JP5068036B2 - Spectral image processing device - Google Patents

Description

  The present invention relates to a spectral image processing method, apparatus, and program for generating a processed image using a spectral image obtained by photographing an object to be measured with an endoscope and acquired for each wavelength.

  In endoscopy, tissue properties and lesions are read from shape changes such as bumps, depressions and surface patterns, and subtle changes in color tone. Whether or not the shape change can be recognized depends largely on the identification of the wavelength of light from the affected area and the observation method. Conventionally, a method of specifying and observing a wavelength using an optical filter has been proposed, but this method can only obtain an image in one wavelength region, and in order to observe at another wavelength, the filter must be replaced. is necessary. Therefore, in recent years, a technique in which a spectral estimation technique for estimating and imaging an image at an arbitrary wavelength from a normal image by computer processing has been developed for endoscopy. Then, by selecting the wavelength of light that the doctor wants to observe according to the contents of the examination, a spectral image of the wavelength selected from the endoscopic image can be viewed.

  By the way, not only the above-described endoscopic image but also a spectroscopic analysis method has been proposed for fundus examination (see, for example, Patent Document 1). In Patent Document 1, spectral data (reference pixel images) of various lesions are stored in advance in the lesion library for each lesion, and there are reference pixel images stored in the lesion library in the input spectral data. By detecting whether or not to do so by pattern matching, the position of the lesioned part in the spectroscopic data is specified, and the identified lesioned part is emphasized and imaged and displayed on the monitor.

Furthermore, as a method of performing image processing on multispectral image data for each color wavelength, three types of virtual spectral aperture images are created by adding different weights to each spectral channel image of the multispectral data, and these three types of virtual spectral aperture images Has been proposed (see Patent Document 2). In this method, when a color processed image is generated from a virtual spectrum aperture image, the user performs it while confirming the image on the screen.
JP 2005-296434 A Japanese Patent Laid-Open No. 11-154220

  As described above, in Patent Document 1, when an image of a lesion is to be displayed on the screen, it is necessary to create a lesion library in which a reference pixel image of the lesion is stored in advance. However, when there is a variation in color of a lesion due to a difference in patients, a difference in color due to a lesion, a variation in color due to a difference in observation model, etc., the lesion library stores only one reference pixel image for each lesion. There is a problem that a lesion may not be detected. On the other hand, in order to accurately detect a lesioned part, it is necessary to store a large number of reference pixel images for one lesion, and there is a problem that a great deal of labor is required for creating a database.

  Furthermore, as in Patent Document 2, if a user has to operate to create a color processed image, there is a problem that it is difficult to confirm the color processed image during imaging of a lesioned part.

  Accordingly, an object of the present invention is to provide a spectral image processing method, apparatus, and program capable of automatically creating a processed image that accurately represents the characteristics of a measurement target such as a lesion.

  The spectral image processing method of the present invention extracts and extracts feature amounts from a plurality of spectral images in an image processing method for generating a processed image composed of primary color signals using a plurality of spectral images obtained by photographing a measurement object. A plurality of spectral images are evaluated using the feature amount, a spectral image having a high evaluation is selected from the plurality of spectral images, and a processed image is generated by assigning the selected spectral image to a primary color signal. To do.

  The spectral image processing apparatus of the present invention is a feature quantity for extracting a feature quantity from a plurality of spectral images in an image processing apparatus that generates a processed image composed of primary color signals using a plurality of spectral images obtained by photographing a measurement object. An extraction unit, an image selection unit that evaluates a plurality of spectral images using the feature amounts extracted by the feature amount extraction unit, and selects a spectral image having a high evaluation among the plurality of spectral images, and a selection by the image selection unit And an image generation means for generating a processed image by assigning the spectral image thus obtained to a primary color signal.

  A spectral image processing program of the present invention is a spectral image processing program for causing a computer to generate a processed image composed of a plurality of primary color signals using a plurality of spectral images obtained by photographing a measurement object. Feature values are extracted from a plurality of spectral images, a plurality of spectral images are evaluated using the extracted feature amounts, a spectral image having a high evaluation is selected from the plurality of spectral images, and the selected spectral image is assigned to a primary color signal. Thus, a processed image is generated.

  Here, the image selection means may select one spectral image, or may select a plurality of spectral images. The processed image may use any color system, for example, an RGB color system composed of primary color signals of red, green, and blue, or a color system of two colors such as green and blue. It may consist of.

  The extracted feature quantity may be any kind, for example, edge intensity, or extracted from a plurality of spatial frequency transformed images that have been spatial frequency transformed by Fourier transform or wavelet transform. , Brightness, color, hue, saturation, texture information, or contrast. The spatial frequency converted image is, for example, a plurality of power spectrum images obtained by performing Fourier transform on each spectral image or a plurality of multi-resolution images obtained by performing wavelet transform on each spectral image. When extracting the feature amount for each of the plurality of spectral images, the feature amount may be extracted from a spatial frequency converted image in a predetermined spatial frequency band among the plurality of multi-resolution images.

  Note that the evaluation of the feature amount is an evaluation of whether or not it includes many characteristic structures indicated by the feature amount, and high evaluation means a relative rank when a plurality of spectral images are evaluated using the feature amount. Say. Here, the image selection means may select a plurality of spectral images in descending order of evaluation of the feature amount, for example, as long as it selects the spectral image using the evaluation of the feature amount. At this time, the image processing unit may assign the R, G, and B primary color signals in order from the longest spectral wavelength of the plurality of spectral images selected by the image selection unit. In addition, when the image selection means selects a plurality of spectral images, it is preferable to select spectral images each having a spectral wavelength separated by 30 nm or more.

  Alternatively, the image selection means may select the spectral image having the highest evaluation for each wavelength band of the R, G, and B primary color signals. At this time, the image processing means may assign a plurality of spectral images selected by the image selection means to R, G, and B primary color signals, respectively. In this case as well, it is preferable that the image selecting means selects a spectral image having a spectral wavelength of 30 nm or more when selecting a plurality of spectral images.

  Furthermore, the feature quantity to be extracted may be one or plural. When the feature quantity extraction unit extracts a plurality of feature quantities from each spectral image, it is preferable to extract a plurality of feature quantities having different properties. For example, the above-described edge intensity, the one extracted by performing spatial frequency conversion, the differential value between spectral images separated by a certain wavelength, or the like may be used as the feature amount.

  At this time, the image selection means may evaluate a plurality of spectral images for each of a plurality of feature amounts, and select a spectral image having a high evaluation among the plurality of spectral images for each feature amount. The image generation means may generate a processed image by assigning a plurality of spectral images respectively selected for a plurality of feature amounts to primary color signals.

  For example, the image selection means may select a spectral image having the highest evaluation for each of a plurality of feature amounts. At this time, the image generation unit may assign the R, G, and B primary color signals in order from the longest spectral wavelength of the plurality of spectral images selected by the image selection unit.

  Alternatively, the image selection unit may select a plurality of spectral images for each feature amount in descending order of the feature amount evaluation. At this time, the image generation means may assign the spectral image having the same rank evaluation to each feature amount to the R, G, and B primary color signals.

  Further, the image processing means may select the spectral image having the highest evaluation for each wavelength band of each primary color signal for each of a plurality of feature amounts. At this time, the image generation means may assign a sum of a plurality of spectral images of each feature amount selected for each wavelength band of each primary color signal to each primary color signal.

  Note that when the image selection unit selects a spectral image using each feature amount, it is preferable to select a spectral image having a spectral wavelength of 30 nm or more.

  According to the spectral image processing method, apparatus, and program of the present invention, feature amounts are extracted from a plurality of spectral images, a plurality of spectral images are evaluated using the extracted feature amounts, and a high evaluation among the plurality of spectral images. By selecting a spectral image and generating a processed image by assigning a spectral image for each selected wavelength band to the primary color signal, without using a database indicating the relationship between the type of lesion and the spectral distribution as in the past It is possible to automatically select a spectral image effective for observation of the measurement object and generate a processed image.

  The image selection unit selects a plurality of spectral images in descending order of evaluation of the feature amount, and the image processing unit selects R, in order from the longest spectral wavelength of the plurality of spectral images selected by the image selection unit. When assigned to the primary color signals of G and B, a spectral image having the most characteristic structure effective for observing the measurement target indicated by the feature value is selected to generate a processed image suitable for the observation of the measurement target. can do.

  Further, the image selection means selects the spectral image having the highest evaluation for each wavelength band of the R, G, and B primary color signals, and the image processing means respectively selects a plurality of spectral images selected by the image selection means. If assigned to the primary color signals of R, G, and B, the spectral image having the most characteristic structure effective for observing the measurement target indicated by the feature value is selected and measured for each wavelength band of the primary color signal. A processed image suitable for observing a measurement object close to the actual color of the object can be generated.

  In addition, the feature amount extraction unit extracts a plurality of feature amounts from each spectral image, and the image selection unit evaluates the plurality of spectral images using each feature amount, thereby evaluating the plurality of spectral images. High spectral images for each feature amount, and the image generation means generates a processed image by assigning a plurality of spectral images selected for each of the plurality of feature amounts to primary color signals. Since the processed image is generated from the plurality of spectral images, it is possible to automatically generate a processed image that is more suitable for observation of the measurement target.

  The image selection means selects the spectral image with the highest evaluation for each of the plurality of feature amounts, and the image generation means sequentially selects the spectral images having the longest spectral wavelengths selected by the image selection means. If it is assigned to the primary color signals of R, G, and B, it is suitable for observation of the measurement object by selecting the spectral image having the most characteristic structure effective for observation of the measurement object indicated by the plurality of feature values. A processed image can be generated.

  Further, the image selection unit selects a plurality of spectral images for each feature amount in descending order of the feature amount evaluation, and the image generation unit adds R spectral images having the same rank in the evaluation of each feature amount. , G, and B are assigned to the primary color signals, the spectral image having the most characteristic structure effective for observing the measurement target indicated by the plurality of feature amounts is selected for each wavelength band of the primary color signals. A processed image suitable for observing the measurement object close to the actual color of the measurement object can be generated.

  Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing a preferred embodiment of the spectral image processing apparatus of the present invention. The configuration of the spectral image processing apparatus 1 as shown in FIG. 1 is realized by executing a spectral image processing program read into the auxiliary storage device on a computer (for example, a personal computer). At this time, the spectral image processing program is stored in an information storage medium such as a CD-ROM or distributed via a network such as the Internet and installed in a computer. The configuration of the spectral image processing apparatus 1 as shown in FIG. 1 is incorporated in the endoscope system itself, not limited to the case where it is realized on the above-described computer, and imaging with the endoscope and display of the processed image are processed in real time. It may be something like this.

  The spectral image processing apparatus 1 generates a processed image CP composed of a plurality of primary color signals using a plurality of spectral images SP obtained by photographing a measurement target such as a lesion using an endoscope, for example. A spectral image acquisition unit 10 that acquires a plurality of spectral images SP, a feature amount extraction unit 20 that extracts a feature amount X from the plurality of spectral images SP, and a feature amount X of each spectral image SP, thereby evaluating each primary color. An image selection unit 30 that selects the spectral image SP having the highest evaluation for each wavelength band of the signal, and an image generation unit 40 that generates the processed image CP by assigning the selected spectral image SP for each wavelength band to the primary color signal. And.

The spectral image acquisition means 10 acquires spectral images SP _{1 to} SP _k (k: the number of spectral images) for each wavelength of light reflected from the measurement object. For example, the spectral image acquisition means 10 A plurality of spectral images SP _{1 to} SP _k acquired in a wavelength band from 400 nm to 700 nm (visible region) as shown in 2 at intervals of 5 nm are acquired. Note that the plurality of spectral images SP _{1 to} SP _k are not limited to those obtained by spectroscopically reflecting light reflected from the measurement target, but use an endoscope as disclosed in Japanese Patent Laid-Open No. 2003-93336. Alternatively, it may be a spectral estimation image generated for each wavelength from an image made up of RGB signals acquired by conventional photographing.

The feature quantity extraction unit 20 extracts the feature quantity X from each of the spectral images SP _{1 to} SP _k , and has a function of extracting, for example, the edge intensity of the spectral image SP as the feature quantity X. Specifically, when there is a spectral image SP as shown in FIG. 3A, the feature quantity extraction means 20 extracts the edge component of the spectral image SP as shown in FIG. It is designed to detect the intensity of. Then, the evaluation of a plurality of spectral images SP _{1 to} SP _k (spectral wavelengths) at the edge intensity as shown in FIG.

Note that the feature amount extraction unit 20 illustrates the case where the edge strength is extracted as the feature amount X. However, by performing a Fourier transform process on the spectral image SP as shown in FIG. From the three-dimensional relationship between the spatial frequency and the number of pixels in the plurality of spectral images SP _{1 to} SP _k (spectral wavelengths) as shown in FIG. The number of pixels of the power spectrum image in a predetermined spatial frequency band or the density dispersion may be extracted as the feature amount X2 from among the power spectrum images.

Alternatively, by applying wavelet transform processing to the spectral image SP as shown in FIG. 5A, a plurality of multi-resolution images as shown in FIG. 5B are generated, and a plurality as shown in FIG. From the three-dimensional relationship of the spatial frequency and the number of pixels with respect to the spectral images SP _{1 to} SP _k (spectral wavelength), for example, the number of pixels or density dispersion of a multi-resolution image in a predetermined spatial frequency band among a plurality of multi-resolution images. May be extracted as the feature amount X2.

Furthermore, FIG. 6 (A), the may be extracted the differential value of the spectral image SP _m spectral image SP m _{+ n} apart by a predetermined wavelength from the spectral image SP _m of Toko as features X3. Then Evaluation of FIG plurality of spectral images _SP 1 to SP _k (spectral wavelength) with respect to differential value as shown in (B) is found.

The image selection means 30 evaluates a plurality of spectral images SP using the extracted feature amount X, and selects a spectral image SP used to generate the processed image CP from among the plurality of spectral images SP. A plurality of spectral images SP are selected in descending order of evaluation of the amount X. In other words, when the feature amount X is extracted as shown in FIG. 3, the image selection means 30 rearranges the spectral images SP in the order of evaluation of the feature amount (in descending order of edge strength) as shown in FIG. The image selection means 30 selects the top three spectral images SP _D , SP _B , and SP _C that are highly evaluated.

Note that the image selection means 30 may select spectral images having a spectral wavelength of 30 nm or more when selecting a plurality of spectral images SP _D , SP _B , and SP _C. For example, when the spectral wavelength of the spectral image SP _D is 680 nm and the spectral wavelength of the spectral image SP _C is 660 nm, the spectral images SP _C and SP _D are not separated by 30 nm or more. Instead of _C, the spectral image SP _A having the next highest evaluation is selected.

The image processing unit 40 generates the processed image CP by assigning the spectral image SP selected by the image selecting unit 30 to the primary color signal. For example, the selected three spectral images SP _D , SP _B , SP _C In order from the longest spectral wavelength of (SP _A ), assigned processed images CP are generated for the primary color signals of R, G, and B. Specifically, as shown in FIG. 8, the image generation unit 40 adds the R signal, the G signal, and the three spectral images SP _D , SP _B , and SP _C (SP _A ) selected by the image selection unit 30, respectively. Assign the B signal. For example, the pixel data of the spectral image SP _{D is} the pixel data of the R signal, the pixel data of the spectral image SP _C (SP _B ) is the pixel data of the G signal, and the pixel data of the spectral image SP _B (SP _A ) is the B signal. A processed image CP composed of RGB color images is generated as pixel data.

Note that when the processed image CP is generated, a known image processing technique such as level adjustment obtained by multiplying each assigned RGB signal by a gain coefficient may be used. Further, the image processing means 40 assigns each of the three spectral images SP _D , SP _B , SP _C (SP _A ) to the R, G, B primary color signals in order from the longest spectral wavelength. R, G, and B primary color signals may be automatically assigned in the descending order of evaluation without depending on the spectral wavelengths of SP _D , SP _B , and SP _C (SP _A ).

In this manner, by selecting the spectral images SP _D , SP _B , and SP _C (SP _A ) for each primary color signal using the feature amount X and combining them to generate the processed image CP, as in the conventional case. Since a database in which the spectral distribution is recorded for each type of lesion is not required, the work for creating the database can be eliminated. Furthermore, even if there is a variation in the color of the lesion due to patient differences, a color change due to a lesion, or a color variation due to differences in the observation model, the characteristic structure of the lesion is extracted as a feature amount X. Since the processed image CP is generated using the spectral image SP selected using the feature amount X, the processed image CP representing the characteristic structure of the lesion can be generated with high accuracy.

  That is, the lesion image has a variation in color of the lesion due to a difference in patients, a change in color due to a lesion, a variation in color due to a difference in observation model, and the like. Therefore, even if an image of a lesioned part is stored in advance as an image library and a lesioned part is detected using template matching or the like, there may be a case where the lesioned part cannot be detected due to the variation. On the other hand, in the spectral image processing apparatus 1 described above, since the spectral wavelength is selected using the feature amount X in consideration of the characteristic structure of the observation site, the spectral image processing apparatus 1 is most suitable for observation without being influenced by the above-described variation. A processed image can be generated using the spectral image SP.

  In addition, since the spectral wavelength assigned to the primary color signal is automatically selected, there is no need to manually select an effective spectral image as in the prior art, and the spectral wavelength is selected in near real time even during inspection. The spectral wavelength can be switched quickly for each observation site.

Further, a plurality of spectral images SP are selected in descending order of evaluation of the feature amount X, and the primary colors of R, G, and B are sequentially selected in descending order of spectral wavelengths of the plurality of spectral images SP _D , SP _B , SP _C (SP _A ). By assigning to the signal, it is possible to select the spectral image SP having the most characteristic structure effective for the observation of the measurement target indicated by the feature amount X and generate the processed image CP suitable for the observation of the measurement target. . Therefore, the spectral image SP can be selected by changing the type of feature amount extracted for each measurement target, and a processed image CP suitable for the type of measurement target can be generated.

FIG. 9 is a flowchart showing a preferred embodiment of the spectral image processing method of the present invention. The spectral image processing method will be described with reference to FIGS. First, a plurality of spectral images SP _{1 to} SP _k to be measured are acquired by the spectral image acquisition means 10 (see step ST1, FIG. 2). Next, the feature amount extraction means 20 extracts feature amounts (edge strength) X from the plurality of spectral images SP _{1 to} SP _k , respectively (step ST2, see FIGS. 3 to 6). Then, in the image selection means 30, the spectral images SP _D , SP _B , and SP _C are selected in descending order of the plurality of feature amounts X (see step ST3, FIG. 7). As described above, when selecting a plurality of spectral images SP, spectral images SP _D , SP _B and SP _A each having a spectral wavelength of 30 nm or more may be selected. Thereafter, in the image generating means 40, each spectral image SP _D , SP _B , SP _C (SP _A ) is assigned to the R signal, G signal, and B signal (step ST4), and the processed image CP is generated (step ST4). Step ST5, see FIG. 8). As a result, even if there is a variation in the color of the lesion due to a difference in the patient, a color change due to a lesion, a color variation due to a difference in the observation model, etc., the characteristic structure of the lesion is extracted as the feature amount X, Since the processed image CP is generated using the spectral images SP _D , SP _B , and SP _C (SP _A ) selected using the feature amount X, the processed image CP representing the characteristic structure of the lesion is accurately obtained. Can be generated.

Here, in the above embodiment, the spectral images SP _D , SP _B , and SP _C (SP _A ) used for generating the processed image CP are selected in descending order of evaluation, but the primary color signals of R, G, and B are selected. The spectral image CP having the highest evaluation may be selected for each wavelength band. That is, the image selection unit 30 evaluates each of the plurality of spectral images SP _{1 to} SP _k using the feature amount X extracted by the feature amount extraction unit 20, thereby obtaining each primary color signal (R signal, G signal, B signal). The spectral image SP having the highest evaluation within the wavelength band is provided for each primary color signal. For example, as shown in FIG. 7, it is assumed that the evaluation of the feature amount X increases in the order of spectral images SP _D (680 nm), SP _B (520 nm), SP _C (660 nm), and SP _A (450 nm). Then, the image selection means 30 has the feature value X most evaluated for each wavelength band (R signal: about 600 nm to 700 nm, G signal: about 500 nm to 600 nm, B signal: about 400 nm to 500 nm) in which a plurality of primary color signals can be expressed. A high spectral image SP is selected. That is, SP _D is selected in the wavelength band of the R signal, SP _B is selected in the wavelength band of the G signal, and SP _A is selected in the wavelength band of the B signal. Even in this case, when selecting the plurality of spectral images SP, the image selecting means 30 selects spectral images whose spectral wavelengths are separated by 30 nm or more.

Then, the image processing unit 40 generates a processed image CP by assigning the plurality of spectral images SP _D selected by the image selecting unit 30 to R signals, the spectral image SP _B to G signals, and the spectral image SP _A to primary signals of B signals. To do. Here, the spectral image SP is selected after all the spectral images SP have been evaluated, but a plurality of spectral images SP _{1 to} SP _k are selected from the wavelength band of the R signal, the wavelength band of the G signal, and the wavelength of the B signal. The spectral image SP having the highest evaluation may be selected from the three classifications.

Even in this case, the processing for creating the database can be eliminated, and the processed image CP representing the characteristic structure of the lesion can be accurately generated, and the spectral wavelength can be measured in near real time even during the examination. Can be selected, and the spectral wavelength can be quickly switched for each observation site. Further, the spectral image SP _D , SP _B , SP _A having the highest evaluation is selected for each wavelength band of the R, G, B primary color signals, and the plurality of spectral images SP are respectively converted into R, G, B primary color signals. By assigning the spectral image SP having the most characteristic structure effective for observing the measurement target indicated by the feature amount X for each wavelength band of the primary color signal, the measurement target close to the actual color of the measurement target The processed image CP suitable for the observation can be generated.

  FIGS. 10 to 13 are schematic views showing another embodiment of processing image generation in the spectral image processing apparatus of the present invention, and the processing image generation embodiment will be described with reference to FIGS. 10 to 13. .

In FIG. 10, the feature quantity extraction means 20 extracts the plurality of feature quantities X, X1, and X2 described above. Note that FIG. 10 illustrates the case where three feature amounts X, X1, and X2 are extracted, but two or more feature amounts are extracted such that four feature amounts including the feature amount X3 are used. If it is. Next, the image selection unit 30 evaluates the plurality of spectral images SP _{1 to} SP _k using the plurality of feature amounts X, X1, and X2, respectively, and creates an evaluation ranking as shown in FIG.

Then, a spectral image SP having the highest evaluation is selected for each feature amount X, X1, and X2, and a processed image CP is generated using the selected spectral image SP. Specifically, in FIG. 10, the image selection unit 30 selects the spectral images SP _D , SP _E , and SP _P having the highest evaluation for each of the plurality of feature amounts X, X1, and X2. Then, the image generating unit 40 includes a plurality of spectral images _SP D _selected, SP E, in order to SP _P from the longest spectral wavelength, assigned to R, G, primary color signals B. Specifically, as shown in FIG. 11, for example, the image generating means 40 assigns the spectral image SP _D to the R signal, assigns SP _E to the G signal, assigns SP _P to the B signal, and generates a processed image CP. When the image selection means 30 selects a plurality of spectral images SP, the spectral images SP _D , SP _E , and SP _P having spectral wavelengths separated by 30 nm or more are selected.

Even in this case, the processing for creating the database can be eliminated, and the processed image CP representing the characteristic structure of the lesion can be accurately generated, and the spectral wavelength can be measured in near real time even during the examination. Can be selected, and the spectral wavelength can be quickly switched for each observation site. Further, the spectral images SP _D , SP _E , SP _P having the highest evaluation among the feature amounts X, X 1, X 2 are selected, and a plurality of spectral images SP _D , SP _E , SP _P are respectively converted into R, G, B primary color signals. By assigning to, it is possible to generate a processed image CP using the spectral images SP _D , SP _E and SP _P having the most characteristic structure effective for observing the measurement target indicated by the feature amount X.

Alternatively, as shown in FIGS. 10 and 12, the image selection unit 30 may select a plurality of spectral images SP for each feature amount X, X1, and X2. Specifically, the image selection means 30 selects three spectral images SP that are highly evaluated in the feature amounts X, X1, and X2. Specifically, the image selecting means 30 uses the spectral images SP _D , SP _B , SP _{C for} the feature quantity X, and the spectral images SP _E , SP _F , SP _{G for} the feature quantity X 1 and the spectral images SP _P , SP for the feature quantity X 2. _Q, to select the SP _R. Note that when the image selection unit 30 selects a plurality of spectral images SP for each of the feature amounts X, X1, and X2, a spectral image SP having a spectral wavelength of 30 nm or more is selected. Then, as shown in FIG. 12, the image generation unit 40 adds the spectral images SP in which the evaluations of the feature amounts X, X1, and X2 are in the same order. That is, the image generation means 40 has the highest evaluation spectral images SP _E , SP _P , SP _G , the second highest evaluation spectral image SP _B , SP _F , SP _Q , and the third highest evaluation spectral image SP _E , SP _P and SP _G are added respectively. Then, for example, the processed image CP is generated by assigning the primary color signals of R, G, and B in descending order of evaluation.

Even in this case, the processing for creating the database can be eliminated, and the processed image CP representing the characteristic structure of the lesion can be accurately generated, and the spectral wavelength can be measured in near real time even during the examination. Can be selected, and the spectral wavelength can be quickly switched for each observation site. Further, the spectral images SP _D , SP _E , SP _P having the highest evaluation among the feature amounts X, X 1, X 2 are selected, and a plurality of spectral images SP _D , SP _E , SP _P are respectively converted into R, G, B primary color signals. By assigning to, it is possible to generate a processed image CP using the spectral images SP _D , SP _E and SP _P having the most characteristic structure effective for observing the measurement target indicated by the feature amount X.

Further, the spectral image SP having the highest evaluation may be selected for each wavelength band of the primary color signal (R signal, G signal, B signal). For example, as shown in FIGS. 10 and 13, spectral image SP _D is selected for ranking using feature amount X in the wavelength band of the R signal, spectral image SP _E is selected for feature amount X1, and spectral image SP _P is selected for feature amount X2. Is done. Similarly, the feature quantity X in spectral image SP _B in the wavelength band of the G signal, the feature amount X1 in the spectroscopic image SP _G, the feature amount X2 in the spectroscopic image SP _R is selected added. Further, in the wavelength band of the B signal, the spectral image SP _A is selected for the feature amount X, the spectral image SP _H is selected for the feature amount X1, and the spectral image SP _S is selected for the feature amount X2. The image generating means 40 generates the processed image CP by assigning the spectral image SP added for each wavelength band selected for each of the plurality of feature amounts X, X1, and X2 to the primary color signal. In this way, the spectral image SP having the highest evaluation is selected for each of the plurality of feature amounts X, X1, and X2 for each wavelength band of each primary color signal, and each feature amount X selected for each wavelength band of each primary color signal. , X1 and X2 are added to each primary color signal, and a plurality of spectral images are selected on the basis of each feature amount in each wavelength band to generate a processed image. A spectral image can be selected with high accuracy, and a processed image suitable for observation of a measurement object can be automatically generated.

According to the foregoing embodiments, by each of a plurality of spectral images SP ₁ to SP _k extracts the feature quantity X, evaluating multiple spectral images SP ₁ to SP _k by using the extracted feature quantity X, By selecting a highly evaluated spectral image SP and assigning the selected spectral image SP for each wavelength band to a primary color signal, the processed image CP is generated, so that the relationship between the type of lesion and the spectral distribution can be obtained as in the past. Without using the database shown, it is possible to automatically select a spectral image effective for observation of the measurement object and generate a processed image.

  Embodiments of the present invention are not limited to the above embodiments. For example, the feature amount extraction unit 20 illustrates the case where the edge strength is detected as the feature amount or is calculated using spatial frequency conversion. For example, brightness, color, hue, saturation, texture information, contrast, etc. You may make it extract.

  Furthermore, the feature quantity extraction unit 20 in FIG. 1 may have a function of changing the feature quantity X used for each measurement target from which the spectral image SP is acquired. Accordingly, the spectral image SP can be selected using the feature amount X that most indicates the characteristic structure of each measurement site (affected part). Further, although the case where the primary color signal is composed of RGB signals is illustrated, it may be composed of, for example, a CYMG color system composed of primary color signals of cyan, yellow, magenta, and green.

1 is a block diagram showing a preferred embodiment of a spectral image processing apparatus of the present invention. The schematic diagram which shows an example of the some spectral image acquired by the spectral image acquisition means of FIG. Schematic diagram showing an example of a feature quantity extracted by the feature quantity extraction means of FIG. The schematic diagram which shows another example of the feature-value extracted by the feature-value extraction means of FIG. The schematic diagram which shows another example of the feature-value extracted by the feature-value extraction means of FIG. The schematic diagram which shows another example of the feature-value extracted by the feature-value extraction means of FIG. Schematic diagram showing how a spectral image is selected in the image selection means of FIG. The schematic diagram which shows a mode that a process image is produced | generated by the image production | generation means of FIG. The flowchart which shows preferable embodiment of the spectral image processing method of this invention The schematic diagram which shows another embodiment about the production | generation of the process image in the spectral image processing apparatus of this invention The schematic diagram which shows another embodiment about the production | generation of the process image in the spectral image processing apparatus of this invention The schematic diagram which shows another embodiment about the production | generation of the process image in the spectral image processing apparatus of this invention The schematic diagram which shows another embodiment about the production | generation of the process image in the spectral image processing apparatus of this invention

Explanation of symbols

1 spectral image processing apparatus 10 spectral image obtainment means 20 feature amount extraction unit 30 image selecting unit 40 image generating unit CP processing image _SP 1 to SP _k spectral image X, X1, X2, X3 feature quantity

Claims (11)

In an image processing apparatus that generates a processed image composed of primary color signals of R, G, and B using a plurality of spectral images obtained by photographing a measurement target,
Feature amount extraction means for extracting a feature amount from each of the plurality of spectral images;
An image selection unit that evaluates the plurality of spectral images using the feature amount extracted by the feature amount extraction unit, and selects three spectral images in order of the high evaluation from the plurality of spectral images;
Image generating means for generating the processed image by assigning the three spectral images selected by the image selecting means to the primary color signals of R, G, and B in order from the longest spectral wavelength. A spectral image processing apparatus. In an image processing apparatus that generates a processed image composed of primary color signals of R, G, and B using a plurality of spectral images obtained by photographing a measurement target,
Feature amount extraction means for extracting a feature amount from each of the plurality of spectral images;
The plurality of spectral images are evaluated using the feature amounts extracted by the feature amount extraction means, and the wavelength of the primary color signal is determined for each of the R, G, and B primary color signals from the plurality of spectral images. An image selection means for selecting a spectral image having a spectral wavelength within the band and having the highest evaluation;
Spectral characteristics, comprising: an image generating means for generating the processed image by assigning each spectral image selected by the image selecting means to a primary color signal in a wavelength band to which the spectral wavelength of the spectral image belongs. Image processing device. In an image processing apparatus that generates a processed image composed of primary color signals of R, G, and B using a plurality of spectral images obtained by photographing a measurement target,
  Feature amount extraction means for extracting a feature amount from each of the plurality of spectral images;
  The plurality of spectral images are evaluated using the feature amounts extracted by the feature amount extraction unit, and (a) a spectral image having the highest evaluation is selected as the first spectral image from the plurality of spectral images. And (b) among the plurality of spectral images excluding the selected first spectral image and having a spectral wavelength separated by 30 nm or more from the spectral wavelength of the first spectral image. The spectral image with the highest evaluation is selected as a second spectral image from (c) the plurality of spectral images excluding the first and second spectral images, and the spectral wavelength thereof is the first spectral image. Image selecting means for selecting, as a third spectral image, a spectral image having the highest evaluation from spectral images separated by 30 nm or more from the spectral wavelengths of the first and second spectral images;
  The processed image is generated by assigning the first, second, and third spectral images selected by the image selection unit to the R, G, and B primary color signals in order from the longest spectral wavelength. Image generation means and
  A spectral image processing apparatus comprising: The spectral image processing apparatus according to claim 1, wherein the feature amount is edge strength. The spectral image processing apparatus according to claim 1, wherein the feature amount is extracted by converting the spectral image into a spatial frequency. In an image processing apparatus that generates a processed image composed of a plurality of primary color signals using a plurality of spectral images obtained by photographing a measurement object,
Feature quantity extraction means for extracting the same quantity of feature quantities as the number of primary color signals from each of the plurality of spectral images;
For each feature quantity extracted by the feature quantity extraction unit, the plurality of spectral images are evaluated using the feature quantity, and the spectral image having a high evaluation among the plurality of spectral images is evaluated for each feature quantity. An image selection means to select;
A spectral image processing apparatus comprising: an image generation unit configured to generate the processed image by allocating the plurality of spectral images selected by the image selection unit to the plurality of primary color signals. The image selection unit selects the spectral image having the highest evaluation for each of the plurality of feature amounts, and the image generation unit converts the plurality of spectral images selected by the image selection unit to have a long spectral wavelength. 7. The spectral image processing apparatus according to claim 6, wherein the spectral image processing apparatus is assigned to the primary color signals of R, G, and B in order from the first one. The image selection means selects a plurality of the spectral images for each feature quantity in descending order of the evaluation of the feature quantity, and the image generation means selects the spectral images having the same rank in the evaluation of the feature quantities. 7. The spectral image processing apparatus according to claim 6, wherein the sum is assigned to the R, G, and B primary color signals. The image processing means selects the spectral image having the highest evaluation for each wavelength band of each primary color signal for each of the plurality of feature amounts, and the image generation means for each wavelength band of the primary color signals. 7. The spectral image processing apparatus according to claim 6, wherein a sum of a plurality of the spectral images of the selected feature quantities is assigned to the primary color signals. In an image processing apparatus that generates a processed image composed of primary color signals of R, G, and B using a plurality of spectral images obtained by photographing a measurement target,
  Feature quantity extraction means for extracting three feature quantities from each of the plurality of spectral images;
  For each feature quantity extracted by the feature quantity extraction unit, the plurality of spectral images are evaluated using the feature quantity, and (a) one of the three feature quantities from the plurality of spectral images. A spectral image having the highest evaluation regarding the first feature amount is selected as a first spectral image. (B) Among the plurality of spectral images, a spectral wavelength of the spectral image of the first spectral image is selected. The highest evaluation is made on the second feature value, which is one of the two feature values excluding the first feature value among the three types of feature values, in the spectral image separated by 30 nm or more from the wavelength. A spectral image is selected as a second spectral image; (c) a spectral image of the plurality of spectral images whose spectral wavelength is 30 nm or more away from the spectral wavelength of the first and second spectral images; Among the three types of feature quantities, the first An image selection means for selecting the evaluation highest spectral image for the third feature quantity is a pre-feature amount other than the second characteristic amount as a third spectral image,
  The processed image is generated by assigning the first, second, and third spectral images selected by the image selection unit to the R, G, and B primary color signals in order from the longest spectral wavelength. Image generation means and
  A spectral image processing apparatus comprising: In an image processing apparatus that generates a processed image composed of primary color signals of R, G, and B using a plurality of spectral images obtained by photographing a measurement target,
  Feature amount extraction means for extracting a feature amount from each of the plurality of spectral images;
  The plurality of spectral images are evaluated using the feature amounts extracted by the feature amount extraction means, and (a) a first of the plurality of spectral images and any one of the R, G, and B A spectral image having the highest evaluation is selected as a first spectral image from spectral images having spectral wavelengths within a wavelength band of one primary color signal, and (b) among the plurality of spectral images, the R, G A spectral image having a spectral wavelength within the wavelength band of the second primary color signal, which is one of the two primary color signals excluding the first primary color signal, and the spectral wavelength is the first wavelength A spectral image having the highest evaluation is selected as a second spectral image from spectral images separated by 30 nm or more from the spectral wavelength of the spectral image, and (c) among the plurality of spectral images, the R, G, and B Of these, other than the first and second primary color signals A spectral image having a spectral wavelength within the wavelength band of the third primary color signal, which is a signal, and the spectral wavelength of which is 30 nm or more away from the spectral wavelength of the first and second spectral images. Image selection means for selecting the spectral image having the highest evaluation as the third spectral image;
  Image generating means for generating the processed image by assigning the first, second and third spectral images selected by the image selecting means to primary color signals in a wavelength band to which the spectral wavelength of the spectral image belongs, respectively;
  A spectral image processing apparatus comprising:

Images