TWI377044B - Method for reconstructing color images - Google Patents

Description

1377044 VIII. Description of the Invention: [Technical Field to Which the Invention Is Applicable] The present invention relates to a spectrum of a light source that is changed such that a spectrum of a light source changes. - Bhutan's current method, especially a color image in which the pixels of the image are reproduced by the original digital color image according to the transformation, and the visual effect is reproduced. [Prior Art] The silk lamp's xenon lamp and halogen lamp are Use various lighting sources, environmental protection, small size, light intensity

Generally, the source of illumination in medicine is tungsten, and the light-emitting diode (LED) has many advantages such as high luminous efficiency, long life, adjustable and rich color gamut compared with the previous one. -pi Tuding η II β, where the 夂 equipment uses a variety of different ray spectrum light sources to produce different time 1 , Wang Wu shirt like 'The creature shadow will be available to the medical staff to go to the six secrets and one king The visual image m S Α is used as a reference to determine whether an abnormal lesion has occurred in the area to be diagnosed.丨 4 sentences lL m ... "分分" 丨 赞 的 的 的 如 如 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各 各', t is more powerful for children under the age of five, if you can go in the early stages of the enterovirus, go shopping. The treatment is to reduce the chance of serious illnesses. 'General pediatrician's identification of cockroaches, & +,, (4) Ways to understand the disease - for the throat and hand, oral examination, due to string. Gas from 〇.〇 ^ ^ , ... for young children, so in the doctor's consultation, the general doctor will ask for a short time to ask if there is a hole in the membrane. If you can set a special light source, you can increase the lesion. The area and the weekly surgical light source can be used to identify the monthly force. If the 4 1377044 can be combined, the doctor can carefully interpret or apply image recognition on the computer screen, and quickly identify whether there is any problem. Enterovirus, which allows health care workers to determine if a child is suffering from a disease. When using different spectral sources to illuminate the area to be diagnosed, there will be different effects of the image, however, if a variety of lighting instruments are really prepared to provide different spectral sources, Will lead to a number of problems, such as expensive hardware purchase costs, maintenance, requiring adequate space, unskilled personnel, etc., so that different spectrum sources are illuminated The detection method will be quite difficult to solve. [Invention] In order to solve the problem that the actual multi-spectral light source illumination device is not easy to obtain, the main object of the present invention is to provide a color image reproduction method, which can make an original by changing the spectrum of the light source. Each pixel of the digital color image is changed according to the spectrum of the converted light source', thereby simulating a color image that produces spectral effects of different light sources as a basis for detecting and determining. To achieve the foregoing objective, the method of the present invention comprises: "a spectrum capturing device" Obtaining the spectral data of the plurality of sample patches, and obtaining the basis function coefficients of the sample patches; and the digital camera capturing the image values of the plurality of sample patches; the vehicle is based on the basis function coefficients and the image capture The value operation finds a conversion moment by which the digital camera captures an image capture value of an original image, and the image value of the original image and the conversion moment are reproduced by the king, according to a new illumination source. Transforming the transformation matrix, and calculating according to the original image"◦44 image capture value and the transformed transformation matrix Students one pair of images to be reproduced new source of illumination. x'

Wherein, when the conversion matrix is transformed according to a new illumination source, the system + comprises: V: the spectrum operation device takes the spectrum of the plurality of sample patches to obtain a set of original spectra; and, the digital sample camera uses the digital sample block Take a frequency spectrum of a set of approximate spectrum; receive eight, the original spectrum and the approximate spectrum of the group are divided by a primary source frequency error, - do not pay - the original reflection spectrum and a set of approximate reflection spectrum; day 3 $ Λ 'And the original reflection spectrum and the set of approximate reflection spectrum are multiplied by the spectrum of the new illumination source, w4 ^ this new material; the new spectrum of the new sample color block is generated. i according to. The new spectral data of the complex sample color block is calculated and found in the nose; when the ώ ώ analysis calculates the basis function, the main axis is based on the base and the total matrix, and is multiplied:: The image interpolation value is calculated and the conversion moment is calculated by the linear regression analysis method. In the digital camera, the digital camera first passes the color correction and the wide image captures the image capture value. The digital camera is used to capture the digital camera. ~ The value is taken when the image is taken from its image. / / The film captures the image of the original image. By the foregoing manner, the present invention can change the spectrum of different light sources, 6 Γ 377044, so that the captured image can produce different effects as the spectrum of different light sources is changed. Color image, so you can design the desired color reproduction effect, such as improving the brightness, sound, A* & color 胤 color difference of the color image to obtain the desired source spectrum, thus reducing the acquisition of the real light source. Cost, when the invention is applied to biomedical or semiconductor testing, by replacing the spectrum of different light sources, the human eye or instrument can directly observe the tissue lesion or the product. [Embodiment] The present invention mainly utilizes a multi-spectrum based method (Μ_...plusbaSeC0 to achieve high-accuracy color measurement and color reproduction technology, and utilizes a digital camera that is readily available as a multi-spectral image county. The sampling device in the system' Μ directly takes the value sensed by the digital camera as image manipulation = _al C〇_s, Dc), and also uses several color stencils in the process (c〇i〇 "Fme "> to manipulate the image and increase the approximation of its approximate spectrum. Therefore, the data value of each pixel in the color image captured by the digital camera can be directly converted into the light source spectrum 1 to reconstruct the color image. The spectral value of each pixel changes with the spectrum of the light source. The spectrum color reproduction technique, the silk analysis method and the multiple linear regression analysis method are first introduced. A·Multi-spectral color reproduction technology color reproduction refers to the color matching principle. Make a copy of the object or the original image. The color is divided into spectrum color and strip: You', IF W color matching requires the original image and the spectral characteristics of the reproduction % - and the condition Equivalent color matching must be in the pursuit of color appearance - "If the specific conditions change, then not - can be 7 1377044 • Guarantee to maintain the same color, such as color printing, photography, television, etc. are all color matching conditions, Unless it is in a special situation, its spectral color matching is difficult to achieve. The method of multi-spectral acquisition was first proposed in the VASARI system of the National Gallery, UK in 1993, and the VASARI system was used in one. The scanner is combined with seven separate filters to create a seven-channel sampling device' to capture images of the artwork, and the sensor on the scanner is a monochrome and high-resolution digital camera. After the image is acquired, the captured image is processed and recorded by some appropriate signals. Therefore, the next step is to illustrate the color reproduction of the spectrum reproduction method: it can acquire the spectrum of each pixel. Color reproduction. Because National : Gallery successfully digitalized the art images in its library, this method is subject to some research The importance of 'the most famous is Munse丨I Color Science Labtoratory (MCSL). Because: VASARI system needs to spend a lot of time on the scanned image, so MCSL proposed to use monochrome cCD (Charge-Coupled Device) digital bit · The sample device is also equipped with seven independent filters to improve the sampling device in the VASAR system. Under the same signal processing, the MCSL experiment proves that such a method can also achieve good results and reduce the cost of sampling. . Time. In addition, 'MCSL does not think it is necessary to use a monochrome digital sampling device. If it is a two-primary digital sampling device, it can achieve good results. The multi-spectral acquisition system is mainly divided into two parts, the first one is image acquisition and the other is multi-spectral analysis and spectrum reproduction. As shown in Annex 1, a schematic diagram of the multi-spectral image capture process is performed by using a digital camera to capture an image of a color filter with a digital camera and then analyzing the image by a computer. Annex 2 shows the process of analyzing and reproducing the spectrum of multi-spectral images. μ is to sample the color patches by m spectrometer, and then perform spectrum analysis and then integrate with the camera to achieve the reproduction of color images. B. Spindle analysis method The main sleeve analysis method is a commonly used method for multivariate statistics. Since 196 = application is important in color technology - the main purpose of the spindle analysis: 'For the first one, the first one is to define a large amount of spectrum information. The main axis of the project is to simplify the data of the information. The first ten 笪 M ^ — to reorganize the original tribute, the relevant and independent variables, and then by the component 'final (4) to explain the original data = Mainly as shown in Annex III, hypothesis (X1 77 variability. ; can be transformed into a central view of the mutual axis analysis method by linear transformation, variable and coordinate two independent variables, but two related variables The variability of the material. This is the main spindle analysis mode derivation: a2 Suppose there are m variables (χ1, χ2 .·, am) such that:: ·.,Xm) 'and want to find the coefficient (a1, var (arx1+a, the value of x2+..+Vx^ is the largest (Var· represents 6 deniers (2-10) takes the largest variable in the original data group. The side-extinguishes in it), in this way, the following Normalization conditions: /, medium (a" ..... am) must be satisfied (ai2+a22+...+am2)=:- i (2-11) 9 U/7044 a 11 condition is to make 2, the formula is very large - group (ai a2',) 'is the unit space vector of the towel space' not only that (ai and 2 in ' Represents the concept of the direction vector, that is, the direction of the main axis. The m-dimensional analysis of the spindle can be expressed by the following equation: PCl=ailXl+ai2X2+-+a1mxm PC2=a21x1+a22x2+...+a2mXm PC C in the basin ~ m (2-12) 八中PC^PC』PCm stands for and m main sleeve. Its first-spindle coefficient (ai1 a spindle is m-type, and can be uar(ai)Xi+a 11χ+12,..., ~) Must be full M. , ^ _ 12* X2 Xm) The value is (4) Mutual ": Spindle to ¥ m spindle and so on, and each coefficient to each coefficient (aa "like m find PUC^PCm Ba... The material is written as a covariation matrix of nXn (c〇va "iance m_X", the eigenvector of this covariation matrix represents (u%), and its corresponding eigenvalue represents the variable eigenvalue in 2_1〇 .|2', the spindle system selected by the user ^, the number (ail, ai2aim) can make the first-spindle % 八八 Γ 异 ' ' 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说 也就是说Ability to source material The mutated part that is not explained by the first - has the largest interpretation of the month." Usually the original data has · m can still find m main components. But the month & this finds p spindles, passed God pays p<m and P is as small as possible, but it can be used in the beginning of the data. More than % variability, this is the main purpose of the present invention to analyze data using a spindle knife analysis method. 1377044 It is known from the foregoing that when a group of data sets is a matrix of non-ancient land p is a single matrix, and a group of random vectors X, such as 2-13, and not, and the covariance matrix C (covariance matrix) can be assumed X= (^, x2> , χη)τ _ _ (2-13) C=E[(XX)(XX)T] (2-14) Assume that the covariation matrix is a Ρ dimension The space 'then will exist' - a group base base system, and assume that the η group base moment _ _ _ π r Ping Ma U, expressed by the following formula: U = [x1t x2, ..., χ ] ( 2-15) The value range to which the variable X is mapped in the dimensional space can be expressed by the following equation: (2-16)

y=UTX Therefore, the covariation matrix C of y can be calculated: C· =E[(y-y>( y-y)T] (2-17)

=UTE[(XX)(XX)T]U =U^GU The new covariance matrix c can be represented by the characteristic equation (Αλ = λ X, where Α is: any square matrix, also the eigenvalue, χ is characteristic Vector), the pattern is represented by the formula 2-18, and the relationship between the spindle bases is orthogonal to each other (un 丨 vector) ' can be expressed as 2-19. COW,...,C-xp] -[λ ·|Χΐ,λ 2 X2,. _ · I into pXp] 11 X\ Γ377044

=[in 1 , =AU λ λρ]

Xp UTU=I Finally combines 2-18 and 2_19, and the equation is obtained. The equation is as follows. C'=UTCU=A". , - Li C, multiple linear regression analysis (2-18) (2-19) its eigenvalues; I i can be characterized by this feature vector (2-20) 'linear riding analysis is a kind The main purpose of the "law", which is widely used in statistical analysis, is to understand whether the "objective variable" can be expressed by a linear equation of some self-variables, and use it to explain the purpose variable: special ", ", that is, variable Whether the relationship between X and γ is close. Finally, the γ value can be obtained from the value of the variable X through the multiple linear regression equation. In the present invention, according to the image/value Dc (digita|C()unts) value of an image taken by the digital camera, the frequency value of the image is obtained through operation, and then the system is characterized and the Dc is understood. After the relationship between the value and the spectral value, a transformation matrix is obtained by multiple linear regression. Xi Xi Therefore, before the multiple linear regression analysis, it is necessary to understand the relationship between the Dc value of the digital camera and the spectral value of the spectroscopic spectrometer. The present invention firstly investigates the spectral value of the spectroscopic spectrometer, assuming that the spectral data of the 丨 group is taken, and the reflection spectrum r( Λ ) of each set of reflected spectral functions after the spindle analysis is reduced can be expressed in the following 12 states: "1(1)=a"Xl(1) ~2u)+...+aiXU) "2(1)=secret (1)% dance·two

n(A)=aMxl(A)+ai2X2U)+ +ajnXnU) W: Medium: For the principal axis analysis, the mapping "the largest major coefficient of the group, and the interval 2·21 formula" is expressed by the following equation: Time

=aX where (2-22) (2-23) (2-24) (2-25) Γ=Ι Γ1. r2.....η]τ a=[ai, a2(..., an] =a.ln

X=lW-,xn]T On the other hand, after the spindle analysis of the spectroscopic spectrometer is performed, the Dg value can be expressed in the heart by calculating the image captured by the digital camera. The Dc value of the 3 outer 'camera cameras captured by k color filters is indicated by 2-27. (2-26) (2-27) 〇〇=|Ρ(λ>(λ)5(λ)^λ

Dc k = iP(X)r(X)Fk(X)S(X)dk where S(A) is the digital camera sensitivity distribution matrix, ρ(λ) is the distribution matrix of the source spectrum, and r( λ ) is the object Reflective spectral distribution matrix, Fk (also) is the spectral distribution matrix of the color crossings. In addition, in order to simplify the 2_27 type, it is assumed that D(A) is a 2-28 type, and a 2-29 type can be obtained by substituting the 2-27 type. D( λ ) = iP(X)Fm(X)S(X)dk (2-28)

Dck=r(A)DU)

=rD (2-29) 13 1377044 The matrix constant 22 is substituted into the 2-29 type to obtain the 2-30 type. The X_D can be obtained by arranging the XD into the number of the car cranes mk ’.

DCk=aXD (2-30) sense a mkDCk (2-31) The analysis of the most master axis of the group shows that 3 ^ in the spindle analysis, the n before the mapping is the second most important, the time coefficient ' and Dc is the digital camera The numerical value, therefore, knows that there is a conversion matrix mk between the spectroscopic spectrometer and the digital camera, so it is calculated by calculation. (4) Moment material, 〇 heavy linear regression month reference - Figure 'In the present invention, the following is mainly included in the spectrum acquisition device for the complex sample color (4) to take its spectrum data ·, transport " find the base of the sample color block The function coefficient (1〇1>; (1〇2) to; ϋ the camera fetches the image sample value of the complex sample color 4 according to the basis function coefficient and the image capture value operation to find a conversion matrix (103); Taking the image capture value of the original image by the digital camera, generating an image according to the image capture value of the original image and the conversion matrix operation (104); converting the conversion matrix according to the new illumination source, according to the original The image reading value of the image and the transformed conversion matrix can be calculated to generate a pair of reconstructed images of the new illumination source (1 05). Referring to the attachment shown in Figure 4, first, a spectrum is used, such as spectroscopic light. MiMOLTA CS1000A measures the spectral data of the sample color ridge. In this embodiment, the U 24 color sample color block and the measurement of each data are separated by ~ 14 U77044, so there are 201 spectrum samples, single color block measurement. Spectral samples to vector and matrix Represented by the following formula: (3-1)

Sr(A)=[Sr(A1)ISr(A2)lSr(A3).......,Sr(Aj] where m = 201. In addition, the 24-color sample patch can be measured by the following formula:

Sr. (M,Sr,^2),Sn(X3), Sr2 (h),Sr2(X2),Sr2(h),Sr^A) =A = ,Sr] (λ plus) .,Sl*2 (in 201) (3-2), 8:24 (in 201)

SnA丨)χλ2)χλ3), where i = 1~24, after obtaining the 24 sets of spectral data, then using the main wheel analysis method for analysis, after the spindle analysis, a set of spindle components can be obtained, such as 3-3 As shown, these components can be used as a basis function to linearly combine each

~ Pen spectrum data, as shown in the equation 3-4, which represents the coefficient corresponding to each basis function. (3-3) (3-4) δη( λ)=α Μ λ,)+α 2χ2( λ 2)+····+ α 201χ201( λ 201) α But since Sr>( λ ) is not a square matrix Therefore, you must use the virtual inverse matrix to do the operation. Firstly, we first take ΑτΑ as the covariance matrix (G〇varjance^trix), and solve the eigenvectors with the eigenfunction (BXm=5mXm), where dm is the eigenvalue and Xm is the eigenvector, which can be expressed by the following formula: Let B = ATA , then (3-5) ' m=1 ～ 201 15 ^77044 and over the 3-5 formula, you can get 2 〇 1 eigenvalue (7) 咱 buckle ... 丨 leaf) corresponding to the 201 sets of eigenvectors (ejgenvect〇 |>), so the main purpose of the spindle analysis is to find a few sets of the largest eigenvectors to form the principal component of the spectral data, and approximate the original spectrum with a few components. After the eigenfunction data is available, the spindle analysis can be performed. Here, the first few eigenvalues are selected to select the spindle component (feature vector), so the cumulative growth ratio is calculated, which represents the first eigenvalue, the equation Such as 3_6

The table is not. And according to Equation 3-6, the comparison between the number of feature vectors and the cumulative growth contribution ratio is calculated, as shown in Table 3 - 彳. ^

(3-6) Cumulative growth ratio of feature vector number (%) 1 84.87 2 94.59 3 99.10 4 99.61 _ 5 99.84 --6 99.90 --_____1 99.94 —__8 99.97 ---9 99.983 ____ 10 99.99 Table 3-1 3-1 knows that when the number of feature vectors is above 6.:., the cumulative growth rate can reach 99_9% or more. When the number of feature vectors reaches 6 or 6 groups, the original spectrum can be expressed. Use the first six sets of features to inward as the basis function of the spectrum. It is known from the previous cumulative growth contribution ratio that the eigenvectors of the first six groups are used to reconstruct the basis function of the spectrum, and the basis vector 1377044 number vector is expressed by the following equation: :η (λΐ) :η (λϊ) for η=1 ～6 (3-7) [η (λ:5) η (λ20ΐ) Therefore, after obtaining the first six sets of maximum eigenvectors as basis functions, they can be linearly combined into the original spectrum data of each data, taking the first set of data as an example. , can be expressed as 3 - 8 where a (i) is the coefficient preceding the unknown linear combination.

S^C λ )= 'Sr, (λ.)" Srz (λ2) =CL ! Xi (λι) X 1 (λ2) + a 2 X 2(λΐ) X 2 (λ2) + . · _ + CK i X 6 (λΐ) X 6 (λ2) Sr201 (λ201) X 1 (into 201) X 2 (λ20ΐ) X 6 (λ20ΐ) (3-8) Then, taking 24 groups as an example, sorting into 3-9, Can be simplified to 3-10.

8ΐΐ(λΐ), 8Γΐ(λ2), Sri(A,3).......,Sri(X201)

Sr2(Xi), 8Γ2(λ2), Sn(X,).......,Sr2(X2〇i) 8η( λ)= . . . 8Γ24(λΐ), Sr24(X2) , Sf24(X3 ).......,8犷24(into 2.1) — '· 17 1377044 ai(1), ai(2),··· 5 a old · Χι(λι),Χι(λζ),... , (in 2〇1) 〇C2(l), (X2(2),···, a2(6) • · • Χ2(λΐ), Χ2(λ2), _ · ·, χ2 (in 2〇1) _a24( i), a24(2),...ja24(6) _Χ6(λ|), Χ6(λ2), · · ·, along (human brew) (3-9)

Sri(A)=a(i)nx(Xn)Ti f〇rn=1-6, i=l~24 The dimension of the original eigenvector is 201 ' However, after the spindle analysis, the eigenvector is reduced to 6 groups' due to Xn It is not a square matrix in calculation, so the corresponding coefficient a is obtained through the virtual inverse matrix, represented by the formula 3-11, and finally the coefficients of the 24 color patches can be arranged, where a jXi 〇 -4, 丨 = 卜 6.

α (ί)η_δπ( λ )x[(Xn)T]1 1J When the coefficient α of the first six sets of eigenvectors is obtained by the principal axis analysis, the Dc value captured by the digital camera is obtained by multiple linear regression analysis. The transformation matrix between systems.

(3-10) The digital camera Dc value can be obtained according to the above description. Therefore, the Dc value of the digital camera can be combined with the 3-10 type to be sorted into the 3_i2 type. ^n(i)=/77n(k) XDCk(i) (3-12) where n = 1~6, ι=1~24, k=1~3. And it can be expressed as a matrix type of 3-1 3, so the conversion matrix m can be calculated. OU(l), OC2(l), ···, CX24(1) ai(2), OC2(2) , ... , 0124(2) • · • · · a 1(6) , a2(6 ) , ... , a24(6) , mm, .,,m3〇)~ mip), mip),..., ni3p) jni(6), πΐ2(6),..., mi(6)

Dci(l),DC2(1),. . ·,〇C24(l) Dci(2) , DC2(2) , . . . } DC24(2) •DCl(3),DC2(3),. .DC24(3) 1377044 3-13) According to the conversion matrix m calculated above, the spectrum can be reconstructed. Description of actual measurement examples: The present invention uses a CANON 860is digital camera as a sampling device, and is divided into two parts during measurement, one without passing through a color filter, and the other through six color filter groups, selected The six color filters are auxiliary for red (R), green (G), blue (B) 'blue (C), purple (M), and yellow (Y) light transmittances of 30%.

The Dc values obtained by shooting the 24-color sample patch with the digital camera are shown in Table 2 below: Patch RGB 1 104.08 66.411 54.813 2 201.83 155.7 142.88 3 114.84 139.35 182.09 4 95.15 112.88 64.94 5 155.88 153.47 202.82 6 133.79 195.34 193.77 7 190.99 114.91 37.85 8 80.117 109.12 185.45 9 207.67 98.618 112.48 10 100.82 64.912 117.44 11 168.33 198.68 78.457 12 205.31 164.18 53.654 13 38.078 63.097 149.52 14 79.889 150.62 83.039 15 193.97 58.711 73.967 16 219.7 200.96 72.81 17 210.81 104.7 171.05 18 73.928 150.3 199.81 19 227.59 225.94 228.03 20 207 205.45 208.7 21 178.78 175.71 180.64 22 141.1 140.57 145.04 19 1377044 23 88.684 89.358 93.089 24 37.341 38.522 38.997

Table 2 shows the six sets of basis function coefficients of the sample color block by spectroscopic spectrometer, as shown in the following three lists: Patch α, α, α4 cts a, 1 -0.6723 0.1118 0.5625 1.6133 -4.8543 18.270 2 -1.7122 3.2687 0.0853 4.7085 -11.328 67.173 3 -0.7310 -0.6049 1.4018 5.9270 20.077 40.559 4 0.6744 -2.1668 -1.7588 -6.3965 0.1159 20.868 5 -0.0247 -2.9428 1.7070 13.964 18.983 57.203 6 -0.9590 0.3607 -6.9596 -13.680 26.902 67.953 7 -2.2466 3.2048 7.5213 1.9661 -30.202 51.268 8 -1.4700 -4.4197 3.8500 13.346 22.860 34.135 9 -1.4106 4.2025 -0.4818 18.274 -24.391 47.891 10 ' -4.6785 -7.7265 -1.3346 11.473 0.3157 21.199 11 -0.9082 -3.9695 -6.7682 -27.958 -8.4103 61.083 12 -1.3483 -0.8906 9.6877 -10.747 -32.743 66.842 13 1.4111 -1.7274 4.5895 10.296 17.891 19.922 14 -1.2473 0.1020 -6.0084 -15.582 5.3700 28.320 15 1.2031 0.6855 -11.3066 24.043 -33.293 40.404 16 1.1547 -1.9846 -1.3881 -22.084 -37.591 89.291 17 1.5127 -3.0763 -4.3986 32.295 -16.284 60.110 18 - 3.1543 7.3086 -4.7347 3.6128 30.537 41 .791 19 •2,1356 1.6207 2.3801 -2.1104 20.022 151.79 20 1.6075 0.4445 2.2658 -0.7335 16.250 102.02 21 1.0377 0.2140 1.8616 -0.3239 10.260 63.040 22 0.6697 0.0252 1.1264 -0.3098 6.6471 36.000 23 0.3190 -0.0989 0.5747 0.0034 3.3792 16.984 24 0.0615 -0.0542 0.1616 0.0778 1.0591 5.9856 Table 3 Therefore, according to the above formula 3-13, the conversion matrix of % can be calculated as shown in the following Table 4: 〇.〇〇〇〇〇〇398 -0.000000428 -0.000000105 -0.000001141 0.000001919 -0.000000752 - 0.000001256 0.000002503 -0.000001204 0.000061688 -0.000062193 〇.〇〇〇〇〇〇114 -0.000027045 0.000024268 0.000002589 20 1377044 0.000022440 0.000019460 -0.000001860 Table 4 In order to further enable the digital camera to provide better shooting results, a color correction step can be performed on the camera. . The standard of color correction is to take the Dc value of the camera as the RGB value of the 24-color sample color block of the spectroscopic spectrometer (CS1000A). The correction method is mainly to measure the six groups of the lowermost row of the sample color block for the camera and the spectroscopic spectrometer. Gray-scale color blocks, respectively, get their RGB Gamma curve, and take the camera as the abscissa, the spectrophotometer as the ordinate, you can get a fitting equation, and finally measure the value of the 24-color block to the splitting The photometer can be used to obtain a new set of camera data. After the gamma curve of the spectrometer and the camera, the RG B color correction of the camera and the spectroscopic spectrometer can be performed. The Dc value after the camera calibration is shown in the following table 5. Its new conversion matrix is shown in Listing 6.

Patch RGB 1 57 40 19 2 190 121 102 3 66 100 157 4 52 66 28 5 118 119 193 6 90 183 175 7 170 68 0 8 45 62 162 9 204 55 67 10 55 39 72 11 125 • 189 39 12 198 134 18 13 0 39 109 14 45 115 44 15 175 36 36 16 229 194 35 21 1377044 17 210 59 138 18 43 114 187 19 252 254 255 20 202 204 205 21 151 153 154 22 100 102 104 23 48 50 52 24 0 0 0 Table 5 0.0000008 〇.〇〇〇〇〇〇〇-0.0000007 0.0000055 -0.0000055 〇.〇〇〇〇〇〇〇0.0000091 -0.0000096 〇.〇〇〇〇〇〇2 0.0001389 -0.0001640 0.0000266 -0.0001768 0.0000847 0.0001032 0.0001338 0.0000343 0.0000089

Table 6 is shown in Appendix V. It shows the color difference comparison table between the two samples before the color correction and the spectroscopic spectrometer. The color difference table is calculated by the Lab color difference formula. know. Conversely, as shown in Annex VI, the color-corrected color difference comparison table can be found from these two tables. The corrected average color difference is smaller than that before the correction, and it should be feasible to display the digital camera as the shooting device. Please refer to Appendix VII. If the digital camera with color correction is further combined with a color filter, the average color difference can be found to be lower, at 5.9797. One of the practical applications of the present invention is to capture a biological image to be judged by a universal three-primary digital camera, and to enhance the color difference and contrast between the normal region and the lesion region of the biological image by changing the light source. Firstly, the spectrum of the 24-color sample color block 22 1377044 can be obtained by the spectroscopic spectrometer and the digital camera. The original spectrum and the -group approximate spectrum can be obtained, and the two sets of frequencies can be divided by the spectrum of the original source in the experimental environment at that time. The broken shot (sample patch) @ a set of raw reflection spectra and a set of approximate reflection spectra. Then, after the original reflection frequency of the group and the set of approximate reflection spectrum are multiplied by the spectrum of the new light source to be replaced, the new 24-color block spectrum of the group can be obtained, and the new conversion after the replacement of the light source can be obtained. matrix. In other words, as long as the spectrum of the different sources is transformed, a new conversion matrix is created, which in turn produces a new image corresponding to the new source.

The attached part shows the fluorescent lamp. When the light source illuminates the hole in the oral cavity, the position of interest is taken. The resolution of the image taken at the couch is 470x672. As shown in Annex IX, the replacement is based on Annex VIII. The oral image of the light source is shown in Annex IX as the oral reconstruction image using the blue light white LED source spectrum plus different color filters. The figures are: (A): The light source is a yellow light illumination device because it is a broadband light source. Therefore, the color rendering is higher, so that the lesion range and the normal lesion-free area are not easy to have too much chromatic aberration. .,... (B) :- The light source is a bluish white LED. The signal is stronger in the blue band, and the color rendering is worse than the broadband source. Therefore, the color difference will be better than the figure (A). Significantly. (C): The light source is bluish white light |_ED with red (R) filter, the oral image is more red, so the original oral image will be red plus: the upper two colors filter. The color will be more Saturated, so the lesion is less likely to be observed by the naked eye. (D): The light source is a bluish white LED with a green (G) light beam. It can be clearly seen that there is a dark red color around the lesion, because the lesion 23 2377044 has a larger suction for this band. Therefore, the color performance will be dull. ...(E)' (F): The white light LED with a blue light source is matched with a blue (Β) filter and a blue light white LED with a yellow (γ) filter, which can be seen when passing through the blue filter. The normal range of the surrounding area will have a whiter effect. Because the background oral tissue is a longer red band, the wavelength difference when adding a blue filter will be larger than that of the yellow filter. This effect will be visually produced. Zero [Simplified description of the drawings] First figure: A flow chart of the present invention. Annex 1: Schematic diagram of the acquisition process of multi-spectral images. - Annex II: The process of analyzing and reproducing the spectrum of multi-spectral images. : Annex 3: Schematic diagram of the main shaft. Annex IV: Schematic diagram of the 24' color sample color block. Annex V · Comparison table of color difference between digital camera and spectroscopic spectrometer. Annex 6: Comparison of the color and difference of the digital camera after the color correction and the spectroscopic spectrometer. .... ' Attachment 7: The color difference between the digital camera and the color corrected by the camera. Attachment 8: Use the led flashlight to capture the original image of the cavity at the hole. Attachment 9: Replace the yellow light flashlight with the bluish white light tED with different color image of the color light film. [Main component symbol description] 24

Claims (1)

丄j/7044 Q?·1] VII. Patent application scope: 1 · The sap image reproduction method 'contains the basis function coefficient of the sample color block by using a spectrum display device for the complex sample color calculation; Obtaining the image operation value of the complex sample color block; arranging; calculating according to the basis function coefficient and the image manipulation value--converting moment U-head machine--the original image is taken (4), the value begins Image shadow_value and the conversion (4) operation generation-reproduction = transforming the conversion matrix according to a new illumination source, according to the (4) brown value of the original image and the transformed transformation matrix, the reproduction of the illumination source can be calculated. When you change the matrix to the spoon: 1 According to the new illumination source, the conversion method includes the following steps: The spectrum acquisition device extracts the spectrum data of the plurality of sample patches to obtain a group The original spectrum; taking the spectral data of the digital sample block of the digital camera to obtain a set of approximate spectra; dividing the original spectrum and the approximate spectrum by a spectrum of the original source simultaneously, respectively obtaining a set An original reflection spectrum and a set of approximate reflection spectra; the set of original reflection spectra and the set of approximate reflection spectra are then multiplied by the spectrum of the new illumination source to generate new spectral data of the plurality of sample patches; according to the plurality of sample colors The new spectral data of the block is computed to find a new transformation matrix. 25 1377044 2·If the scope of patent application is item j; the method for calculating the basis function coefficient of the sample color block is calculated by operation. ', Axis analysis method 3. The color image reproduction method according to item 1 of the patent application scope is based on the basis function coefficient and the shadow (4) value calculation = method 'Dong root is calculated by multiple linear regression analysis. _When changing matrix, the color image phase #4·, V image reproduction method according to the scope of the patent application, the image capture value of the original image is firstly corrected by the machine. . ^, the digital camera 5 · as claimed in the scope of patents! In the color camera pair, the original image is imaged, and the digital image is captured by the color camera. The digital camera captures the original image and captures the image capture value. Eight patterns: (such as the next page) 26