.

  FIG. 3 shows a time chart of digital watermark detection processing in the watermark detection system 100 to which the present invention is applied.

  Image data captured by the camera-equipped mobile phone 2 is encoded by DWT (Discrete Wavelet Transform, discrete wavelet transform), and converted into image data having a layered structure with a resolution as shown in FIG. As shown in FIG. 4, the conventional server 1 that receives such image data performs decoding processing, feature point detection processing, and the like after receiving all of the image data. For the user, there is a problem that a long time is required from the transmission of the watermarked image until the response from the server 1 is obtained.

  On the other hand, the server 1 of this embodiment does not perform the decoding process, the feature point detection process, etc. after receiving all of the image data, but a part of the arrived image data (for example, 2LL component image data, etc.) ) To perform feature point candidate detection processing, etc., as needed.

(Step S1: Initial setting process)
Referring to FIG. 3 again, when the transmission / reception unit 11 of the server 1 receives the header, the control unit 17 analyzes the contents of the header, and the arrived data is JPEG2000 format image data, and the watermarked image is received. It recognizes that it is data (step S1). And the control part 17 starts the program for performing the watermark detection process demonstrated by this embodiment.

(Step S2: feature point candidate detection processing)
When the transmission / reception unit 11 receives 2LL component image data, the data is transferred to both the feature point candidate detection processing unit 121 of the feature point detection processing unit 12 and the decoding processing unit 15. The feature point candidate detection processing unit 121 performs feature point candidate detection processing from 2LL component image data (step S2). When the feature points are corner points of the image area, for example, as shown in FIG. 5A, a plurality of corner points are detected from the image data of 2LL components. This feature point candidate detection process will be described in detail later.

(Step S3: Decoding process)
When the transmission / reception unit 11 receives 2LH component image data, 2HL component image data, and 2HH component image data following the 2LL component image data, these data are transferred to the decoding processing unit 15. The decoding processing unit 15 generates one LL component image data from these three image data and the four image data of the already transferred 2LL component image data (step S3). A synthesis bank is used for this decoding process.

  In the above description, it has been described that the decoding process is performed when the three image data of the 2LH component, the 2HL component, and the 2HH component arrive. However, when the image data of the 2LH component arrives first, it has already been transferred to the decoding processing unit 15. Vertical processing (vertical one-dimensional convolution processing) may be performed using 2LL component image data and 2LH component image data. Subsequently, when the image data of 2HL component and 2HH component arrive, vertical processing is performed using the image data of 2HL component and the image data of 2HH component, and the two image data obtained by the vertical processing are processed horizontally (1 in the horizontal direction). Image data of the LL component may be generated by performing (dimensional convolution processing).

(Step S4: feature point candidate detection process)
When LL component image data is generated by the decoding processing unit 15, the feature point candidate detection processing unit 121 performs feature point candidate detection processing from the LL component image data (step S4). When the feature points are corner points of the image area, for example, as shown in FIG. 5B, a plurality of corner points are detected from the image data of the LL component. The feature point candidate detection process will be described in detail later, but is exactly the same as the process in step S2 except that the target image data is different.

(Step S5: feature point determination processing)
When the feature point candidate detection process (step S4) by the feature point candidate detection processing unit 121 ends, the feature point determination unit 122 determines whether the feature point candidate is detected in step S4 and the feature point candidate detected in step S2. Based on this, feature points are determined (step S5). This feature point determination process will also be described in detail later.

(Step S6: Decoding process)
When the transmission / reception unit 11 receives LH component image data, HL component image data, and HH component image data, these data are transferred to the decoding processing unit 15. From these three pieces of image data and the four pieces of image data of the LL component image data already generated by the decoding processing unit 15, the decoding processing unit 15 generates original image data (image data before encoding). (Step S6).

(Step S7: Image region cutout process)
In step S6, the image area segmentation processing unit 13 performs a watermarked image area segmentation process on the decoded original image data (step S7). In this cut-out process, the feature points determined in step S5 are used.

(Step S8: Watermark detection processing)
From the image area cut out in step S7, the watermark detection processing unit 14 performs processing for detecting a watermark. Based on the information detected by the watermark detection process, the server 1 recognizes the service content (for example, music content distribution) desired by the user (user of the camera-equipped mobile phone 2), and the server 1 receives appropriate content data. It transmits to the mobile phone 2 with a camera.

  The above is the outline of the processing of the server 1 in the watermark detection system 100 of the present embodiment. Next, details of the feature point candidate detection process in steps S2 and S4 and the feature point determination process in step S5 will be described.

(Details of feature point candidate detection processing in steps S2 and S4)
A detailed example of the feature point candidate detection process in steps S2 and S4 will be described with reference to FIGS. Assuming that the image size of the original image is 1024 pixels vertically and 1280 pixels horizontally, the size of the image data of the 2LL component is 256 pixels vertically and 320 pixels horizontally.

  With reference to the flowchart of FIG. 6, the process procedure of step S2 which the feature point candidate detection process part 121 performs is demonstrated. In step S2, feature point candidate detection processing is performed on the image data of the 2LL component. In step S21, calculation processing by a feature point detection filter is performed for all coordinate points of 2LL component image data.

  In step S22, the coordinates of the point whose filter calculation result is equal to or greater than a predetermined value are added to the first feature point candidate list. This first feature point candidate list is as shown in FIG. 7 and is temporarily stored in the storage area of the feature point candidate detection processing unit 121. In this embodiment, four points A2 (80, 40), B2 (240, 40), C2 (80, 200), and D2 (240, 200) (see also FIG. 8a) are used as feature point candidates. The feature point candidates are detected by the point candidate detection processing unit 121 and stored in the first feature point candidate list.

  Note that the process of step S5 is different from step S2 only in that the image that is the target of the feature point candidate detection process is LL component image data, and the processing content is the same as step S2. With reference to FIGS. 7 and 8B, in this embodiment, A1 (160, 80), B1 (480, 80), C1 (320, 280), D1 (160, 400), E1 (480, 400) are detected as feature point candidates by the feature point candidate detection processing unit 121, and these feature point candidates are also stored in the first feature point candidate list.

(Details of feature point determination processing in step S5)
A detailed example of the feature point determination process in step S5 performed by the feature point determination unit 122 will be described with reference to FIGS. Here, description will be made on the assumption that feature point determination is performed based on the feature point candidates detected in the feature point candidate detection processing in step S2 and step S5 described with reference to FIGS.

  With reference to the flowchart of FIG. 9, the process procedure of step S5 is demonstrated.

  In step S51, coordinate conversion processing is performed on all feature point candidates in the first feature point candidate list of FIG. In other words, the 2LL component image data and the coordinates indicating the position of the feature point candidate detected in the LL component image data are converted into coordinates indicating the position in the original image. For 2LL component image data, the x coordinate value and the y coordinate value are each multiplied by four. For the LL component image data, the x coordinate value and the y coordinate value are each doubled.

  In step S52, the calculation result is stored in the second feature point candidate list as shown in FIG. In the second feature point candidate list, A′2, B′2, C′2, and D′ 2 obtained by performing coordinate transformation on the coordinates of the feature point candidates A2, B2, C2, and D2 in the 2LL component image data. The coordinate values of four points and the coordinate values of five points A′1, B′1, C′1, D′ 1, and E′1, which are coordinate-converted with respect to the coordinates of the feature point candidates in the LL component image data, are as follows. Stored.

  In step S53, the second feature point candidate list is referred to, and feature point candidates whose coordinate values detected from both image data match are examined. A feature point candidate whose coordinate values match is determined to be a feature point. Referring to the second feature point candidate list, four sets of feature point candidates A'2 and A'1, B'2 and B'1, C'2 and D'1, D'2 and E'1, The coordinate values match (see also FIG. 11). These four sets of coordinate values are regarded as the coordinate values of the feature points in the original image data, and are stored in the feature point list shown in FIG.

  Note that the process of step S5 is different from step S2 only in that the image that is the target of the feature point candidate detection process is LL component image data, and the processing content is the same as step S2. With reference to FIGS. 7 and 8B, in this embodiment, A1 (160, 80), B1 (480, 80), C1 (320, 280), D1 (160, 400), E1 (480, 400) are detected as feature point candidates by the feature point candidate detection processing unit 121, and these feature point candidates are also stored in the first feature point candidate list.

Note that the difference between the image size of the 2LL component image data and the image size of the LL component image data even though the feature point candidate in the 2LL component image data and the feature point candidate in the LL component image data should match. Due to this, the coordinate values of feature point candidates after coordinate conversion may be slightly different. For example, in the second feature point candidate list of FIG. 10, the coordinate value related to the feature point candidate in the 2LL component image data is only a multiple of 4, whereas the coordinate value related to the feature point candidate in the LL component image data is 2 Only takes a multiple of. Therefore, in such a case, even when the difference between the coordinate value related to the feature point candidate in the 2LL component image data and the coordinate value related to the feature point candidate in the LL component image data is 2, the feature point candidate in the 2LL component image data You may consider that a coordinate and the coordinate of the feature point candidate in LL component image data correspond.
(Feature point detection processing from 3-stage decomposed image data)
An example of performing feature point detection from an encoded image that has been decomposed in three stages as shown in FIG. 13 will be described with reference to FIGS. 14 to 18.

  FIG. 14 is a first feature point candidate list indicating the coordinates of feature point candidates detected by the feature point candidate detection unit 121 from 3LL component image data, 2LL component image data, and 1LL component image data. Three feature point candidates are detected from 3LL component image data, five points from 2LL component image data, and seven points from LL component image data. The 2LL component image data is reconstructed by combining four image data of 3LL, 3LH, 3HL, and 3HH.

  FIG. 15 is a flowchart illustrating a processing procedure in the feature point determination process (modified example of step S5) performed by the feature point determination unit 122 in such a case.

  In step S61, coordinate conversion processing is performed for all feature point candidates in the first feature point candidate list of FIG. That is, a process of converting the coordinate value indicating the position of the feature point candidate detected in the 3LL, 2LL, and LL component image data into the coordinate value indicating the position in the original image is performed. In step S62, the calculation result is stored in the second feature point candidate list as shown in FIG. The second feature point candidate list stores coordinate values obtained by performing coordinate conversion on the feature point candidate coordinates in the 3LL, 2LL, and LL component image data.

  In step S63, the second feature point candidate list is referred to, and the number of image data whose coordinate values stored in the list match is counted. The counting result is stored in the third feature point candidate list shown in FIG. For example, since the coordinate values (320, 160) are coordinate values of coordinate point candidates detected based on the three image data of 3LL, 2LL, and LL, the coordinate value ( 320, 160) and the number of detected images 3 are stored as feature point candidate data of feature point number 1.

  Similarly, since the coordinate value (960, 160) is detected from two images of 3LL and 2LL, the coordinate value (960, 160) and the number of detected images 2 are feature point candidates of feature point number 2. Stored as data. Similarly, data relating to feature point candidates from feature point numbers 3 to 10 are stored.

  In step S64, the third feature point candidate list is referred to, and a feature point candidate whose detected image number is equal to or larger than a predetermined value is determined as a feature point. Here, assuming that feature point candidates whose detected images are two or more are feature points, four feature point candidates of feature point numbers 1, 2, 3, and 4 are determined as feature points. The determined feature points are stored in a feature point list as shown in FIG.

(Processing when the number of feature points determined by the feature point determination unit 122 is small)
A process when the number of feature points determined by the feature point determination unit 122 is small will be described. For example, when it is necessary to detect four feature points prior to the image region cut-out process performed by the image region cut-out processing unit 13, only three feature points can be detected. As processing in such a case, the following can be considered.

  The first is to reduce the number of detected images in step S64 in FIG. As a result, there is a possibility that the fourth feature point is determined in step S64.

  Secondly, the feature point detection processing performed by the feature point candidate detection processing unit 121 is performed again. For example, the threshold value of the feature point detection filter processing result in step S22 of FIG. 6 is lowered so that more feature point candidates are added to the first feature point candidate list.

(Processing when there are too many feature points determined by the feature point determination unit 122)
A process when the number of feature points determined by the feature point determination unit 122 is too large will be described. For example, in the case where four feature points need to be detected prior to the image region cut-out process performed by the image region cut-out processing unit 13, six feature points have been detected. As processing in such a case, the following can be considered.

  The first is to increase the number of detected images in step S64 of FIG. As a result, the fifth and sixth feature points may not be determined in step S64.

  Secondly, the feature point detection processing performed by the feature point candidate detection processing unit 121 is performed again. For example, the threshold value of the feature point detection filter processing result in step S22 of FIG. 6 is raised so that fewer feature point candidates are added to the first feature point candidate list.

Third, based on the coordinate value of the feature point determined by the feature point determination unit 122, a process of estimating whether the detected feature point is appropriate is performed. For example, when four corners of the image region are determined as feature points, arbitrary four points are selected from n (n: natural number, n> 4) feature point candidates determined by the feature point determination unit 122. Here, there are _n C ₄ combinations of selections. The combination having the highest similarity between the shape of the figure formed from the four selected points and the shape of the image region can be determined as the feature point.

  The embodiment disclosed this time should be considered as illustrative in all points and not restrictive. The scope of the present invention is shown not by the above description of the embodiments but by the scope of claims for patent, and is intended to include meanings equivalent to the scope of claims for patent and all modifications within the scope.

  In the above-described embodiment, the feature point (square of the image area) detection for extracting a watermarked image area has been described in detail. However, the present invention can also be applied to feature point detection in a general image.

  In the above embodiment, the feature point detection from the image data by the JPEG2000 encoding method has been described. However, the present invention can also be applied to image data that has been hierarchically encoded by another method.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Server 11 Transmission / reception part 12 Feature point detection process part 13 Image area cut-out process part 14 Watermark detection process part 15 Decoding process part 17 Control part 121 Feature point candidate detection process part 122 Feature point determination part

1 is a diagram illustrating a watermark detection system 100. FIG. It is the structure of the server 1 of the watermark detection system 100 to which this invention is applied. 2 is a time chart of digital watermark detection processing in the watermark detection system 100 to which the present invention is applied. It is a time chart of digital watermark detection processing in a conventional watermark detection system. (A) It is a figure showing an example of a feature point candidate detected from 2LL component image data, and (b) an example of a feature point candidate detected from LL component image data. It is a flowchart which shows the procedure of a feature point candidate detection process (step S2, step S4). It is a 1st feature point candidate list. It is the figure shown about the position of the feature point candidate. It is a flowchart which shows the procedure of a feature point determination process (step S5). It is a 2nd feature point candidate list. It is the figure which showed about the position of the feature point candidate when converting the coordinate value of the feature point candidate detected from 2LL component image data and LL component image data into the coordinate value in an original image size. It is a figure of a feature point list. It is a figure which shows the encoding image decomposed | disassembled into 3 stages. It is a figure of the 1st feature point candidate list. It is a flowchart which shows the detail of a feature point determination process (step S5 modification). It is a figure of the 2nd feature point candidate list. It is a figure of the 3rd feature point candidate list. It is a figure of a feature point list. It is a figure explaining the extraction process of an image area. 2 is a timing chart in the conventional watermark detection system 100. It is a figure which shows 1 stage decomposition | disassembly in wavelet transformation. It is a figure which shows 2 stage decomposition | disassembly in a wavelet transform. It shows the data structure of an image file in JPEG2000 format when it is decomposed in two stages.

Claims (8)

A feature point detection device for detecting feature points from hierarchically encoded image data,
Detecting feature points based only on image data on the low frequency side among a plurality of image data having different frequency bands from the low frequency side to the high frequency side constituting the hierarchically encoded image data. A feature point detection device. A feature point detection device for detecting feature points from hierarchically encoded image data,
Among n (n: natural number) image data having different frequency bands from the low frequency side to the high frequency side constituting the hierarchically encoded image data,
First feature point candidate detection means for detecting first feature point candidates obtained from l (l <n, l: natural number) image data on the low frequency side;
Second feature point candidate detection means for detecting second feature point candidates obtained from m (l <m <n, m: natural number) image data on the low frequency side;
The coordinates of the first feature point candidate detected by the first feature point candidate detecting means coincide with the coordinates of the second feature point candidate detected by the second feature point candidate detecting means. And a feature point determining means for determining that the feature point candidate is a feature point of the hierarchically encoded image data. A feature point detection device for detecting feature points from hierarchically encoded image data,
Among n (n: natural number) image data having different frequency bands from the low frequency side to the high frequency side constituting the hierarchically encoded image data,
First feature point candidate detection means for detecting first feature point candidates obtained from l (l <n, l: natural number) image data on the low frequency side;
Second feature point candidate detection means for detecting second feature point candidates obtained from m (l <m <n, m: natural number) image data on the low frequency side;
The coordinate value of the first feature point candidate detected by the first feature point candidate detection unit and the coordinate value of the second feature point candidate detected by the second feature point candidate detection unit , Coordinate conversion means for converting the coordinate value in the original image size,
The coordinate value in the original image size of the first feature point candidate converted by the coordinate conversion unit and the coordinate value in the original image size of the second feature point candidate converted by the coordinate conversion unit. And a feature point determining unit that determines that the coordinate value is the coordinate value of the feature point of the hierarchically encoded image data when the two coincide with each other. A receiving unit for receiving the hierarchically encoded image data;
A feature point detection unit that detects a feature point from the image data received by the reception unit,
When the receiving unit receives image data on the low frequency side among a plurality of image data having different frequency bands from the low frequency side to the high frequency side constituting the hierarchically encoded image data,
The feature point detection unit starts a feature point detection process from only the low-frequency image data without waiting for reception of high-frequency image data by the reception unit. apparatus. A feature point detection method for detecting a feature point from hierarchically encoded image data,
A feature point is detected only from low-frequency side image data among a plurality of pieces of image data having different frequency bands from a low frequency side to a high frequency side constituting the hierarchically encoded image data. A feature point detection method. A reception step of receiving image data on the low frequency side among a plurality of image data having different frequency bands from the low frequency side to the high frequency side constituting the hierarchically encoded image data;
A feature point detecting step of detecting a feature point from the image data received in the receiving step;
In the feature point detection step, before receiving the high-frequency side image data of the plurality of image data, the feature point detection step is performed only from the low-frequency side image data received in the reception step. An image processing method characterized by starting a detection process. A feature point detection program characterized by causing a computer to realize a function of detecting feature points from hierarchically encoded image data,
A function of detecting feature points based only on low-frequency side image data among a plurality of image data having different frequency bands from the low frequency side to the high frequency side constituting the hierarchically encoded image data A feature point detection program characterized by being realized by a computer. A reception step of receiving image data on the low frequency side among a plurality of image data having different frequency bands from the low frequency side to the high frequency side constituting the hierarchically encoded image data;
A feature point detecting step of detecting a feature point from the image data received in the receiving step;
In the feature point detection step, before receiving the high-frequency side image data of the plurality of image data, the feature point detection step is performed only from the low-frequency side image data received in the reception step. An image processing program for starting a detection process.

Images