WO2001065488A1

WO2001065488A1 - Apparatus for retrieving image and method for retrieving image

Info

Publication number: WO2001065488A1
Application number: PCT/JP2001/001498
Authority: WO
Inventors: Makoto Kato; Tsuyoshi Suzuki
Original assignee: Hamamatsu Photonics K. K.
Priority date: 2000-02-29
Filing date: 2001-02-28
Publication date: 2001-09-07
Also published as: AU2001235992A1; JP2001243474A

Abstract

Image data on an original image for comparison and an objective image to be retrieved is converted into optical images and the optical images are written, side by side, in a spatial optical modulator. The pattern of the written optical images then read out with coherent light, a correlation image is formed by transforming the read out modulation output by means of Fourier transform, and the relative positions of the optical images are altered. The difference is determined between the correlation image before the relative positions are altered and the correlation image formed after the relative position is altered. The correlation value of the objective image to be retrieved to the original image for comparison is determined by integrating the output signal of that difference and compared with a specified threshold value.

Description

Description Image retrieval apparatus and image retrieval method

The present invention relates to an image search apparatus and an image search method, and more particularly to an image search apparatus and an image search method for searching for an image by generating a correlation signal between images. Background art

In recent years, in medical practice, many image diagnoses using MRI images, CT images, pathological photographs, X-ray photographs, ultrasonic images, and the like have been performed. When performing image diagnosis, past cases are often referred to, and an image search system has been conventionally used. The image retrieval system makes a database of digitized diagnostic images along with supplementary information such as patient ID, name, and imaging site, and performs image retrieval using this database. Disclosure of the invention

As a search method, a method of searching for an image using information attached to the image such as a patient name as a key can be considered.

However, in such a search method, an image similar to the image to be diagnosed is not directly searched, and thus the searched image is not always useful for diagnosis. In addition, since search conditions are limited based on supplementary information, an almighty search cannot be performed. The search accuracy can be improved by increasing the types of supplementary information and increasing the conditions of the supplementary information at the time of the search. It takes a lot of time and effort to input data, which is not realistic. Therefore, a similar image search method that directly searches for an image similar to the image to be diagnosed can be considered. As a method of searching for similar images, for example, a method of searching based on the degree of similarity using pattern matching can be considered. However, a search method using pattern matching is `` close to perfect match but slightly If there are robust factors such as "there is a difference", "the size is different", or "the viewing angle is different", even if the images are similar, it is determined that the combination is dissimilar May not.

Furthermore, another method of searching for similar images is to generate a correlation image by the Joint Transform Correlation method and detect the peak value to determine similarity. . Japanese Patent Laid-Open No. 2000-111178 discloses such a similar image search method.

However, in the similar image search method according to the publication, laser light that is coherent light is used to generate a correlation image, and this laser light reacts with irregularities of a lens and dots of pixels on a recording surface. , Causing speckle noise. For this reason, similar images cannot be accurately extracted.

In addition, due to the influence of the medical imaging network standard DICOM (Digital Imaging and Communications in Medicine), it is expected that the use of image networks will spread and the sharing of image data will progress in the future. With this, the number of images stored in the database is expected to be enormous. Therefore, it is desired to quickly and accurately search for a desired image in such a large-scale database.

The present invention has been made to solve the above-described problem, and can quickly and accurately search for a similar image from a huge number of images, and It is an object of the present invention to provide an image search device and an image search method in which cost is reduced.

In order to solve the above problems, an image search device of the present invention includes: an image data storage unit configured to store image data of a plurality of search target images; an image data of one comparison original image; Image data conversion means for converting the image data of one of the search target images into an optical image, and writing the optical image of the comparison original image and the optical image of the one search target image side by side A spatial light modulator for reading out the written two optical image patterns with coherent light, and a correlation for forming a correlation image on a correlation output surface by Fourier-transforming the modulation output read from the spatial light modulator. Image forming means for changing a relative position between the optical image of the comparison original image and the optical image of the one search target image and writing the two changed optical images to the spatial light modulator; An image position changing means for performing a Fourier transform of the modulation output read from the spatial light modulator to the correlation image forming means to form a correlation image on the correlation output surface; and A correlation image formed for the two optical images before the relative position is changed, and a correlation image formed for the two optical images after the relative position is changed by the image position changing unit Difference detecting means for obtaining a difference from the correlation image, and detecting the correlation value of the one search target image with respect to the comparison original image by integrating the detection result of the difference detecting means within a certain area on the correlation output surface. A comparison unit that compares the correlation value with a predetermined threshold value and determines whether the one search target image is a similar image similar to the comparison original image based on the comparison result. means An output unit that outputs the one search target image as a similar image when the comparison unit determines that the one search target image is a similar image; and The plurality of search targets stored in the image data storage unit By sequentially switching the image data of at least one search target image in the image data of the image one by one and converting it into an optical image, the at least one search target image and the comparison original image are sequentially converted. Similar image extraction control means for outputting, as a similar image, a search target image determined to be similar to the comparison original image by comparison.

Here, the correlation image formed on the correlation output surface by the correlation image forming means includes not only a correlation signal formed based on the comparison original image and the search target image itself, but also speckle noise. . Here, the location of the correlation signal based on the image itself changes when the relative position of the image is changed. On the other hand, speckle noise that appears due to the optical system does not change its location even if the relative position of the image is changed. Therefore, the correlation signal can be separated from speckle noise by taking the difference between the correlation image before changing the relative position and the correlation image after changing the relative position.

In order to find the same image, the peak height of the luminescent spot of the primary light (+ primary light or primary light) that appears at a position uniquely determined by the relative positions of the two images is detected. do it. However, in order to find a similar image, it is not enough to simply detect such a bright spot peak of the primary light, and it is necessary to detect feature points at the same or similar but different positions in the similar image. Therefore, it is necessary to collect correlation signals that appear in various places on the correlation output surface. In the present invention, since the intensity of the correlation signal is integrated within a predetermined area, correlation signals appearing based on feature points of the same or similar but different positions of the similar image can be collected. It is possible to determine with high accuracy whether or not the image is a similar image. In addition, since the correlation signal can be separated from speckle noise, even weak correlation signals that were initially buried in the speckle noise can be accurately picked up and integrated. This makes it possible to determine with high precision whether or not the two images are similar.

Furthermore, since a correlation image is formed directly using a spatial light modulator, a correlation image can be formed more quickly than when a correlation operation is performed on a combination basis. Therefore, similar images can be quickly searched from a huge number of images. Also, the cost can be reduced as compared with a case where a computer performs a correlation operation.

Here, it is preferable that the image search device of the present invention further includes an integration region designating unit that designates the region where the correlation value detection unit performs integration. For example, the integration region is defined as a small range in the correlation output plane. If the specified area is specified, the area outside the specified area will be excluded from the similarity judgment even if there is a similar part. Therefore, in this case, it is possible to search for identical / similar images that have the same degree of identity and have the same or similar feature points with little displacement.

On the other hand, if the integration region is specified to be a large range, the same or similar portions whose positions are greatly deviated from each other can be included in the similarity determination, so that the similar range can be expanded and searched.

In addition, by preventing the zero-order light from being included in the integration region, detection of the autocorrelation signal can be avoided, and highly accurate detection can be performed. In addition, by preventing the integration region from including the primary light (+ 1st-order light and 1st-order light), it is possible to search only similar images and not search the same image. . Here, the zero-order light appears at a predetermined position on the correlation output surface that is uniquely determined by the optical system of the correlation image forming means. The first-order light (+ 1st-order light and 1st-order light) also appears on the correlation output surface at a predetermined position determined by the positional relationship between the comparison original image and the search target image. Therefore, when specifying the integration region, the 0th order light and the 1st order light What is necessary is just to exclude the appearance position.

It is preferable that the image search device of the present invention further includes a relative position change amount designating unit that designates a relative position change amount changed by the image position changing unit. By adjusting the amount of change, similar images can be searched with higher accuracy.

Further, the image search device of the present invention further includes: a search range specifying unit that selects image data of the at least one search target image to be sequentially compared with the comparison original image from image data of the plurality of search target images. It is preferable to further include For example, if additional information is added to the image data of each search target image in advance, based on this additional information, at least one image data to be searched can be selected and provided for the search.

Further, the correlation image forming means includes: first Fourier transform means for forming a first Fourier transform pattern by Fourier-transforming the modulated output read from the spatial light modulator; and Another spatial light modulator for reading out the written and written Fourier transform pattern with coherent light, and second Fourier transform means for Fourier transforming the modulated output read from the other spatial light modulator to form a correlation image It is preferable to consist of

According to another aspect, an image search method according to the present invention includes an image data conversion step of converting image data of one comparative original image and image data of one search target image into optical images, respectively. The optical image of the comparison original image and the optical image of the one search target image are written side by side in a spatial light modulator, and the written two optical image patterns are written in coherent light using the spatial light modulator. A correlation image forming step of performing a Fourier transform on a modulation output read from the spatial light modulator in the modulation step to form a correlation image on a correlation output surface; Image and the one search pair The relative position of the elephant image with respect to the optical image is changed, the two optical images after the change are written into the spatial light modulator, and the modulation output read from the spatial light modulator is Fourier-transformed. An image position changing step of forming a correlation image on the correlation output surface, a correlation image formed on the two optical images before a relative position is changed in the image position changing step, and the image A difference detection step of obtaining a difference between the two optical images after the relative position is changed in the position change step and a correlation image formed, and a detection result of the difference detection step on the correlation output surface. A correlation value detection step of integrating within a certain area to detect a correlation value of the one search target image with respect to the comparison original image, and comparing the correlation value with a predetermined threshold value, based on the comparison result, The one search target image is compared with the one A comparison step of determining whether or not the image is similar to the original image; a step of outputting the search target image determined to be a similar image in the comparison step as a similar image; and at least one search By repeatedly performing the image data conversion step while sequentially switching the image data of the target image, the at least one search target image and the comparative original image are sequentially compared and determined to be similar to the comparative original image. And a similar image extraction control step of outputting the retrieved search target image as a similar image.

The correlation signal can be separated from the speckle noise by taking the difference between the correlation images before and after changing the relative position, so that the weak correlation signal buried in the speckle noise can be accurately picked up and used for integration. This makes it possible to determine with high precision whether or not the two images are similar. BRIEF DESCRIPTION OF THE FIGURES

FIG. 1 shows an overall configuration of an image search device according to an embodiment of the present invention. FIG.

FIG. 2 is a diagram showing a structure of an optical address type SLM used in the image search device according to the above embodiment.

FIG. 3 is a diagram showing a detection image of the CCD image sensor.

FIGS. 4 (a) and 4 (b) are diagrams showing the principle of similar image retrieval by the image retrieval device according to the above embodiment, and FIG. Fig. 4 (b) is a diagram showing a state where the target image is side by side and is incident on the first SLM, and Fig. 4 (b) shows the comparison original image and the search target image as shown in Fig. 4 (a). FIG. 3 is a diagram showing a correlation image obtained on a correlation output surface (on a CCD image sensor) when the light is incident on a first SLM.

FIG. 5 is a flowchart showing an image search operation of the image search device according to the above embodiment.

FIG. 6 is a diagram showing a display state of a parameter window screen displayed first when an image search operation of the image search device according to the above embodiment is started.

Fig. 7 shows the display status of the "Open" window screen that appears when you click the "Ref" button on the "Ref" side in "Mu 1ti Division" in the parameter overnight window screen in Fig. 6. FIG.

Fig. 8 shows the display status of the directory selection window screen that appears when you click the "Browse" button on the "Dir" side of "Mu 1 ti division" in the parameter window window of Fig. 6. It is. FIG. 9 is a diagram showing a display state of a comparison window screen appearing when the “Start” button in “Mu1ti division” on the parameter overnight window screen of FIG. 6 is clicked.

FIG. 10 is a diagram showing a display state when parameters are set on the parameter window screen of FIG. The first 1 figure best mode for carrying out the _t invention is a diagram illustrating a display state of the search result image display window screen displayed in the image retrieval operation of the image retrieval apparatus according to the above embodiment

An image search device according to an embodiment of the present invention will be described with reference to FIGS. 1 to 11.

In the description of the drawings, the same elements will be denoted by the same reference symbols, without redundant description. Also, the dimensional ratios in the drawings do not always match those described.

FIG. 1 shows a configuration diagram of an image search device according to an embodiment of the present invention. The image retrieval apparatus 100 of the present embodiment includes a central processing unit (hereinafter, “CPU”) 7, a computer tomographic imaging apparatus (hereinafter, “CT”) 1, a magnetic resonance imaging apparatus (hereinafter, “MRI”). 2, network server 3, digital camera 4a, magnetic disk 5, computer monitor 6, input device 12, RAM 14, ROM 16, hard disk device 17, image memory 18, video RAM 8, and A / D It is composed of an optical operation unit 20 connected to the CPU 7 via the converter 14.

The CT 1, MR 12, network server 3, and digital camera 4a are means for capturing images, and are connected to the CPU 7 via the computer bus 9. The digital camera 4a is connected to the computer bus 9 via a digital camera interface (I / F) 4b. The image data captured by these image capturing means is stored on a magnetic disk 5 as image data storage means via a computer bus 9.

Now, multiple search target images 10 a, 10 b, 10 c, 10 d, 10 e,

It is assumed that 10 f,..., 10 z are stored in the magnetic disk 5. Further, it is assumed that the comparison original image 10 a ′ is fetched from the CT 1, MR 12, or the digital camera 4 a (in this case, the digital camera 4 a) and stored in the magnetic disk 5. In this case, the image search device 100 sends a plurality of search target images 10a, 10b, 10c, 10d, 10e, 10f,... Stored in the magnetic disk 5. · Images similar to the comparison original image 10a, can be searched from among 10z.

The comparison original image 10a 'and the multiple search target images 10a, 10b, 10c, 10d, 10e, 10f, Additional information such as patient information (patient ID, name, measurement site, etc.) and scale (length information per pixel) are added. Therefore, the operator can specify the range of the search target image based on the image header information (sex, age, measurement site, etc.) before executing the search. As a result, among the plurality of search target images 10 a, 10 b, 10 c, 10 d, 10 e, 10 f,. Images belonging to a specific group that meet the conditions specified by the operator (for example, 10a, 10b, 10c, 10d, 10e, and 10f) are processed in the current search process. Is the range of the search target.

The video RAM 8 is a memory for temporarily storing two images to be compared. More specifically, one image selected as a search target among the search target images 10a, 10b, 10c, 10d, 10e, and 10f within the current search target range (for example, , 10 a) and the comparison original image 10 a ′ are stored in the video RAM 8 via the computer bus 9. The images selected as search targets from among the images 10a, 10b, 10c, 10d, 10e, and 10f in the search range this time are sequentially replaced during work. Each time it is compared with the comparison original image 10a '.

The optical operation unit 20 performs optically congruent conversion correlation (Joint Transf orm Correlation) processing to directly detect the correlation value (an index of the degree of similarity between the two images) between the two images in the video RAM 8 as the light intensity. The optical operation unit 20 includes a liquid crystal panel 21 a, 21 b, a zoom lens 22 a, 22 b, an image rotation mirror 23, a reflection mirror 24, a spatial light modulator (hereinafter, “SLM”) 25. a, 25 b, laser diode (hereinafter, “: LD”) 26, collimated lens 27, half mirror 28, 29, 31 1, mirror 30, Fourier transform lens 32 a, 32 b, and CCD It consists of an image sensor 33.

The liquid crystal panel 21a, the zoom lens 22a, and the image rotation mirror 23 are for pre-processing the comparison original image 10a 'stored in the video RAM 8, and are provided after the liquid crystal panel 21a. In addition, a zoom lens 22a is provided, and an image rotation mirror 23 is further provided behind the zoom lens 22a. The liquid crystal panel 21a is for converting a digital image into an optical image. The zoom lens 22 a is for enlarging or reducing the optical image displayed on the liquid crystal panel 21 a. The image rotation mirror 23 is for rotating the optical image enlarged or reduced by the zoom lens 22a.

The liquid crystal panel 21b and the zoom lens 22b are for preprocessing the search target image (in this case, 10a) stored in the video RAM 8, and the zoom is performed after the liquid crystal panel 21b. The lens 22b is arranged. The liquid crystal panel 2 lb is for converting a digital image into an optical image. The zoom lens 22b is for enlarging or reducing the optical image displayed on the liquid crystal panel 21b.

A reflection mirror 24 having two reflection surfaces is disposed downstream of the image rotation mirror 23 and the zoom lens 22b. On one reflection surface of the reflection mirror 24, the optical image of the comparison original image 10a 'supplied from the liquid crystal panel 21a, the zoom lens 22a, and the image rotation mirror 23 enters, and the other. The optical image of the search target image 10a supplied from the liquid crystal panel 21b and the zoom lens 22b is incident on the other reflecting surface. The two incident optical images are reflected by the respective reflecting surfaces, and have a predetermined positional relationship (the optical address section) on the writing side (optical address section) of the first SLM25a disposed after the reflection mirror 24. Parallel to the X-axis of each other and at a distance M from each other. A first half mirror 29 and a first free-transform lens 32a are arranged at a stage subsequent to the first SLM 25a, and a second SLM 25b is arranged at a stage subsequent thereto. Have been. A second half mirror 31 and a second Fourier transform lens 32b are arranged downstream of the second SLM 25b.

The LD 26 is for continuously outputting laser light (hereinafter referred to as “reading light”), which is a coherent light, which is incident on the reading side (light modulator) of the SLMs 25a and 25b. Between the LD 26 and the first half mirror 29, a collimator lens 27 and a third half mirror 28 are arranged. The collimating lens 27 is for collimating the readout light output from the LD 26 into parallel light. The third half mirror 28 partially divides the reading light (parallel light) from the collimating lens 27 partially by transmitting and partially reflecting the light. The read light transmitted through the third half mirror 28 is reflected by the first half mirror 29 and is incident on the read side of the first SLM 29. The mirror 30 changes the optical path of the readout light reflected by the third half mirror 28 and guides it to the second half mirror 31. The reading light that has entered the second half mirror 31 is reflected and enters the reading side of the second SLM 25 b.

Here, the structure and operation of the SLMs 25a and 25b will be described with reference to FIG. S LM2 5a and 25b are connected to the reading side (optical modulation side (lower side in Fig. 1) according to the image input to the writing side (optical address side (upper side in Fig. 1; left side in Fig. 2)). Readout light incident on the right side of Fig. 2)) Is output after phase modulation. In the present embodiment, an optically-addressable SLM is used. In this example, a parallel-aligned nematic liquid crystal spatial light modulator (PAL-SLM) is used as a specific example of the SLM25a, 25b.

In this SLM, a nematic liquid crystal layer 416 (in this case, a parallel alignment nematic liquid crystal layer) as a light modulating section is provided between a pair of alignment layers 415 and 417. On the side of the alignment layer 415 opposite to the liquid crystal layer 416, a mirror 414, a photoconductive layer 413 as an optical address portion, a transparent conductive film 412, and a writing-side glass substrate 411 are provided. , Are provided in this order. Further, a transparent conductive film 418 and a reading-side glass substrate 419 are provided on the side of the alignment layer 417 opposite to the nematic liquid crystal layer 416. Photoconductive layer 4 1 3 is, for example, amorphous silicon (a- S i), C d S, B i 12 0 2. And organic photoconductor (PVK). The mirror layer 414 is made of, for example, a dielectric mirror. The alignment layers 415 and 417 may be formed by coating PVA or polyimide on the mirror 414 and the transparent conductive film 418 and then rubbing the film, oblique evaporation of SiO or LB film. It may be created by the formation of. An AC driving voltage is applied between the pair of conductive films 412 and 418 of the SLMs 25a and 25b, for example, from AC power supplies 500a and 500b.

When writing light is incident on the photoconductive layer 413 via the writing-side glass substrate 4 1 1 1, the resistance at that portion decreases according to the writing light intensity, and the resistance of the liquid crystal layer 416 depends on the writing light intensity. Voltage is applied. The liquid crystal molecules move at a rate corresponding to the voltage, and the refractive index changes. Therefore, when coherent read light polarized in the length direction of the liquid crystal molecules from the LD 26 enters the liquid crystal layer 416 via the read-side glass substrate 419, this light is refracted. The phase is modulated according to the rate change.

The first Fourier transform lens 32a performs a Fourier transform on the readout light modulated and output on the readout side (the liquid crystal layer 416) of the first SLM25a, and converts the first Fourier pattern to the focal plane thereafter. It is for forming into. The second SLM 25b has its write side (optical address side (photoconductive layer 413)) disposed on the back focal plane of the first Fourier transform lens 32a. Therefore, the first Fourier pattern is imaged on the writing side of the second SLM25b.

The second Fourier transform lens 32b performs a Fourier transform on the readout light modulated and output on the readout side (liquid crystal layer 416) of the second SLM25b, and then performs a Fourier transform on the focal plane (correlation output plane). 2 to form a Fourier pattern (correlated image).

A CCD image sensor 33 is disposed on the back focal plane (correlation output plane) of the second Fourier transform lens 32b. The CCD image sensor 33 is configured by a two-dimensional array of a plurality of pixels, and can detect the light intensity at each pixel position. Therefore, the CCD image sensor 33 detects the light intensity (luminance) at each pixel position of the second Fourier pattern and outputs an electric signal corresponding to the detected light intensity. This electric signal is converted into a digital signal (luminance value (gradation value)) by the AZD converter 14 and then supplied to the CPU 7. The digital signal thus supplied to the CPU 7 is temporarily stored in the image memory 18 as correlation image data. The correlation image data can be displayed on a computer monitor 16 and, for example, as shown in FIG. 3, can be displayed as a LIFE image.

As described above, in the present embodiment, the optical operation unit 20 optically converts the image data of the comparison original image 10 a ′ and the image data of the search target image 10 a into optical data, respectively. After conversion into an image, these correlation patterns are directly formed by optically performing joint conversion correlation processing. Here, if such a joint conversion correlation process is to be performed by a computer, a high-performance computer is required, and the manufacturing cost of the entire image search device is increased. On the other hand, in the present embodiment, since the correlation operation processing is optically performed by the optical operation unit 20, the manufacturing cost can be significantly reduced. In addition, since the optical operation unit 20 can perform the operation processing at extremely high speed, the operation processing can be performed in a short time even if the number of search target images is enormous.

Next, the principle of similar image retrieval according to the present embodiment will be described with reference to FIG.

(b) Specific description will be made with reference to the drawings.

Now, it is assumed that an image as shown in FIG. 4 (a) is input to the video RAM 8 as a comparison original image 10a 'and a search target image 10a. Here, the comparison original image 10a 'and the search target image 10a are images in which the same person appears but has locally different portions. Optical images corresponding to these two images are generated by the liquid sub-panels 21a and 21b, and are incident on the writing side (photoconductive layer 413) of the SLM25a. Here, as shown in FIGS. 1 and 4 (a), these optical images 10a ′ and 10a have a relative positional relationship parallel to the X axis and shifted by a distance M. Injected into SLM2 5a. The XY plane in the drawing is the writing-side incident surface of the SLM 25a (the writing light incident surface of the photoconductive layer 413), and the X-axis and the Y-axis are perpendicular to each other. Further, as shown in FIG. 1, the optical axes (Z-axis) of the first and second Fourier transform lenses 32a and 32b extend perpendicular to the XY plane.

The two images 10a 'and 10a are Fourier-transformed by the Fourier transform lens 32a, and the obtained first Fourier transform pattern is written into the SLM25b. After reading the conversion pattern, When the Fourier transform is performed by the Fourier transform lens 32b, the correlation output surface (CCD image sensor 33), which is the back focal plane of the Fourier transform lens 32b, has a second Fourier transform shown in FIG. A conversion pattern (correlated image) is obtained. The XY plane in FIG. 4 (b) shows the back focal plane (correlation output plane) of the Fourier transform lens 32b, and this X axis is the X axis (the second plane) on the writing side entrance plane of the SLM 25a. 4 (a)), and the Y axis is parallel to the Y axis (Fig. 4 (a)) on the writing-side entrance surface of the SLM25a. The Z axis is the optical axis of the first and second Fourier transform lenses 32a and 32b, and extends perpendicular to the XY plane in the figure. As shown in Fig. 3, the autocorrelation between the image 10a 'and the image 10a is located on the Z axis (the origin of the figure) on the focal plane (correlation output plane) after the Fourier transform lens 32b. The zero-order light, which is the signal light, appears. As is evident from Fig. 4 (a), the parts with a high degree of coincidence in both images, for example, the part a in image 10a 'and the part a in image 10a It is displaced substantially parallel by a distance that is approximately equal to the distance M. Therefore, on the correlation output surface (on the CCD image sensor 33), as shown in FIG. 4 (b), a pair of correlation signal lights A based on these image portions a and a ' It appears at a pair of positions that are displaced from the position of the optical axis Z) of the conversion lens 32b substantially parallel to the X axis by a distance substantially equal to M · (f2 / f1). Here, f1 is the focal length of the Fourier transform lens 32a, and f2 is the focal length of the Fourier transform lens 32b.

Similarly, a portion having a high degree of coincidence, for example, a portion b 'of the image 10a' and a portion 13 of the image 10 & are also shifted by a distance substantially parallel to the X axis and substantially equal to the distance M. Therefore, a pair of correlation signal lights B based on these image parts b and b 'are also M' (f2 / f1) substantially parallel to the X axis from the position of the origin (optical axis Z of the Fourier transform lens 32b). Appear at a pair of positions displaced by a distance approximately equal to. Similarly, other parts having a high degree of coincidence, for example, the part c 'of the image 10a' and the part c of the image 10a are also shifted by a distance substantially parallel to the X axis and substantially equal to the distance M. Therefore, a pair of correlation signal light C based on these image portions (:, c ′) is also substantially parallel to the X axis from the position of the origin (optical axis Z of the Fourier transform lens 32b) by M * (f2 / f1 ) Appear at a pair of positions displaced by a distance approximately equal to).

In this way, these correlation signal lights are shifted from the origin (optical axis Z of the Fourier transform lens 32b) to a pair of positions substantially parallel to the X axis and shifted by a distance substantially equal to Μ · (f2 / f1). A strong peak (primary soil light) appears due to the gathering.

However, although the part d 'of the image 10a' and the part d of the image 10a are mutually identical image parts, their positions are different. Specifically, the image parts d 'and d are shifted by a distance L parallel to the X axis and a distance N parallel to the Y axis. Therefore, the correlation signal light D based on these image parts d and d 'is distanced from the origin (the optical axis Z of the Fourier transform lens 32b) in parallel to the X axis by a distance L · (f2Zf1), Y It is generated at a pair of positions parallel to the axis and offset by a distance N · (f 2 / f 1).

If there are a plurality of image parts that match but have different positions, such as these image parts d and d ', on the images 10a and 10a', the correlation signal light based on these matching image parts It will occur discretely.

When the images are the same as in the image portions d and d ′, the intensity of the correlation signal light D based on these image portions increases. On the other hand, if the image parts d and d ′ are not the same as each other but are slightly different, that is, if the images are similar, the intensity of the correlation signal light D based on these image parts is not so high.

Although not shown in FIG. 4 (b), the optical Since laser light, which is lent light, is used, speckle noise is generated in response to the unevenness of the lens and the dots of the pixels on the recording surface (LCD panels 21a and 21b). Therefore, among the correlation signals based on the similar image portion, those smaller than the speckle noise are buried in the speckle noise.

Here, since the speckle noise is caused by the optical system, the position where the speckle noise appears does not change even if the positions of the comparative images 10a and 10a 'are shifted. On the other hand, since the correlation signal is formed on the basis of the image itself, the position at which the image appears is shifted by shifting these images 10a and 10a '. Therefore, in the present embodiment, the following measures are taken to remove the spike noise.

(1) First, a Fourier transform is performed on the optical image of the comparison original image 10 a ′ to be compared and the search target image 10 a to form a second Fourier pattern (correlation image).

(2) Next, the relative position of these two optical images 10a and 10a 'is changed and Fourier transform is performed again to obtain a second Fourier pattern (correlation image). That is, one of the two images is displaced by about 10% of the size of the image in any direction of up, down, left, and right on the XY plane in FIG. 4 (a). More specifically, the position of any one of the images 10a 'and 10a is shifted by adjusting the storage position in the video RAM8. (3) Two second Fourier patterns obtained in (1) and (2) above

(Correlation image)

As described above, since speckle noise is caused by the optical system, the position where the speckle noise appears does not change even if the image is shifted. On the other hand, the correlation signal light caused by the image itself moves when the image is shifted. Therefore, by taking the difference between the two second Fourier transform images obtained in (1) and (2) in (3), speckle noise is removed and only the shifted correlation signal is obtained. Can be obtained.

The correlation image detected by the CCD image sensor 33 includes noise due to the CCD image sensor 33 itself and quantization noise of the comparative image, in addition to the above speckle noise. By taking the difference between the two correlation images before and after the movement, these noises can also be removed.

In the present embodiment, the absolute value of the difference between the correlation signals remaining after the speckle noise has been removed in this manner is integrated in a predetermined area in the correlation output plane, and based on the integration result (sum of luminance values). , The similarity between the two images 10 a and 10 a ′ is determined.

Here, the reason for integrating the luminance in a predetermined region in the correlation output plane in the present embodiment will be described.

When the comparison original image 10a 'and the search target image 10a shown in Fig. 4 (a) are the same image, that is, the image parts a, b, c and a', bc ' Thus, when the same image portion is an image in which the same image portion is arranged at the same position, the relative position of the two images is displayed on the correlation output surface (the focal plane after the Fourier transform lens 32b). At a pair of positions (positions that are away from the origin parallel to the X-axis by soil M · (f 2 / f 1)), a high luminance peak (+ Primary light, primary light). Therefore, by detecting a luminance peak at one of these paired positions and comparing it with a predetermined threshold value, it is determined that the luminance is greater than the threshold value, and it is determined that these comparative images are the same image. Can be determined.

As described above, in the same image, the locally coincident image portions exist at the same position. Therefore, the relative positions of the image parts that match each other are equal to each other, and are equal to the relative positions of the entire images. New For this reason, the luminances (correlation signals) that appear when the respective image portions match each other are gathered at a predetermined pair of positions to generate ± 1 order light with extremely high luminance. Therefore, for the same image, a correlation signal with high luminance is generated at a predetermined pair of positions corresponding to the relative positions.

On the other hand, if the comparison original image 10a 'and the search target image 10a are images similar to each other, the same or similar image portions such as image portions d and d' are placed in different locations. There are various types of distribution. For this reason, the correlation signals based on these image portions also appear at various positions on the correlation output surface. Therefore, since these correlation signals do not collect at the above-mentioned pair of predetermined positions (positions separated from the origin by a distance M · (f 2 / f 1) parallel to the X axis), the predetermined pair of positions High peaks do not appear as in the case of the same image. Therefore, simply detecting the luminance peak at this position and comparing it with the threshold value can only determine that the luminance peak is equal to or less than the threshold value and therefore is not the same, and cannot detect a similar image. -As described above, although images that are similar to each other have parts that are locally identical or similar, these parts that are identical or similar are distributed to different parts of each image. Existing. Therefore, the luminance (correlation signal) generated by these coincident or similar parts is scattered and generated in various places.

Therefore, in the present embodiment, correlation signals generated and scattered in various places as described above are gathered by integration to be a factor for similarity determination. For this reason, in the present invention, integration is performed not for a peak waveform but for a region having a predetermined size.

In the present embodiment, the integration area is variable. During the operation, the position and size of the integration area can be set using the parameters. For example, if the integration area is reduced, the position of matching image parts An image with little misalignment and high identity can be detected. On the other hand, if the integration area is made large, it is possible to detect an image having many parts that are the same or similar when viewed locally even if the position is different.

However, the integration region is limited to a position and a range that does not include the zero-order light so as not to detect autocorrelation that causes noise.

For example, in the case of the examples of FIGS. 4 (a) and 4 (b), if the integration area is set to the area surrounded by the broken line の in FIG. 4 (b), discrete generation is performed. Since the correlation signal light, for example, the correlation signal D, can also be included in the integration, the parts d and d ′ in FIG. 4 (a) can also be used as elements for similarity determination. For example, as the integration area ①, on the CCD image sensor 33, the position of the first-order primary light where the correlation signal light is gathered (the position shifted by 1 M · (f2 / f1) parallel to the X axis from the origin). It is sufficient to set a rectangular area with one side in the X-axis direction having a length of, for example, 1.5 times the distance M · (f2 / f1). In this way, the autocorrelation signal (zero-order light) can be excluded from the integration region ①.

On the other hand, if the integration area is reduced as shown by the area surrounded by another broken line ② in Fig. 4 (b), it is possible to search only images with a high degree of identity with little misalignment of the matching image part. . For example, as the integration area 、, on the CCD image sensor 33, the first side of the first-order optical position where the correlation signal light is converged, and one side has a length equal to the distance M · (f2 / f1). It is conceivable to set it in the range of a square.

Furthermore, if the integration area is set to an area other than the area ① in the area の in Fig. 4 (b), the position of the primary light (in this case, — primary light) where the correlated signal light is collected will be within the range of integration. Can be excluded. Therefore, it is necessary to identify and search for similar images that include image parts that are locally identical or similar but are displaced from the same image or dissimilar images. Becomes possible. Therefore, only similar images can be searched. Further, the integration area may be set to an area surrounded by another broken line ③ in FIG. 4 (b). Also in this case, the vicinity of the primary light position where the correlation signal light is gathered can be excluded from the integration range, so that only similar images can be searched. Thus, the integration region can be set to an arbitrary region.

In this embodiment, as described above, instead of detecting the peak of the high-brightness primary light at a predetermined pair of positions, the integrated value of the luminance in a predetermined region is measured, and thus the speckle noise It is extremely important to remove these in advance.

In the image search apparatus 100 of the present embodiment, in order to search for similar images by the above-described method, a comparison original image (in this case, 10a ') as shown in FIG. For the image (in this case, 10a), a correlation image as shown in FIG. 4 (b) is generated by the optical operation unit 20 and stored in the image memory 18. Then, by changing the storage position of either the comparison original image 10a or the search target image 10a in the video RAM 8, the position of either of these two images 10a 'and 10a is changed. The modified image is recorded in the SLM 25a again, and the correlation image is generated again and stored in the image memory 18. The CPU 7 obtains the difference between the correlation image before the relative position change and the correlation image after the relative position change, which are obtained in this way. That is, at each pixel position on the imaging surface of the CCD image sensor 33, the CPU 7 calculates the difference between the light intensity of the correlation image obtained before the relative position change and the light intensity of the correlation image obtained after the relative position change. Ask for. Next, the CPU 7 integrates the difference result with respect to a predetermined area (for example, the area の in FIG. 4 (b)) on the imaging surface of the CCD image sensor 33, and sets the integration result to a predetermined threshold value. Compare with If the integration result is larger than the threshold value, the comparison original image 10 a 'and the search target image 10 a Judge that they are similar. On the other hand, if the integration result is equal to or less than the threshold value, it is determined that the comparison original image 10a 'and the search target image 10a are not similar.

Here, the CPU 7 integrates the difference result over a predetermined area as follows, and compares the integration result with a predetermined threshold value.

First, the CPU 7 compares the difference result (gradation value) of each pixel in a predetermined area (for example, the area の in FIG. 4 (b)) with a first threshold, and determines a first threshold. The number of pixels having a difference result greater than the value is counted to determine the similarity. The similarity obtained in this way is determined as an integrated value in a predetermined area of the difference result. Next, the CPU 7 compares the integrated value (similarity) of the difference result in the predetermined area obtained in this way with a second threshold value, and compares the integrated value (similarity) with the second threshold value. If it is larger, it is determined that the comparison original image 10a 'and the search target image (10a) are similar.

The input device 12 is composed of a keypad mouse (not shown), and an operator can input various instructions to the image search device 100. For example, the operator can input an instruction to perform an image search. In addition, the range of the search target image can be specified as a condition of the image search. In addition, the operator determines the other conditions for the image search, such as the amount of movement of the relative position of the image to be compared, the position and range of the integration area used in the integration operation and the first threshold value, and the comparison operation. You can also enter the value of the second threshold to be used.

The ROM 16 stores a program (search software (Fig. 5)) for image search described later. When the CPU 7 executes the image search program, the image search processing of the present embodiment is performed. The RAM I 4 stores the result calculated by the CPU 7 during the image search processing. The image search program (FIG. 5) is stored in the hard disk drive 17 and may be read into the RAM 14 and executed. Next, the image retrieval apparatus 10 according to the present embodiment configured as described above is used.

The image search processing performed by 0 will be described with reference to FIG.

When the operator instructs the input device 12 to execute the image search processing, the CPU 7 starts executing the image search program stored in the ROM 16 and starts the image search processing.

When the image search program starts, first, the operating environment narrows down the parameters of the parameter file attached to the image based on the verbal index condition (step S 1).

More specifically, when the image search program starts, the CPU 7 controls the computer monitor 6 to display a parameter window shown in FIG. The operator clicks the “Ref” button on the “Ref” side in the “Mu 1 ti division” in this parameter overnight window. Then, a dialog ("Open" window) as shown in Fig. 7 appears, and the operator selects the comparison original image (in this case, image 10a 'in Fig. 4 (a)) to be searched. Select a file name and click the “Open” button. Next, the operator clicks the “Browse” button on the “Dir” side in the “Mu1ti division” in the parameter overnight window in FIG. Then, a dialog as shown in Fig. 8 ("Directory selection window") appears. The operator selects the name of the file that contains the image to be compared in the current search process, and clicks the “SelectDir” button. As a result, the current search target range is determined. For example, if the operator selects a file named “Cytology”, the file “Cytology” contains images 10a, 10b, 10c, 10d, 10e, 10f. Is stored, the images 10a, 10b, 10c, 10d, 10e, and 10f are the images to be searched in the current search processing. In this way, the dialog shown in Fig. The search condition specifies the file name attached to the image and specifies the range of the search target image.

When the necessary input is completed as described above, the operator clicks the “Start” button in the “Mu1ti division” in the parameter setting window (FIG. 6). As a result, one of the search target images 10a, 10b, 10c, 10d, 10e, and 10f determined as described above in step S1 (for example, The image 10a in FIG. 4 (a) is taken out and stored in the video RAM 8 together with the comparison original image 10a '(step S2). As a result, the comparison window shown in FIG. 9 is displayed on the combo overnight monitor 6, and the one retrieval target image 10a retrieved this time and the comparison original image 10a 'are arranged side by side. (a) As shown in the figure, they are displayed at a distance M in the horizontal direction (X-axis direction).

Before the operator clicks the “Start” button on the parameter window in step S1, the operator can set parameters as shown in FIG. 10 while viewing the parameter window. Here, in the parameter overnight window, “Δ1ι”, “Δν”, “Threshold”, “judgment value”, “range”, and “position” are set as parameters that can be set (parameters). Is displayed.

Here, “Δ1ι” and “Δν” are parameters indicating the amount of movement of the position of the image. More specifically, Δί is arbitrarily set in the comparison window of FIG. 9 by the amount by which the comparison image (the comparison original image 10a 'or the search target image 10a) is shifted in the horizontal direction (X-axis direction). be able to. Δ ν can be set arbitrarily in the comparison window in Fig. 9 by the amount by which the comparison image (comparison original image 10a 'or search target image 10a) is shifted in the vertical direction (Y-axis direction). it can. Here, the values of Δ1ι and Δν are the number of pixels on the computer monitor 6. As these values Δh and ΔV, the comparison image (ratio The distance can be set to about 10% of the size of the comparison image or the search target image).

In the relative position changing step (step S5) described later, based on the values Δh and ΔV set in this way, either the comparison original image 10a 'or the search target image 10a is On the corresponding liquid crystal panels 2 la and 21 b, they are shifted by a distance corresponding to Δh in one of the left and right directions and by a distance corresponding to ΔV in one of the up and down directions. In this way, the image to be compared 10a or 10a 'is shifted in the horizontal direction (X axis in FIG. 4 (a)) and in the Z or vertical direction (Y axis in FIG. 4 (a)). Will be.

“Position” and “Range” are defined by the CCD image sensor 3 as shown in Fig. 3.

This is for defining the area where integration is to be performed on the imaging plane (correlation output plane) 3. Specifically, the integration region is set as a region of a size determined by “range” with reference to a coordinate determined by “position”. For example, it is set as a square or rectangular area having a center at a certain coordinate determined by “position” and a side having a length determined by “range”. Further, it can also be set as a circular area having a radius defined by the “range” and the center at a certain coordinate determined by the “position”. In this way, an area of an arbitrary shape and size can be set by the “position” and the “range”.

“Thresshο1d” is a first threshold value. In other words, it is a reference for integrating by counting the number of pixels whose luminance value of each pixel in the integration area defined by “position” and “range” is larger than this threshold, and can be set arbitrarily. it can.

The “judgment value” is the second threshold, which is compared with the integration result (similarity) in the integration area. In other words, it is a parameter for determining how low the similarity is to be selected as a similar image. It can be set arbitrarily. Therefore, only images whose similarity (integration result) obtained by counting the number of pixels on the basis of “Thresho 1 d” is larger than this “determination value” will be displayed in S 11 described later. It will be displayed on the window (Fig. 11). The “judgment value” can be weighted by patient information such as age and gender. Note that the above Δh, Δν, Threshold, judgment value, range, and position are previously set to initial values and stored in the ROM 16. Therefore, if the operator does not make any settings in S1, the initial values will be used. In addition, these parameters Ah, Δν, Threshold, judgment value, range, and position can be changed at any time during the search operation in FIG.

In addition, as a parameter that can be set for the operation overnight, the rotation angle of the comparison original image 10a 'may be specified.

In step S2, when one search target image 10a and the comparison original image 10a 'are stored in the video RAM 8, these images are

The first correlation operation processing is performed on 10a and 10a '(step S3). The first correlation image (1) obtained is detected by the CCD image sensor 33 and stored in the image memory 18 (step S4). Next, the relative position between the comparison original image 10a 'and the search target image 10a is changed (step S5). Specifically, the CPU 7 controls the video RAM 8 to store one of the comparison original image 10 a ′ and the search target image 10 a in the distance Δh set by the initial setting or by the operator. Shift up, down, left, and right by ΔV.

Next, a second correlation operation process is performed by the optical operation unit 20 on the two images 10 a ′ and 10 a whose relative positions have been changed in this way (step S 6). . The obtained second correlation image (2) is a CCD image. Detected by the sensor 33 and stored in the image memory 18 (step S 7). Next, the CPU 7 compares the first correlation image (1) stored in the image memory 18 with the second The difference from the correlation image (2) (Step S

8). That is, the difference (absolute value) between the intensity value (gradation value) of the first correlation image (1) and the intensity value (gradation value) of the second correlation image (2) is calculated for each pixel. calculate. As a result, noise such as speckle noise caused by the optical system can be removed.

Next, the CPU 7 obtains the absolute value of the correlation signal remaining after the noise is removed by calculating the difference in this manner, by the initial setting or the “position” and “range” specified by the operator. The integration is performed within the region of, and the integrated value is used as the similarity (step S9).

Specifically, the CPU 7 first calculates the difference result of each pixel within the set area (for example, the area の in FIG. 4 (b)) as “Thresho 1 d” (first threshold). Value), and count the number of pixels that have a difference result greater than “Threhold” (first threshold value) to determine the similarity.

Next, the CPU 7 determines whether or not this similarity is greater than the “determination value” (second threshold) set by the initial setting or the operator (step S10), and the similarity is determined. If it is equal to or less than the “judgment value” (second threshold value) (No in step S10), go to step 12.

On the other hand, if the similarity is larger than the second threshold (Y es in step S 10), it is determined that the comparative image 10 a ′ is similar to the search target image 10 a and displayed as a candidate on the monitor 6. (Step S 1 Do At this time, the similarity is also displayed. That is, the current search target image 10 a is displayed together with the similarity in the search result image display window の in FIG. 11. Then, the operator selects the search target image 10 0 extracted as a similar image in this manner. While observing a, you can change “Th resho 1 d” (first threshold) on the parameter overnight window (Fig. 10). The operator can also change the “judgment value” (second threshold value) on the parameter overnight window to an arbitrary value and click the “Apply” button. In this case, the search result image display window (Fig. 11) is redisplayed, and only the search target images having a similarity greater than the changed “judgment value” (second threshold) are displayed. Become.

Next, it is determined whether or not the search for all the search candidate images (in this case, 10a, '·, 10f) has been completed (step S12), and the search for all the search candidate images has been completed. If the processing has not been completed (No in step S12), it is determined whether or not the first threshold (Threshold) is requested to be changed by the operation (step S13), and the first threshold (Threshold) is determined. If a request to change Th resh 0 1 d) has been made (Yes in step S 13), change the first threshold (Th reshold) (step S 14) and proceed to step 15 . If there is no request to change the first threshold (Threshold) (No in step S13), the process directly proceeds to step 15. In step 15, the next search candidate image (for example, 10b) is selected, and the process proceeds to step S2.

If the search for all the search candidate images has been completed in Step S12 (Yes in Step S12), the search ends.

As described above, in the present image search apparatus 100, when the operator completes the necessary input and clicks the “Start” button in the parameter overnight window, all the search target images within the current search range are displayed. , 10b, "', and 10 are sequentially compared with the comparison original image 10a' one by one as shown in FIG. 9, and the result is as shown in FIG. The search result images are displayed sequentially in the search result image display window. When the search for 10 a, 10 b,..., And 10 f is completed, all the search target images determined to be similar are displayed in the search result image display window in FIG. Will be.

After the search for all the search target images in the search range is completed, the operator observes the search result image display window shown in FIG. It is determined whether the image is displayed as a similar image together with the degree.

Here, if favorable results are not obtained, Thresho 1 d (first threshold value), judgment value (second threshold value), image movement amount A h, ΔV, and integration area Change the position and range of, and repeat the image search process (Fig. 5) again from step S1.

For example, when it is considered that an accurate image search is not performed due to the influence of noise, it is conceivable to change the image movement amount Δh or ΔV. Furthermore, if a high similarity is shown despite images with low similarity, or conversely, if a low similarity is shown even though images with high similarity, T threshold ( It is conceivable to change the position or range of the first threshold value) or the integration region.

In addition, when the similarity is considered to indicate similarity in a favorable tendency, and when not only similar images but also images considered to be dissimilar images are displayed as similar images, the determination value (the Change the threshold (second threshold) to a higher value, and conversely, if the number of images displayed as similar images is too small, change the judgment value (second threshold) to a lower value Just do it. As described above, in the present embodiment, the image data of the comparison original image and the search target image are each converted into an optical image, and the optical image of the comparison original image and the search target image are juxtaposed to form a spatial light modulator. And write the written optical image pattern with coherent light, and Fourier transform the read modulated output. In other words, a correlation image (Fourier transform pattern) is formed, the relative position of the optical image between the comparison original image and the search target image is changed, and the optical image is formed with respect to the optical image before the relative position is changed. The difference between the correlation image and the correlation image formed for the optical image after the relative position has been changed is detected, and the output signal is integrated to detect the correlation value of the search target image with respect to the comparison original image. The correlation value is compared with a predetermined threshold value (second threshold value). As described above, according to the image search device of the present embodiment, a correlation signal between images can be directly generated, and further, speckle noise can be removed. Therefore, similar images can be quickly and accurately searched from a huge number of images. In addition, since the image data of the comparison original image and the search target image are each converted to an optical image and a correlation image is formed to obtain a correlation value, an optical operation is employed. In comparison, costs can be significantly reduced. In addition, since the optical calculation is employed, the calculation processing can be performed at extremely high speed, and the search can be performed in a short time even if the number of search target images is enormous.

The image search device and the image search method according to the present invention are not limited to the above-described embodiments, and various modifications and improvements can be made within the scope described in the claims.

For example, the magnetic disk 5 may be another type of external storage device such as a CDROM, or the network server 3 may be used directly instead of the magnetic disk 5. Further, the SLMs that can be used in the present invention are not limited to those shown in FIG. 2, and may have other configurations.

In addition, when the image search device and the image search method of the present invention are used for image diagnosis at a medical site, they can be used for X-ray film image search, cytology image search, helical scan CT image search, and the like. However, these images can be linked to electronic medical record information to keep track of past medical histories. Furthermore, besides its application in the medical field, in the field of security, when a suspicious person is photographed, it can be compared with a wanted person (especially an internationally problematic person). In this case, remove the outline of the face and examine. It is also useful in archeology, such as searching for similarities in excavated patterns.

As for the integration area, the difference results may be integrated in one continuous integration area as shown in each integration area (1) to (3) in Fig. 4 (b), or in a plurality of integration areas separated from each other. The difference results may be integrated and the integration results may be added. Further, integration may be performed over the entire area of the correlation image detected by the CCD image sensor 33. However, it is desirable to eliminate the zero-order light.

In S3 and S4 in FIG. 5, the correlation image is captured for a plurality of frames by the CCD image sensor 33, and the result of adding the luminance values (gradation values) of each pixel for all the frames to the correlation image (1). ) May be stored in the memory 18. In this case, the correlation image (2) is generated and stored in the memory 18 in the same manner in S6 and S7. In S8, the difference between the correlation images (1) and (2) thus obtained is taken. According to this method, since the difference between the pixels in the correlation images (1) and (2) can be obtained after the luminance value of each pixel is sufficiently increased, more accurate detection can be performed.

In the above embodiment, the similarity is obtained by counting the number of pixels whose luminance value is greater than Th resho 1 d. However, the similarity degree is obtained by counting the number of pixels whose luminance value is greater than Th resho 1 d. You may ask. Similarly, the similarity is determined to be similar when the similarity is greater than the determination value, but the similarity may be determined when the similarity is equal to or greater than the determination value.

In the above embodiment, the difference result of each pixel in the predetermined area is compared with Th resho 1 d, and a difference result larger than Th reshold is obtained. By counting the number of pixels having the pixel, the integration within the predetermined area was performed.However, it is also possible to simply add up the difference results of the pixels within the predetermined area without performing the comparison with Tresho 1 d. Good.

The coherent light source that irradiates the SLM to obtain the correlation signal is not limited to a laser diode, and various coherent light sources can be employed.

The means for changing the relative position of the image is not limited to the video RAM 8, and various image display control devices can be used.

In the above embodiment, the parameters Δίι, Δν, Threshold, judgment value, range, and position can be changed at any time during the search operation, but these parameters can be changed during the search operation. May be changed only when the search operation is terminated or interrupted. Industrial applicability

INDUSTRIAL APPLICABILITY The image search device and the image search method according to the present invention are widely used for searching similar images in the medical field, the security field, the archeology field, and the like.

Claims

The scope of the claims

1. Image data storage means for storing image data of a plurality of search target images;

Image data conversion means for converting each of the image data of one comparison original image and the image data of one of the plurality of search target images into an optical image,

A spatial light modulator that writes the optical image of the comparison original image and the optical image of the one search target image side by side, and reads out the written patterns of the two optical images with coherent light;

Correlation image forming means for performing a Fourier transform on a modulation output read from the spatial light modulator to form a correlation image on a correlation output surface;

By changing the relative position between the optical image of the comparison original image and the optical image of the one search target image and writing the two changed optical images to the spatial light modulator, the correlation image forming means Image position changing means for Fourier-transforming a modulation output read from the spatial light modulator to form a correlation image on the correlation output surface;

A correlation image formed for the two optical images before the relative position is changed by the image position changing unit; and the two correlation images after the relative position is changed by the image position changing unit. Difference detection means for obtaining a difference between the optical image and the correlation image formed,

Correlation value detection means for integrating a detection result of the difference detection means within a certain area on the correlation output surface to detect a correlation value of the one search target image with respect to the comparison original image;

The correlation value is compared with a predetermined threshold value, and based on the comparison result, it is determined whether the one search target image is a similar image similar to the comparison original image. Means for determining

An output unit that outputs the one search target image as a similar image when the comparison unit determines that the one search target image is a similar image; and controls the image data conversion unit to store the image data. The image data of at least one search target image among the image data of the plurality of search target images stored in the means is sequentially switched one by one to be converted into an optical image, whereby the at least one search is performed. Similar image extraction control means for sequentially comparing a target image with the comparison original image and outputting a search target image determined to be similar to the comparison original image as a similar image. Search device.

2. The image retrieval apparatus according to claim 1, further comprising an integration area specifying means for specifying the area where the correlation value detection means performs integration.

3. The image search device according to claim 2, wherein the integration region designation means designates a region that does not include zero-order light as the region.

4. The image search device according to claim 2, wherein the integration region designation means designates a region not including primary light as the region.

5. The image search device according to claim 1, further comprising a relative position change amount designating unit for designating a relative position change amount changed by the image position changing unit.

6. It further comprises a search range specifying means for selecting image data of the at least one search target image to be sequentially compared with the comparison original image from image data of the plurality of search target images. The image search device according to claim 1, wherein:

7. The correlation image forming means,

First Fourier transform means for performing a Fourier transform on a modulation output read from the spatial light modulator to form a first Fourier transform pattern; Another spatial light modulator that writes the first Fourier transform pattern and reads out the written Fourier transform pattern with coherent light;

2. The image retrieval apparatus according to claim 1, further comprising second Fourier transform means for Fourier-transforming a modulation output read from said another spatial light modulator to form a correlation image.

8. An image-to-data conversion process for converting the image data of one comparison original image and the image data of one search target image into optical images, respectively.

The optical image of the comparison original image and the optical image of the one search target image are juxtaposed and written into the spatial light modulator, and a pattern of the two written optical images is written by the coherent light from the spatial light modulator. A modulation step for reading, a Fourier transform of a modulation output read from the spatial light modulator in the modulation step, a correlation image forming step of forming a correlation image on a correlation output surface, and an optical image of the comparison original image and The relative position of one search target image with respect to the optical image is changed, the changed two optical images are written in the spatial light modulator, and the modulation output read from the spatial light modulator is subjected to Fourier transform. An image position changing step of forming a correlation image on the correlation output surface using the two optical images before a relative position is changed in the image position changing step. A difference detecting step of obtaining a difference between the two optical images after the relative position is changed in the image position changing step and a correlation image formed,

A correlation value detection step of integrating a detection result of the difference detection step within a certain area on the correlation output surface and detecting a correlation value of the one search target image with respect to the comparison original image;

A comparing step of comparing the correlation value with a predetermined threshold value, and determining whether the one search target image is a similar image similar to the comparison original image based on the comparison result; Outputting the one search target image as a similar image when the one search target image is determined to be a similar image in the comparing step; and sequentially switching image data of at least one search target image. By repeatedly executing the image data conversion step, the at least one search target image is sequentially compared with the comparison original image, and the search target image determined to be similar to the comparison original image is output as a similar image. A similar image extraction control step of causing the image to be retrieved.