JP5168185B2 - Image processing method, image processing apparatus, and watermark detection system - Google Patents

Description

  The present invention relates to the field of image processing, and in particular, searches for or specifies a common pattern whose shape, color, position, etc. are the same in any of the plurality of document images from a plurality of images to be processed. The technology used for

  As computer technology and digital technology advance, there is an increasing demand for finding a common pattern between a plurality of images. For example, for the purposes of various types such as signage and copyright protection for documents, numbers, letters, figures, etc. are used on the background of many Microsoft (registered trademark) Office Word documents or PowerPoint (registered trademark) documents. The watermark is embedded. However, subsequent processing is required for paper documents obtained by printing electronic documents. For example, when copying or scanning, often extract a watermark from a document image and authenticate against the extracted watermark and / or keep only the body part to ensure the integrity of the document Thus, it is required to remove the watermark from the document image.

  In addition, when photographing or scanning a scene having a large size or a wide range with a device such as a digital camera or a scanner, images of the target or scene cannot be acquired at a time. Or scan the scene continuously from many angles to acquire multiple images, find common parts between the multiple images, and based on them, It is often necessary to stitch together images. In addition, finding a common pattern between a plurality of images has many potential applications.

  For this reason, many methods for finding a common pattern from an image have been proposed. For example, Non-Patent Document 1 discloses a method of extracting a common confidential pattern from a plurality of document images based on the identity of pattern colors. Here, the colors are first classified for each document image, the first document image is used as a reference image, the other document images are combined with each other in the classification for each color, all the images are accumulated, and Based on the color identity of the common pattern, a composite image having the highest overlay probability is specified as the common pattern.

  In the prior art, many methods and systems that are used to realize the stitching of images have also been proposed. For example, in Patent Literature 1 and Patent Literature 2, it is used for joining or synthesizing a plurality of partial images based on the calculated amount of overlap between two partial images. A method and apparatus is disclosed.

  However, in various conventionally proposed methods and apparatuses, two images are processed for each of a plurality of images to be processed, or two images are used as a reference image. Processing is performed on the above images, but in any case, the relevance between a plurality of images to be processed is not considered.

  Further, the case where the common pattern between images is deteriorated to some extent is not considered. Actually, a situation in which the common pattern always deteriorates appears in a plurality of images to be processed. For example, a common pattern such as a watermark pattern in each document image may have a difference in position, angle, and / or size due to an error when performing processing such as scanning or copying on the document image. The omission of the watermark image due to obstruction by the body part of the document, and the common part (that is, the common pattern) of the plurality of images to be stitched together is missing or defocused due to causes such as occlusion or defocusing. Clouding occurs and similar situations.

  FIG. 1 shows an example in which a watermark pattern is present in six document images. As shown in this figure, the contents of the same watermark are included in all of the six document images, but since they are obstructed by the body part, any one image has the complete watermark character string “ CONFIDENTIAL ”is not included. Thus, in a situation where a certain degree of deterioration has occurred in the common pattern, none of the common patterns can be satisfactorily searched from a plurality of images even if various conventional methods and apparatuses are used.

  Therefore, a technique for finding or identifying a common pattern in a plurality of (three or more) images to be processed relatively accurately and / or satisfactorily, that is, the above-described conventional technique. There is an urgent need for a technique that can overcome the defects and obtain a satisfactory result even in a situation where the deterioration of the common pattern occurs due to various causes.

US Pat. No. 6,690,482 US Pat. No. 7,145,596

Yusaku Fujii, Hiroaki Takebe, Katsuhito Fujimoto, Satoshi Naoi, "Secret pattern extraction from document images based on color uniformity between documents", IEICE Technical Report SIS2006-81, pp.1-5, 2007

  In the following text, a brief summary description is provided in this application to provide a basic understanding of some aspects, but this summary description is to be understood as an example. It is not intended to identify key points or important parts of the present application, nor is it intended to limit the scope of the present application. Its purpose is merely to provide some concepts according to the present application in a simplified form which are the prelude to the more detailed description that is presented later.

  One of the purposes of the technology disclosed in the present application for the above-described problems existing in the conventional technology is to search for a common pattern from among three or more images to be processed, Another object is to provide an image processing method and apparatus used for specifying. By considering the relationship between multiple images to be processed, it is possible to find the common pattern comparatively reliably and accurately from multiple images even when the common pattern is degraded. Can be obtained.

  Another object of the technology disclosed in the present application is to specify, as a reference image, one image that is the best for matching three or more images to be processed with other images. It is an object of the present invention to provide a method and apparatus used to do this.

  Still another object of the technique disclosed in the present application is to provide a method and an apparatus used for calculating an average similarity of three or more images to be processed.

  Another object of the technology disclosed in the present application is to provide a computer-readable storage medium in which a program code is stored. When the program code is executed by a computer, the computer stores the program code in the computer. Any one of the methods described above is performed.

  Still another object of the technology disclosed in the present application is to provide a watermark detection system used to extract a watermark from three or more document images.

  In order to achieve the above object, an image processing method used by an image processing apparatus disclosed in the present application is, in one aspect, from N images to be processed, an image including a common pattern in the N images. (N is a natural number of 3 or more). In the image processing method, image feature extraction is performed on N images, and the N images are divided into the C layers based on the result of the feature extraction (where C is equal to or greater than 2). A natural number), a step of calculating an average similarity of N images in each layer, and an image having an average prediction error smaller than a preset threshold in N images in the layer having the highest average similarity And specifying a composite image obtained by combining N images in the hierarchy based on the reference image as an image including a common pattern.

  In one embodiment, the image processing apparatus disclosed in the present application specifies an image including a common pattern in the N images from N images to be processed (where N is a natural number of 3 or more). Is). The image processing apparatus performs image feature extraction on N images in order to concentrate the image of the common pattern on any one of the C layers, and based on the result of the feature extraction An image feature extraction unit that divides N images into the C layers (where C is a natural number of 2 or more), and an average similarity calculation unit that calculates the average similarity of N images in each layer A reference image specifying unit that specifies, as a reference image, an image having an average prediction error smaller than a preset threshold value in N images in the layer having the highest average similarity, and N images in the layer based on the reference image And an image synthesis unit that identifies the synthesized image as an image including a common pattern.

  In one aspect, the watermark detection system disclosed in the present application specifies an image including a watermark embedded in the N document images from N images to be processed (where N is 3 or more). Is a natural number). The watermark detection system performs document image feature extraction on N document images, and divides the N document images into the C hierarchies based on the feature extraction results (here, , C is a natural number of 2 or more), an average similarity calculation unit for calculating an average similarity of the N document images in each layer, and an average prediction in N images in the layer having the maximum average similarity A reference image specifying unit that specifies a document image whose error is smaller than a preset threshold as a reference image, and N document images in the hierarchy based on the reference image are combined, and the combined image is combined into a common pattern. An image compositing unit that identifies the image to include.

  In the technique disclosed in the present application, in the process of specifying the reference image and / or calculating the average similarity, the relationship between a plurality of images is taken into account. For this reason, parameters of the average prediction error and the prediction accuracy probability are used. (The specific meaning and calculation method will be described in detail later). Accordingly, even when the common pattern shown in FIG. 1 is missing or blurred, the common pattern can be searched for accurately and reliably.

  Another advantage of the technique disclosed in the present application is that the present technique can be used not only for processing gray-level images but also for processing color images.

  Another advantage of the technique disclosed in the present application is that it can be used in many practical applications such as stitching together a plurality of images, such as using an image processing method and apparatus according to the present technique for watermark detection on a document image. It can be used for.

FIG. 1 is a diagram illustrating an example of a gray-scale image of a plurality of documents to be processed including a watermark pattern. Here, each document image includes “CONFIDENTIAL” which is a similar watermark character string. FIG. 2 is a block diagram showing an example of a data processing system to which the image processing method and apparatus according to this embodiment can be applied. FIG. 3 is a flowchart of an image processing method 300 for searching for a common pattern (for example, a common watermark character string) from the gray-scale images of a plurality of documents to be processed shown in FIG. Is shown. FIG. 4 shows six documents in a hierarchy where a common pattern is located when edge detection is performed on the six document images shown in FIG. 1 according to the method shown in FIG. It is a figure which shows an edge image. FIG. 5 is a diagram showing translation matching parameter values obtained by performing the matching for each of the six images shown in FIG. FIG. 6 is a diagram illustrating predicted translation matching parameter values for the image 1 illustrated in FIG. 4 obtained by calculation based on the translation matching parameters illustrated in FIG. FIG. 7 is a diagram showing translation prediction error values for the image 1 shown in FIG. 4 acquired based on the values shown in FIGS. 5 and 6. FIG. 8 is a diagram showing a combined document edge image obtained by combining the edge image of the document shown in FIG. 4 based on the reference image specified according to the method shown in FIG. FIG. 9 is a diagram showing an edge image obtained by performing noise removal processing (that is, background removal) on the synthesized document edge image shown in FIG. FIG. 10 is a diagram showing a composite document edge image acquired using one document image arbitrarily selected using a conventional method as a reference image. FIG. 11 shows a similarity value between two images in the first layer after layering the six document images shown in FIG. 1 by edge strength using the method shown in FIG. It is a figure which shows the prediction accuracy probability value of each document edge image in the said hierarchy. FIG. 12 is a diagram illustrating a similarity value between two images in the second hierarchy (the hierarchy is a hierarchy including a common pattern) and a prediction accuracy probability value of each image in the hierarchy. . FIG. 13 is a diagram illustrating an image processing method 1300 according to another embodiment. It is a variation of the method 300 shown in FIG. FIG. 14 is a conceptual block diagram of an image processing apparatus 1400 according to an embodiment.

  In the following, an exemplary embodiment of the technology according to the present application will be described with reference to the drawings. For clarity and simplicity, the specification does not describe all features of actual embodiments. However, the following should be understood. That is, in the process of developing such an actual embodiment, in order to realize a specific goal of the developer, many things specified in the embodiment must be determined. And the development work is very complex and can be time consuming, but for those skilled in the art using the content disclosed here, such a development work is just a routine task. is there.

  Here, what needs to be described is an apparatus closely related to the plan according to the technology according to the present application in order to prevent the technology according to the present application from becoming obscured by detailed description more than necessary. Only the configuration and / or processing steps are shown, and other details not so relevant to the technology according to the present application are omitted.

  Those skilled in the art should understand that the elements in the drawings are merely for simplicity and clarity and are not necessarily drawn to scale. For example, the size of certain elements in the drawings has been expanded relative to other elements to contribute to an understanding of embodiments of the technology according to the present application.

  For the sake of simplicity, an example of searching for the watermark pattern common to the six images, that is, the character string “CONFIDENTIAL” from the six document images shown in FIG. An image processing method and apparatus according to the present invention will be described. However, it goes without saying that the present invention is applicable to other situations.

  FIG. 2 shows a block diagram of a data processing system 200 as an example to which the image processing method and apparatus according to the present embodiment can be applied.

  As shown in FIG. 2, the data processing system 200 may be a symmetric multiprocessor (SMP) system that includes a plurality of processors 202 and 204 connected to a system bus 206. However, as a choice, a single processor system (not shown) may be employed. A memory controller / high-speed buffer 208 is also connected to the system bus 206 and is used to provide an interface with the local memory 209. An I / O bus bridge 210 is connected to the system bus 206 and provides an interface with the I / O bus 212. The memory controller / fast buffer 208 and the I / O bus bridge 210 may be integrated together as described. A Peripheral Component Interconnect (PCI) bus bridge 214 connected to the I / O bus 212 provides an interface with the PCI local bus 216. The modem 218 and the network adapter 220 can be connected to the PCI local bus 216. A typical PCI bus implementation can support four PCI expansion slots or insertable connectors. Added PCI bus bridges 222 and 224 provide interfaces for added PCI local buses 226 and 228. This makes it possible to support additional modems or network adapters. In accordance with this method, the data processing system 200 is connected to many peripheral devices, such as network computers. The graphics adapter 230 and hard disk 232 to which the memory maps can be connected directly or indirectly to the I / O bus 212 as described in the figure.

  The image processing apparatus according to the technology according to the present application can be integrated in, for example, the processors 202 and 204 shown in FIG. 2, or peripheral devices are connected to the data processing system 200 via an I / O bus. Can do.

  Those skilled in the art should understand that the hardware described in FIG. 2 can be modified. For example, other peripheral equipment such as an optical disk drive or the like can be used in addition to or instead of the hardware described. The example described in FIG. 2 does not limit the configuration of a system to which the technology according to the present application can be applied.

  FIG. 3 shows N document images (for example, six document gray scales shown in FIG. 1) of N images (where N is a natural number of 3 or more) in one embodiment according to the technology of the present application. A flowchart of an image processing method 300 for finding a common pattern in the image (for example, a common watermark character string) is shown.

  As shown in FIG. 3, the method 300 begins in step S305, and in step S310, processes all N document images so as to extract features for each image. In the prior art, many methods can be used for extracting features from a document image. In the method used here, first, all edges in all N document images are extracted using the CANNY operator, and then the magnitude of the edge strength of each edge point is calculated. Here, the CANNY operator is a common edge detection operator suitable for processing gray-scale images. For more details, see JF Canny, "A computational approach to edge detection", IEEE trans. PAMI, 8 (6), pp. 679-698, 1986. For more details, refer to http://www.pages.drexel.edu/~weg22/can_tut.html.

Subsequently, in step S315, the hierarchization is performed on the N images based on the magnitudes of the edge strengths of all the edge points of the N images calculated and acquired in step S310. Assuming that N images are divided into C hierarchies, each document image I _i (i = 1, 2,..., N) is divided into C document edge images (the first image is the first image). To C). In other words, after dividing N images into C layers, a total of N × C document edge images can be acquired, and each layer has an edge of N documents. Has an image. Subsequent processing such as copying or scanning changes the parameters such as gray level or chromatic aberration of different document images, and even if they are different from each other, the edge intensities of the common patterns in the different document images are the same. (Both strengthened or weakened together). That is, the edge strengths of common patterns in different document images match each other. Therefore, when layering N images, the edge of the common pattern appears in one of the C layers almost at the same time.

  FIG. 4 shows a case where feature extraction (that is, edge detection) is performed on the six document images shown in FIG. 1 according to the above-described method and then divided into three layers (that is, C = 3). 6 shows six document edge images in one layer where the common pattern is located. FIG. 4 shows that any of the common patterns in the six document images shown in FIG. 1, that is, the “CONFIDENTIAL” edge of the common character string appears in the hierarchy.

  Referring back to FIG. 3, in step S320 to S345, the method 300 performs processing on N document edge images in each layer (representing the currently processed layer using l) from the first layer. Starting from this, one image (which may be referred to as an image with the highest reliability) that is most excellent in matching with the other N-1 images is searched for as a reference image. Then, the N−1 images and the reference image are combined to obtain a combined edge image.

を計算する。 Specifically, as shown in FIG. 3, in step S320, average prediction is performed for each image (represented using Ii) in the first hierarchy (where 1 is a natural number of 0 <1 <C + 1). error

Calculate

の計算過程は、以下の通りである。まず、Ｎ枚の文書エッジ画像に対して、２枚ずつのマッチングを行い、２枚ずつのマッチングパラメータを取得する。２枚ずつの画像間の差異は、並進、回転及び／又は拡大縮小変換から見積もることができると仮定する。このようにして、第i枚と第j枚の２枚の画像間のマッチングパラメータM_ij (P_t, P_r, P_s)を計算することができ、ここで、P_t, P_rと P_sは、それぞれ、並進、回転及び拡大縮小変換に対応するパラメータであり、iとjはいずれも１とＮの間（二つの端点を含む）に介在している自然数であり、かつ、i≠jである。 Average prediction error

The calculation process of is as follows. First, two document matching is performed on N document edge images, and two matching parameters are obtained. It is assumed that the difference between every two images can be estimated from translation, rotation and / or scaling conversion. In this way, the matching parameter M _ij (P _t , P _r , P _s ) between the i-th and j-th images can be calculated, where P _t , P _r and P _s is a parameter corresponding to translation, rotation, and scaling conversion, i and j are natural numbers between 1 and N (including two endpoints), and i ≠ j.

Here, the matching parameter between the two images can be calculated using any one conventional method. For example, B. Srinivasa Reddy and BN Chatterji, "An FFT-Based Technique for Translation, Rotation and Scale-Invariant Image Registration", IEEE Transactions on Image Processing, Vol. 5, No. 8, 1996, pp. 1266-1271. The matching parameter M _ij can be calculated using the method disclosed in.

As for N images, N × (N−1) / 2 matching parameters can be calculated through matching of each two images. It goes without saying that N × (N−1) / 2 matching parameters are redundant if correct matching parameters can be calculated after matching two images. For each document image, the other N-1 images and two matching parameters can be calculated, and the N-1 matching parameters are used to calculate the other N-1 The matching parameter between the images can be predicted. For example, a matching parameter M ₁₂ between the first sheet and the second images, the first sheet and by using the matching parameter M ₁₃ between the three images, and a second sheet to the first sheet of image No. the value of the matching parameter M ₂₃ between the three images (expressed using the M ₂₃ ^1e) can be predicted. That is, based on the matching parameter M _mi between the m-th image and the i-th image and the matching parameter M _mj between the m-th image and the j-th image, the i-th and j-th images for the m-th image. The value M _ij ^me of the matching parameter M _ij between the images can be predicted. Here, m is a natural number (including two end points) interposed between 1 and N, and m ≠ i ≠ j.

In practice, as shown in FIG. 1, all common patterns in many document images are missing. Accordingly, there is a certain error between the actually calculated two matching parameters and the predicted two matching parameters. That is, a matching parameter M _ij calculated by matching two images between the i-th image and the j-th image and a matching parameter M _ij ^me predicted based on the matching parameters M _mi and M _mj . There is a certain error between them. Hereinafter, this error is referred to as “prediction error between the i-th image and the j-th image with respect to the m-th image” and is represented by ε _ij ^m (P _t , P _r , P _s ).

  As described above, the prediction error between the other N-1 images can be acquired for each one image, so that the prediction error matrix in translation, rotation, and / or scaling for the image can be obtained. Each will be able to get.

を計算する。ここで、従来の技術における任意の一つの知られている計算方法を用いて、１枚の画像の平均予測誤差

を取得することができる（後でさらに詳細に説明する）。 Then, for each image, based on the prediction error matrix for each of translation, rotation, and / or scaling for the image, the three factors of translation, rotation, and scaling are considered together, Average prediction error of image

Calculate Here, an average prediction error of one image using any one known calculation method in the prior art

Can be obtained (discussed in more detail later).

  As shown in FIG. 3, when the average prediction error of all N images in the current layer is acquired in step S320, the process flow of the method 300 proceeds to step S325, where the average prediction error in the layer is the smallest. An image is specified as a reference image. Here, the image with the smallest average prediction error is the image that is most excellent in matching with other N-1 images among the N images, that is, the image with the highest reliability.

  Here, an image having the smallest average prediction error in N images (that is, an image having the highest matching with other N-1 images in N images and having the highest reliability) However, a person skilled in the art will recognize that the object of the present invention can be realized in the same manner by using another appropriate image (for example, an image having the second best matching) as the reference image. Should be understood. For example, when the calculated average prediction errors of N images are arranged in ascending order, the first 1 / n (n is a natural number greater than 0 and less than or equal to N, and the value of n can be set by experience. ) May be recognized as one image excellent in matching with other N-1 images (or may be referred to as a reliable image). ). Alternatively, one threshold value is set in advance for the average prediction error based on experience, and any image whose average prediction error is smaller than the threshold value is an image excellent in matching with the other N−1 images (that is, Thus, an image excellent in matching with the other N-1 images (that is, a reliable image) is specified as a reference image. It will be.

  Subsequently, in step S330, translation, rotation, and / or enlargement is performed on the basis of the reference image using two matching parameters of the reference image calculated previously and the other N−1 images. By performing the reduction conversion, the other N−1 images are converted so as to have the same position, angle, and size as the reference image (that is, the other N−1 images are matched with the reference image). . Then, the combined N images are combined to obtain a combined document edge image.

  Here, N images are synthesized using an arbitrary method known in the prior art. For example, as a relatively simple method, when N images after conversion and combination are accumulated for each pixel point, the numerical value at each pixel point is the edge in all the superimposed images at that pixel point. The total number of points. Then, for display, there is one that obtains a synthesized image by linearly transforming the obtained numerical values of 0 to N at each pixel point (for example, the numerical value of 0 to N is converted to a gray scale of 0 to 255). Linearly converted to key values to obtain a composite gray-scale image). For example, P. Shivakumara and G. Hemantha Kumar and DS Guru and P. Nagabhushan, "Sliding window based approach for document image mosaicing", Image Vision Comput., Vol. 24, No. 1, pp.94-100, 2006 and Anthony Zappal and Andrew H. Gee and Michael Taylor, "Document mosaicing", Image Vision Comput., Vol. 17, No. 8, 1999, pp.589-595. Synthesis may be performed.

For the sake of simplicity, here, the calculation method of the average prediction error will be described in more detail with reference to FIGS. That is, here, the rotation parameter P _r and the enlargement / reduction parameter P _{s in} the matching parameters M _ij (P _t , P _r , P _s ) of every two images in each layer are both 0. Then, the matching parameter can be simplified to M _ij (x, y). Here, the values of x and y represent translation amounts in the x and y directions, respectively.

図７には、図５と図６に示している値に基づいて取得した、画像１に対する並進予測誤差ε_ij ¹(x, y)の値が示されている。ここで

図５〜７に示すように、（NA, NA）或いはN/Aは無効な値を表し、それらの値に対しては計算する必要がないことを示す。 FIG. 5 shows values of translation matching parameters M _ij (x, y) obtained by performing calculation by performing matching for each of the six edge images shown in FIG. 4. . FIG. 6 shows values of predicted translation matching parameters for image 1 obtained by calculation based on the translation matching parameters shown in FIG. 5 according to the method described above. here

FIG. 7 shows the value of the translation prediction error ε _ij ¹ (x, y) for the image 1 obtained based on the values shown in FIGS. 5 and 6. here

As shown in FIGS. 5 to 7, (NA, NA) or N / A represents an invalid value and indicates that it is not necessary to calculate for these values.

を求めることができる。ここで使用された比較的簡単な方法は、図７に示す、全ての有効な予測誤差値に対してxとy方向の総平均値を求めることである。具体的にいうと、図７に示したマトリックスにおいて、有効な２０個の位置でのxとyの値をそれぞれ加算して得られた和sum(x)とsum(y)に対して、総平均値を求める。即ち

従って、求められた画像１の平均予測誤差は１．０５となる。このようにして、同様な方法に従って、全てのＮ枚の画像の平均予測誤差を取得することができる。 As described above, the average prediction error of the image 1 is calculated by calculating the translation prediction error ε _ij ¹ (x, y) shown in FIG. 7 using any known method.

Can be requested. The relatively simple method used here is to obtain the total average value in the x and y directions for all valid prediction error values shown in FIG. More specifically, in the matrix shown in FIG. 7, the sums sum (x) and sum (y) obtained by adding the values of x and y at 20 valid positions are summed up. Find the average value. That is

Therefore, the average prediction error of the obtained image 1 is 1.05. In this way, the average prediction error of all N images can be acquired according to a similar method.

  As described above, taking only translation conversion as an example, based on FIGS. 5 to 7, how to calculate the average prediction error of one image has been described. However, those skilled in the art can translate, rotate and / or enlarge at the same time. When considering reduction, it is possible to easily conceive how to calculate the average prediction error of one image. For example, as a relatively simple method, the average translation prediction error, the average rotation prediction error, and the average enlargement / reduction prediction error of each image are calculated according to the above-described method, and then these three error values are calculated. Some perform a weighted average to obtain an average prediction error of the image.

  FIG. 8 shows a combined document edge image obtained by combining the edge image of the document shown in FIG. 4 based on the reference image specified according to the above method. FIG. 8 shows that in addition to the common pattern, there is always some noise in the synthesized document edge image.

  Referring back to FIG. 3, in order to further remove the influence of noise and obtain an ideal result, as shown in FIG. 3, in step S <b> 335, with respect to the synthesized document edge image acquired in step S <b> 330, Noise removal processing is performed.

  For example, as described above, when the images are synthesized by accumulating the converted N images for each pixel point, any one pixel in the synthesized document edge image is used. If the numerical value of the point is smaller than a certain threshold value T, it can be seen that the number of edge points superimposed at the position of the pixel point is not sufficient. Therefore, the pixel point is recognized as a noise point, and the numerical value of the point is set to 0 (that is, placed in the background). FIG. 9 shows an edge image obtained after performing noise removal processing (that is, removing the background) on the synthesized document edge image in the hierarchy where the common pattern shown in FIG. 8 is located. It goes without saying that other noise removal methods known in the prior art may be used.

  FIG. 10 shows a common image obtained by selecting any one document image (a document image with the smallest average prediction error) as a reference image according to the method disclosed in Non-Patent Document 1. A composite document edge image including a pattern (here, noise removal processing has already been performed) is shown. Comparing the images shown in FIGS. 9 and 10, some watermark characters such as “C”, “N” and “F” in FIG. 9 are more than the result obtained using the conventional method (FIG. 10). You can easily see that it is clear.

  Further, referring back to FIG. 3, as shown in the figure, in the processing flow of the method 300, the noise removal process (that is, step S <b> 335) is performed, and then the process proceeds to step S <b> 340. The average similarity is calculated.

  As mentioned above, the present invention takes into account the relevance between multiple images. Therefore, when calculating the image similarity, a parameter called a prediction accuracy probability P is introduced and used to represent the influence of the average prediction error of the image on the similarity of the image.

ここで、

は第i枚の画像の平均予測誤差、

は予め設定された最大の平均予測誤差値である。ここで、

は、平均予測誤差のパラメータが、並進、回転及び／又は拡大縮小で取り得る値の範囲を表す。このように、計算された予測精度確率P_iの値は０と１（二つの端点を含む）間に介在する。 The relatively simple method for calculating the prediction accuracy probability P _i used here is expressed by the following equation.

here,

Is the average prediction error of the i-th image,

Is a preset maximum average prediction error value. here,

Represents the range of values that the average prediction error parameter can take in translation, rotation and / or scaling. Thus, the calculated value of the prediction accuracy probability P _i is interposed between 0 and 1 (including two end points).

ここで、CONF2(i, j)は、第i枚の画像と第j枚の画像との類似度（この２枚の画像は既に並進、回転及び／又は拡大縮小変換して互いに合わせたものである）を表す。 Using the calculated prediction accuracy probability P _i , the similarity of the i-th document image is defined as follows.

Here, CONF2 (i, j) is the similarity between the i-th image and the j-th image (the two images have already been translated, rotated and / or scaled and matched to each other). Is).

In the ideal state, since the average prediction error of the image is 0, P _i = 1 according to the equation (4). At this time, the similarity between the image calculated in the present invention and the similarity obtained according to the conventional method is obtained. Is the same, but in the non-ideal state, the average prediction error of the image is not 0, and thus P _i <1, and therefore, the prediction accuracy probability P _i determines the average prediction error of the i-th image. The influence on the similarity of images is expressed.

For binary images, a relatively simple method of calculating the similarity between two images (assuming they are already superimposed on each other) is as follows.

  In the method 300 according to the invention, any other known method may be used to calculate the similarity between two images. Furthermore, the above equation (5) may be modified as necessary.

Then, the average similarity of all the images in the current hierarchy can be defined as follows.

  Referring again to FIG. 3, when the average similarity of the current hierarchy is calculated in step S340, the process of method 300 proceeds to step S345, where all the processes for C hierarchy have been completed or not. Is judged. If it is determined in step S345 that the processing for all the C hierarchies has not yet been completed, 1 is added to l, and the process of method 300 is returned to step S320, in the next hierarchy (ie, the (l + 1) th hierarchy). The processing from step S320 to step S340 is repeated for all N images.

  As described above, in the image feature extraction process, N images are divided into layers, and each image is divided into C layers according to different image edge strengths. A common image such as a watermark is concentrated in any one of the C layers, but it is not yet known which layer is concrete. Therefore, the processing from step S320 to step S335 is performed on all of the N document edge images in each layer, and as shown in step S340, for each layer, the layer The average similarity is calculated.

  As shown in FIG. 3, if it is determined in step S345 that the processing for all C hierarchies has already been completed, the process flow of method 300 proceeds to step S350, where the average similarity is the highest. Is identified as the layer where the common pattern is located, so that the synthesized document edge image (having already been subjected to noise removal processing) having the maximum average similarity can be identified as an edge image including the common pattern. Can be found or identified.

As noted above, the similarity algorithm used in step S340 of method 300 takes into account the relevance between multiple images. That is, the influence of the similarity of the prediction error on the accuracy is considered. Therefore, compared with the conventional method, the method according to the technique of the present application can obtain a more accurate result. For example, in the case of an application for searching for a common pattern from a plurality of images as described above, if two images are accidentally very similar to each other, The similarity between images is considerably high. This may increase the average similarity of N images (when the influence of the prediction error on the accuracy of the similarity is not considered). However, if these two images do not match well with the other N-2 images, their average prediction error is quite high. Thus, according to the present invention, the prediction accuracy probability can be reduced, and the similarity CONF _i of these two images can be reduced, thereby reducing the average similarity of N images. can do.

  For example, FIG. 11 shows two sheets in the first layer after the layering is performed on the six document images shown in FIG. 1 according to the edge strength using the method 300 shown in FIG. FIG. 12 shows a second hierarchy (the hierarchy is a hierarchy including a common pattern). FIG. 12 shows a similarity value between images and a prediction accuracy probability value of each document edge image in the hierarchy. ) Shows the value of the similarity between two images and the value of the prediction accuracy probability of each image in the hierarchy.

  In the situation shown in FIG. 11, when the prediction accuracy probability is not taken into consideration, the average similarity obtained by averaging after obtaining the sum is 0.0484. However, considering the influence of the prediction accuracy probability, the similarity of the six images is 0.0032, 0.0178, 0.0334, 0.0207, 0.0298, and 0.0246, respectively, according to the above equation (5). In this way, according to the above equation (7), the average similarity of the first hierarchy shown in FIG. 11 is 0.0216.

  In the situation shown in FIG. 12, if the prediction accuracy probability is not taken into consideration, the average similarity obtained by averaging after obtaining the sum is 0.0364. However, considering the influence of the prediction accuracy probability, the similarity of the six images is 0.0299, 0.0347, 0.0334, 0.0271, 0.0315, and 0.0326, respectively, according to the above equation (5). In this way, according to the above equation (7), the average similarity of the second hierarchy shown in FIG. 12 is 0.0315.

  Therefore, if the selection is made according to the average similarity value without considering the prediction accuracy probability, the first hierarchy shown in FIG. 11 is erroneously recognized as a hierarchy including a common pattern therein. . However, if the prediction accuracy probability is considered, the average similarity of the document edge images in the hierarchy including the common pattern (that is, the second hierarchy shown in FIG. 12) is the hierarchy not including the common pattern (represented in FIG. 11). Higher than the average similarity of the document edge images in the first hierarchy). That is, since a hierarchy having a large average similarity calculated when the prediction accuracy probability is taken into consideration can be accurately specified as a hierarchy including a common pattern therein, an accurate result can be obtained.

  As described above, the image processing method according to the present invention has been described based on the flowchart shown in FIG. 3 using the six document gray-scale images shown in FIG. 1 as an example. It should be clear that it is merely exemplary and that corresponding modifications to the flow of the method shown in FIG. 3 can be made depending on the actual application and specific requirements.

  If necessary, the execution order of some steps in the method 300 shown in FIG. 3 may be adjusted, or some processing steps may be omitted or added. For example, FIG. 3 illustrates that the process of calculating the average similarity (ie, step S340) is performed after performing the image synthesis and noise removal processes (ie, steps S330 and S335). Of course, they may be performed in parallel or in reverse order.

  FIG. 13 shows an image processing method 1300 according to another embodiment of the present invention. It is a variation of the method 300 shown in FIG.

  From the flowchart of the method shown in FIGS. 3 and 13, it can be seen that the processing steps in steps S1305 to S1345 are similar to the processing steps in steps S305 to S325, S340 to S345, and S330 to S335 shown in FIG. The differences are only in the following. That is, as shown in FIG. 13, the image composition processing in step S1340 and the noise removal processing in step S1345 are executed after step S1330 of calculating the average similarity. At this time, it is not necessary to perform processing for all the layers, and synthesis and noise removal processing are performed on N images only in the layer having the highest average similarity (the layer in which the common pattern is located). Step S350 is omitted. Thereby, the amount of calculation can be reduced as compared with the method shown in FIG. In order to prevent duplication, a specific processing process in each step shown in FIG. 13 is not described here again.

  Of course, other modifications may be made to the method 300 shown in FIG. 3 or the method 1300 shown in FIG. 13, for example, step S1325 shown in FIG. 13 is executed between step S1335 and step S1340. May be. A person skilled in the art can easily draw a completely corresponding flow chart, but for simplicity it will not be described in detail here.

  Then, the image feature extraction process, the edge detection process, the process of layering the image, the process of calculating the average prediction error of the image using the matching parameters of each image, Processing of image synthesis, image noise removal processing, processing for calculating the similarity between two images, and using the prediction accuracy probability, the similarity between one image and other images is calculated. The calculation processing and the like can of course be performed using any one of known techniques, and is not limited to any of the specific methods described above.

  In the above, the document gray gradation image is taken as an example to describe the image processing method according to the technique according to the present application. However, the method is not limited to the processing for the document image, and any gray gradation is possible. The present invention can be applied to image processing, and can also be applied to color image processing. For example, for an application for searching for a common pattern from a plurality of color images, before the step S310 shown in FIG. 3, after converting a plurality of color images into a gray gradation image by color → gray conversion, Processing is performed using the method shown in FIG. Alternatively, feature extraction is performed directly from the color image in step S310 or step S1310. For example, the edge information is directly extracted from the color image and the edge strength is calculated (for example, the document "Color edge detection by color information segmentation" published by Kagehide Hideki et al., Computer Application, No. 8, 2001. Can be referred).

  In the above description, the processing related to image feature extraction and layering in the image processing method according to the technique according to the present application has been described using the example of extracting edge information and calculating the edge strength. However, the technique according to the present application is applicable to various known image feature extraction methods and methods for layering images according to feature extraction results. Any method may be used as long as the common pattern image is divided into almost the same hierarchy according to the result of feature extraction. For example, similar results are similarly obtained when layering is performed according to the color of a common pattern (for a color image) or the value of pixel gray gradation (for a gray gradation image) instead of the magnitude of edge strength. be able to. For example, after performing feature extraction on a color image or gray-scale image to be processed, the color or gray-scale is based on the assumption that the colors of the common pattern or the pixel values of the gray-scale are consistent with each other. By performing layering according to gradation pixel values, it is also possible to divide the image of the common pattern into almost the same layer.

  And when describing the method by the technique which concerns on this application above, the concept of an average prediction error and prediction accuracy probability was introduced, and those calculation methods were also given. A person skilled in the art who uses the contents disclosed in the present application can expand the above-described concept and calculation method as necessary, but each of them will not be described in detail here.

  In the above description, only an application for searching for a common pattern such as a watermark from a plurality of document images has been described. However, the image processing method 300 shown in FIG. 3 or the method 1300 shown in FIG. You may apply to the application which connects. In a normal case, a plurality of images to be joined may change in size (enlargement / reduction) and angle (rotation) in addition to translation, and similar perspective deformation or curvature deformation This situation may occur. In such a case, before searching for a common pattern using the method 300 or 1300 described above, a pre-processing portion is required, so that any of the plurality of images has the same size, angle, and deformation coefficient. Or a common pattern in a plurality of images is made to coincide in a high-dimensional parameter space composed of position, size, angle, deformation coefficient, and the like. When a common pattern in a plurality of images to be joined is searched for, a common “one origin” in the plurality of images is found. Then, by specifying the relative positions of a plurality of images to be joined together and performing composition by joining the plurality of images, joining of the images can be realized. . There are a large number of currently known methods for image stitching techniques. For example, joining may be performed using the method disclosed in Patent Document 1 or Patent Document 2. Of course, other methods may be used. For simplicity, each one is not described in detail.

  A person skilled in the art will be able to select a part of the process in the image processing method described based on FIGS. 3 and 13, for example, one image that best matches each of the other images from a plurality of images. Of course, the process used to find the reference image and the process used to calculate the average similarity for multiple images when considering the relevance between multiple images. It should be understood that the present invention can be used for various applications other than the image processing shown in the embodiments of the present application as necessary.

  In the following, based on FIG. 14, an image processing apparatus used to search for or identify a common pattern from a plurality of images to be processed according to one embodiment of the present invention will be described.

  FIG. 14 is a conceptual block diagram of an image processing apparatus 1400 according to one embodiment of the present invention. The image processing apparatus 1400 can use the method 300 shown in FIG. 3 or the method 1300 shown in FIG. As shown in FIG. 14, the image processing apparatus 1400 includes an image feature extraction unit 1410, a reference image specifying unit 1420, an average similarity calculation unit 1430, an image composition unit 1440 and a noise removal unit 1450.

  In this case, the image feature extraction unit 1410 performs N-level division on a plurality of received images (for example, N images) to be processed (can be gray-scale images or color images). It is used to extract image features from the images. For example, as described above, the image feature extraction unit 1410 extracts all edges in the image using an edge detection operator, calculates the edge strength of each edge point, and calculates the edge strength. N images are divided into C hierarchies based on the size, and N × C images are acquired.

  The reference image specifying unit 1420 uses a method described above from N images (for example, edge images) in each layer acquired by the image feature extraction unit 1410 to obtain one appropriate image ( This is used to specify the image having the best matching with the remaining N-1 images, specifically, for example, the image with the smallest average prediction error) as the reference image of the hierarchy.

  The average similarity calculation unit 1430 considers the influence of the prediction error on the accuracy of the accuracy according to the method described above (for example, calculates the prediction accuracy probability using the average prediction error of each image). Used to calculate the average similarity of all images in each hierarchy.

  The image composition unit 1440 performs translation, rotation, and / or enlargement / reduction conversion on all the images in one layer on the basis of the reference image, thereby obtaining N−1 images other than the reference image. Used to synthesize N images according to the reference image (for example, in the case of an edge image, the numerical value is accumulated for each pixel point, and the numerical value at each pixel point of the synthesized image is Is the total number of all superimposed edge points in). Here, the image synthesizing unit 1440 may synthesize the respective images of the respective layers, or in order to simplify the calculation, the image synthesizing unit 1440 may calculate the average similarity according to the calculation result of the average similarity calculating unit. Only the image of the highest level may be synthesized.

  The noise removal unit 1450 is used to perform a noise removal process on the image synthesized by the image synthesis unit 1440 in order to remove more than necessary noise present in the synthesized image.

  In view of the detailed description of the specific processing steps already described above, the specific processing steps of the units will not be described in detail again to avoid duplication.

  What needs to be described here is that the configuration of the image processing apparatus 1400 shown in FIG. 14 is merely an example, and those skilled in the art can use the block diagram of the configuration shown in FIG. It may be corrected. For example, if the quality of the synthesized image of the image synthesis unit 1440 can satisfy a predetermined requirement, the noise removal unit 1450 can be omitted. When a plurality of images to be processed are color images, a color-gray gradation conversion unit may be added before the image feature extraction unit 1410. It is used to convert N color images into N gray tone images by color-gray tone conversion.

  As mentioned above, the image processing methods 300 and 1300 and the image processing apparatus 1400 according to the technique according to the present application can be applied to the general-purpose data processing system shown in FIG. However, the image processing method and apparatus according to the technique according to the present application may of course be used for a system or device different from that shown in FIG. For example, they can be applied to devices such as scanners, copiers or multi-function integrated machines. Then, the device such as the scanner, the copier or the multi-function integrated device can extract the watermark embedded in a plurality of document images, manage the document, and manage the confidential information inside the company. It can be used to monitor and alert against copying or duplicating simple documents.

  Of course, each operation process in the method according to the technique according to the present application can also be realized by a computer-executable program stored in various storage media readable by the device.

  And the objective of the technique which concerns on this application can also be achieved with the following system. That is, a storage medium storing the code of the executable program is directly or indirectly provided to a system or device, and a computer or a central processing unit (CPU) in the system or device receives the code of the program. Is read and executed.

  In this case, as long as the system or device has a function of executing the program, the implementation method of the technology according to the present application is not limited to the program, and the program may be in any format. For example, an object program, a program executed by an interpreter, or a scenario program provided to an operation system.

  The storage media readable by these devices include various memories, memory cells, semiconductor devices, for example, disk units such as optical, magnetic and magneto-optical disks, and other media suitable for storing information. However, it is not limited to them.

  The technology according to the present application can also be obtained by connecting a client computer to a corresponding website on the Internet, downloading the computer program code according to the technology according to the present application to the computer, and executing the program after being installed. Can be realized.

  Finally, it is necessary to explain that in this application the terms first and second, for example, are used to distinguish one entity or operation from another entity or operation. It does not require or imply that there is any such actual relationship or order between entities or operations. And “have”, “include”, or any other variation, covers non-exclusive “include”. Thus, a process, method, article, or device that includes a series of elements includes not only those elements, but also other elements not explicitly described, or such processes, methods, articles, or devices Includes unique elements. In the absence of more restrictions, an element that is limited by the phrase “including one” is that there are other similar elements in the process, method, article or device that include the element. Are not excluded.

  As mentioned above, although the Example of the technique which concerns on this application was described in detail based on drawing, it should be understood that said embodiment only demonstrates the technique which concerns on this application, and the technique which concerns on this application It is not a restriction on Those skilled in the art can make various modifications and changes to the above-described embodiment without departing from the spirit and scope of the technology according to the present application. Therefore, the scope of the technology according to the present application is limited only by the appended claims and their equivalent meanings.

(Supplementary Note 1) An image processing method in which an image processing apparatus specifies an image including a common pattern in N images from N images to be processed (where N is a natural number of 3 or more). And
Image feature extraction is performed on N images, and N × C images are obtained by dividing the N images into C hierarchies based on the result of the feature extraction (where C is 2 or more natural numbers)
Calculating an average similarity of the N images in each hierarchy;
A composite image obtained by combining, as a reference image, an image having an average prediction error smaller than a preset threshold value in N images in the hierarchy having the maximum average similarity, and synthesizing N images in the hierarchy based on the reference image. Identifying as an image containing a common pattern;
Is executed by the image processing apparatus.

(Supplementary note 2) The method further includes the step of calculating the average prediction error for each of the N images in the hierarchy,
Calculating the average prediction error comprises:
Based on matching parameters between two images of the image and the remaining N−1 images, a prediction matching parameter between two images of the remaining N−1 images for the image is calculated. Obtaining a predicted matching parameter matrix for the image;
Calculating the average prediction error of the image based on the prediction matching parameter matrix for the image;
The image processing method according to claim 1, further comprising:

(Supplementary Note 3) The step of calculating the average similarity of N images in each layer includes:
Calculating a prediction accuracy probability of each image based on the average prediction error of each image;
Calculating the degree of similarity of each image using the prediction accuracy probability of each image based on the degree of similarity between two images of each image and the remaining N−1 images;
Calculating the average similarity of N images based on the similarity of each image;
The image processing method according to claim 2, further comprising:

(Supplementary Note 4) The step of performing feature extraction of an image and obtaining N × C images by dividing the N images into the C layers based on the result of the feature extraction includes:
Extracting all edges in the N images using an edge detection operator;
Calculating the magnitude of the edge strength at each edge point;
Dividing the N images into the C layers based on the calculated magnitude of the edge strength;
The image processing method according to any one of appendices 1 to 3, further comprising:

(Supplementary note 5) The image processing method according to supplementary note 4, wherein the N images to be processed are gray gradation images, and the edge detection operator is a CANNY operator.

(Supplementary note 6) The image processing method according to supplementary note 4, wherein the N images to be processed are color images, and the edge detection operator extracts edge information from the color image.

(Supplementary note 7) The N images to be processed are color images, and further include a step of obtaining N gray gradation images by color-gray gradation conversion before performing image feature extraction, The image processing method according to appendix 4, wherein the edge detection operator is a CANNY operator.

(Supplementary Note 8) The method further includes the step of acquiring the composite image, and the step of acquiring the composite image includes:
Aligning the remaining N-1 images in the hierarchy with the reference image;
By accumulating the combined N images for each pixel point, the total number of all overlaid edge points at the pixel point is calculated, and the synthesized edge image is acquired as the synthesized image based on the total number. Steps,
The image processing method according to appendix 4, characterized by comprising:

(Supplementary note 9) The method further includes the step of performing noise removal on the composite image, wherein the step of specifying the composite image as an image including a common pattern specifies the composite image after noise removal as an image including a common pattern. The image processing method according to any one of appendices 1 to 3, wherein:

(Supplementary note 10) The image according to any one of supplementary notes 1 to 3, wherein the N images to be processed are document images, and the common pattern is a watermark embedded in the document image. Processing method.

(Supplementary note 11) The image processing method according to Supplementary note 2 or 3, wherein the matching parameter between two images each includes a matching parameter relating to translation, rotation, and / or scaling conversion.

(Supplementary Note 12) An image processing apparatus that identifies an image including a common pattern in N images to be processed (N is a natural number of 3 or more) from N images to be processed,
An image feature extraction unit (here, N × C images are obtained by performing image feature extraction on N images and dividing the N images into C layers based on the result of the feature extraction). , C is a natural number of 2 or more)
An average similarity calculator for calculating an average similarity of the N images in each layer;
A reference image specifying unit that specifies, as a reference image, an image having an average prediction error smaller than a preset threshold in N images in the hierarchy having the maximum average similarity;
An image combining unit that combines N images in the hierarchy based on the reference image, and identifies the combined image as an image including a common pattern;
An image processing apparatus comprising:

(Supplementary Note 13) The reference image specifying unit includes:
Two of the remaining N-1 images for each image based on matching parameters between each of the N images in the one layer and the remaining N-1 images. A prediction parameter matrix acquisition unit that calculates a prediction matching parameter between each image and acquires a prediction matching parameter matrix for each image;
An average prediction error calculation unit that calculates an average prediction error of each image based on the prediction matching parameter matrix for each image;
The image processing apparatus according to appendix 12, characterized by comprising:

(Supplementary Note 14) The average similarity calculator
A prediction accuracy probability calculation unit that calculates a prediction accuracy probability of each image based on the average prediction error of each image in N images;
Similarity that calculates the similarity of each image using the prediction accuracy probability of each image based on the similarity between each of the N images and the remaining N−1 images. A degree calculator,
With
14. The image processing apparatus according to appendix 13, wherein the average similarity of N images is calculated based on the similarity of each image in N images.

(Supplementary Note 15) The image feature extraction unit
An edge extraction unit that extracts all edges in the N images using an edge detection operator;
An edge strength calculator that calculates the magnitude of the edge strength at each edge point;
With
The image processing apparatus according to any one of appendices 12 to 14, wherein the N images are divided into the C layers based on the calculated magnitude of the edge intensity.

(Supplementary note 16) The image processing apparatus according to supplementary note 15, wherein the N images to be processed are gray gradation images, and the edge detection operator is a CANNY operator.

(Supplementary note 17) The image processing apparatus according to supplementary note 15, wherein the N images to be processed are color images, and the edge detection operator extracts edge information directly from the color image.

(Supplementary Note 18) The N images to be processed are color images,
The image processing apparatus further includes a color-gray gradation conversion unit that converts N color images into N gray gradation images by color-gray gradation conversion,
The image processing apparatus according to appendix 15, wherein the edge detection operator is a CANNY operator.

(Supplementary Note 19) The image composition unit combines the remaining N−1 images and the reference image in the hierarchy, and accumulates the combined N images for each pixel point, whereby all the image points at the pixel point are accumulated. 16. The image processing apparatus according to appendix 15, wherein the total number of superimposed edge points is calculated, and a combined edge image is acquired as the combined image based on the total number.

(Additional remark 20) It further has a noise removal part which performs a noise removal process with respect to the synthesized image synthesize | combined by the said image synthetic | combination part, The said image synthetic | combination part specifies the synthesized image after noise removal as an image containing the said common pattern The image processing device according to any one of appendices 12 to 14, characterized in that:

(Supplementary note 21) In any one of Supplementary notes 12 to 14, the N images to be processed are document images, and the common pattern is a watermark embedded in the document image. Image processing device.

(Supplementary note 22) The image processing apparatus according to supplementary note 13 or 14, wherein the matching parameter between two images each includes a matching parameter relating to translation, rotation, and / or scaling conversion.

(Supplementary note 23) A watermark detection system for identifying an image including a watermark embedded in the N document images from N images to be processed (where N is a natural number of 3 or more),
Image feature extraction is performed on N document images, and N × C images are obtained by dividing the N document images into the C hierarchies based on the feature extraction result. Part (where C is a natural number of 2 or more),
An average similarity calculator for calculating an average similarity of the N document images in each layer;
A reference image specifying unit that specifies, as a reference image, a document image having an average prediction error smaller than a preset threshold in N images in the hierarchy having the maximum average similarity;
An image synthesis unit that synthesizes N document images in the hierarchy based on the reference image, and identifies the synthesized image as an image including a common pattern;
A watermark detection system comprising:

(Supplementary note 24) The watermark detection system according to supplementary note 23, wherein the watermark detection system is integrated in a scanner, a copier, or a multifunction integrated device.

200 Data Processing System 202 Processor 204 Processor 206 System Bus 208 Memory Controller / High Speed Buffer 209 Local Memory 210 I / O Bridge 212 I / O Bus 214 PCI Bus Bridge 216 PCI Bus 218 Modem 220 Network Adapter 222 PCI Bus Bridge 224 PCI Bus Bridge 226 PCI bus 228 PCI bus 230 graphics adapter 232 hard disk 300 image processing method 1300 image processing method 1400 image processing device 1410 image feature extraction unit 1420 reference image specifying unit 1430 average similarity calculation unit 1440 image synthesis unit 1450 noise removal unit

Claims (8)

An image processing method in which an image processing apparatus specifies an image including a common pattern in N images from N images to be processed (where N is a natural number of 3 or more),
Image feature extraction is performed on N images, and N × C images are obtained by dividing the N images into the C layers based on the result of the feature extraction (here, C × Is a natural number of 2 or more)
Calculating an average prediction error calculated based on matching parameters between two images of the image and the remaining N−1 images in each hierarchy;
Calculating an average similarity of the N images in each hierarchy;
The average degree of similarity is set to N pieces of said average prediction error is the reference image small images than a preset threshold in the image at the maximum hierarchy, were based on the reference image by combining the N images in the hierarchy synthesis Identifying the image as an image containing a common pattern;
Is executed by the image processing apparatus. Calculating a pre-Symbol average prediction error,
Based on matching parameters between two images of the image and the remaining N−1 images, a prediction matching parameter between two images of the remaining N−1 images for the image is calculated. Obtaining a predicted matching parameter matrix for the image;
Calculating the average prediction error of the image based on the prediction matching parameter matrix for the image;
The image processing method according to claim 1, further comprising: The step of calculating the average similarity of N images in each hierarchy includes:
Calculating a prediction accuracy probability of each image based on the average prediction error of each image;
Calculating the degree of similarity of each image using the prediction accuracy probability of each image based on the degree of similarity between two images of each image and the remaining N−1 images;
Calculating the average similarity of N images based on the similarity of each image;
The image processing method according to claim 2, further comprising: The step of performing feature extraction of an image and obtaining N × C images by dividing the N images into the C layers based on the result of the feature extraction includes:
Extracting all edges in the N images using an edge detection operator;
Calculating the magnitude of the edge strength at each edge point;
Dividing the N images into the C layers based on the calculated magnitude of the edge strength;
The image processing method according to claim 1, further comprising: The method further includes obtaining the composite image, and obtaining the composite image includes:
Aligning the remaining N-1 images in the hierarchy with the reference image;
By accumulating the combined N images for each pixel point, the total number of all overlaid edge points at the pixel point is calculated, and the synthesized edge image is acquired as the synthesized image based on the total number. Steps,
The image processing method according to claim 4, further comprising:   The image processing method according to claim 1, wherein the N images to be processed are document images, and the common pattern is a watermark embedded in the document image. An image processing apparatus that identifies an image including a common pattern in the N images (N is a natural number of 3 or more) from N images to be processed,
An image feature extraction unit (here, N × C images are obtained by performing image feature extraction on N images and dividing the N images into the C layers based on the result of the feature extraction) C is a natural number of 2 or more)
In each layer, an average prediction error calculated based on matching parameters between two images of the image and the remaining N−1 images is calculated, and the average similarity of the N images in each layer An average similarity calculator for calculating the degree,
A reference image selecting unit which the average degree of similarity to identifying images smaller than a threshold the average prediction error is preset in N images at the maximum hierarchy as a reference image,
An image combining unit that combines N images in the hierarchy based on the reference image, and identifies the combined image as an image including a common pattern;
An image processing apparatus comprising: A watermark detection system that identifies an image including a watermark embedded in the N document images from N images to be processed (where N is a natural number of 3 or more),
Image feature extraction is performed on N document images, and N × C images are obtained by dividing the N document images into the C hierarchies based on the feature extraction result. Part (where C is a natural number of 2 or more),
An average prediction error calculated based on matching parameters between two images of the image and the remaining N−1 images in each layer is calculated, and an average of the N document images in each layer An average similarity calculator for calculating the similarity,
A reference image selecting unit which the average degree of similarity is to identify a small document image than a threshold the average prediction error is preset in N images at the maximum hierarchy as a reference image,
An image synthesis unit that synthesizes N document images in the hierarchy based on the reference image, and identifies the synthesized image as an image including a common pattern;
A watermark detection system comprising:

Images