The fractional calculus filter of the digital picture of high precision computation proposed by the invention is a kind ofly digital picture complex texture minutia to be carried out the high accuracy fractional order strengthens or level and smooth circuit arrangement.The order of the fractional calculus that the present invention relates to is not traditional integer rank, but non-integral order is generally got mark or reasonable decimal in the practical applications.See Fig. 1, the fractional calculus filter of the digital picture of this high precision computation is to adopt RGB to constitute with cascade system to RGB transducer 14 with HSI to HSI transducer 9, line storage group 10, lock phase/shift circuit group 11, fractional calculus mask convolution circuit 12, maximum comparator 13.The first algorithm unit circuit, 1 to the 8th algorithm unit circuit 8 in the fractional calculus mask convolution circuit 12 adopts distinctive fractional calculus mask convolution algorithm to realize the high precision computation of fractional calculus.The fractional calculus filter of the digital picture of this high precision computation have can fractional order strengthen again can fractional order level and smooth, computational accuracy is high, the distortionless remarkable advantage of color.The fractional calculus filter of the digital picture of high precision computation proposed by the invention is specially adapted to that the complex texture minutia of HD digital TV, biomedical image, bank money, satellite remote sensing images and biometric image etc. is carried out the high accuracy fractional order to be strengthened or level and smooth application scenario.The invention belongs to the technical field of applied mathematics, Digital Image Processing and digital circuit cross discipline.
Summary of the invention
The objective of the invention is to construct a kind of fractional calculus filter of digital picture of high precision computation, this filter can carry out that the high accuracy fractional order strengthens or level and smooth to digital picture complex texture minutia.Applicant of the present invention has furtherd investigate basic principle and the operation rule thereof with fractional calculus enhancing or level and smooth digital image texture minutia; On this basis according to the input characteristics of the character of digital image fractional order calculus and Digital Image Processing, digital circuit, serial digital video code stream; New departure of the circuit arrangement of a kind of high accuracy fractional order enhancing or level and smooth digital picture complex texture minutia has been proposed, i.e. the fractional calculus filter of the digital picture of high precision computation.The fractional calculus filter of the digital picture of this high precision computation have can fractional order strengthen again can fractional order level and smooth, computational accuracy is high, the distortionless remarkable advantage of color.See Fig. 1, the fractional calculus filter of the digital picture of this high precision computation is to adopt RGB to constitute with cascade system to RGB transducer 14 with HSI to HSI transducer 9, line storage group 10, lock phase/shift circuit group 11, fractional calculus mask convolution circuit 12, maximum comparator 13.The first algorithm unit circuit, 1 to the 8th algorithm unit circuit 8 in the fractional calculus mask convolution circuit 12 adopts distinctive fractional calculus mask convolution algorithm to realize the high precision computation of fractional calculus.
Before specifying content of the present invention; Be necessary used symbol connotation of this specification and span thereof are carried out 3 explanations: the 1st point; Continue to use the longitudinal axis and the transverse axis coordinate (different with the general mathematical notation of Euclidean space, its custom is represented transverse axis and ordinate of orthogonal axes respectively with x and y coordinate) be accustomed in the traditional images processing with x and y coordinate difference presentation video pixel, with S (x; Y) denotation coordination (x, the gray value or the rgb value of the pixel on y); When x and y get the continuous analogue value, S (x, y) expression analog image; When x and y got discrete digital value, (it was a picture element matrix to S for x, y) expression digital picture (x and y represent row-coordinate and row coordinate respectively); The 2nd point; In order to make fractional calculus mask (it is the square formation of (n+2) * (n+2)) that clear and definite axial symmetry center arranged; The minimum value of n is 3, the maximum occurrences of n+2 less than the size number N of the digital picture of pending fractional calculus (if (x y) is the picture element matrix of L * H to the digital picture S of pending fractional calculus; When L=H, its size number N=L; When L ≠ H, its size number N is the minimum value among L and the H, i.e. N=min (L, H)), n get any odd number between [3, min (L, H)-2], and (L H) gets minimum value among L and the H to min; The 3rd point, in practical engineering application, the digital picture S (x of pending processing; Y) (it is the picture element matrix of a L * H, and L representes S (x, line number y); H representes that (promptly every row has H pixel to S for x, columns y); X gets the integer between 0~(L-1), and y gets the integer between 0~(H-1)) gray value or the rgb value of the capable pixel of L generally be not parallel input (input simultaneously of each row of the gray value of the capable pixel of L or rgb value), but the serial input (gray value of the capable pixel of L or rgb value one-row pixels connect the one-row pixels input; The gray value or the rgb value of H pixel of every row input form the serial digital video code stream) image processing apparatus; According to the input characteristics of serial digital video code stream, use S
x(k) (subscript x representes each frame of digital image S (x to the pixel in the expression serial digital video code stream; Y) be to form the serial digital video code stream with the mode that one-row pixels connects one-row pixels input; (x is y) from its nethermost delegation (L is capable) beginning input from bottom to up, k remarked pixel S for S
x(k) the pixel sequence number in the serial digital video code stream, k begins counting from L * H-1, subtracts one by pixel input k value, until being zero); If S
x(k) coordinate before the corresponding serial input (x, the pixel S on y) (x, y), S then
x(coordinate before the corresponding serial input of k ± aH ± b) (the pixel S on the x ± a, y ± b) (x ± a, y ± b).
See Fig. 1, the fractional calculus filter of the digital picture of high precision computation of the present invention is to be formed to 14 cascades of RGB transducer with HSI to HSI transducer 9,
line storage group 10, lock phase/
shift circuit group 11, fractional calculus
mask convolution circuit 12,
maximum comparator 13 by RGB; Serial digital video code stream S
x(k) behind the fractional calculus filter of the digital picture of input high precision computation; Process RGB is divided into three the tunnel with gray value or brightness I component after HSI transducer 9 is handled: after the first via is handled through
line storage group 10, lock phase/
shift circuit group 11, fractional calculus
mask convolution circuit 12 in proper order, and difference output pixel S
xThe approximation of (k+n (H+1)) inclined to one side calculus of v rank fractional order on diagonal, upper right diagonal, upper left diagonal and 8 directions of diagonal line under x axle negative direction, x axle positive direction, y axle negative direction, y axle positive direction, the left side; After handling through
maximum comparator 13, the value of exporting the mould value maximum in above-mentioned 8 approximations is as pixel S again
xThe v rank fractional calculus value approximation of (k+n (H+1))
The second the tunnel triggers sequential control circuit produces corresponding timing control signal; The output of Third Road and
line storage group 10 together the feed-in lock mutually/
shift circuit group 11 generates the pel array of (2n+1) * (2n+1).Wherein, the order v of the fractional calculus
mask convolution circuit 12 in the fractional calculus filter of the digital picture of this high precision computation can-m~+ (m is any positive integer to get mark or reasonable fractional value between the m.Because the computational length of digital circuit is limited, when v is unreasonable decimal, can approximate approximate reasonable decimal); According to the demands of different of engineering precision, order v is divided into three types floating data, and the calculated data type of its first algorithm unit circuit, 1 to the 8th
algorithm unit circuit 8 also is divided into corresponding three types: the 1st type; Single (accounts for 4 byte of memorys; Computational length 32bit, 6~7 of significant digits,
evaluation scope 10
-37~10
38); The 2nd type, double (accounts for 8 byte of memorys, computational length 64bit, 15~16 of significant digits,
evaluation scope 10
-307~10
308); The 3rd type, long double (accounts for 16 byte of memorys, computational length 128bit, 18~19 of significant digits,
evaluation scope 10
-4931~10
4932).The fractional calculus filter of the digital picture of the high precision computation that the present invention proposes comprises following circuit block, its concrete structure as follows:
See Fig. 1, according to the different in kind of processing digital images, when handling gray level image, RGB cuts little ice to RGB transducer 14 with HSI to HSI transducer 9, directly the gray value of output digital video image; When handling digital color image, among RGB is transformed into the HSI color space with digital video image from rgb color space to HSI transducer 9.Among HSI is transformed into rgb color space with digital video image from the HSI color space to RGB transducer 14.
See Fig. 1, line storage group 10 is made up of sequential control circuit, read/write address generator and two-port RAM group; Sequential control circuit produces control corresponding read/write address generator, two-port RAM group, lock phase/shift circuit group 11, fractional calculus mask convolution circuit 12 down in the triggering of the row of input digit video flowing, a useful signal and operates required timing control signal with maximum comparator 13; The read/write address generator produces the read/write address of two-port RAM under the effect of timing control signal, and is responsible for handling read/write address initialization and rotating problem; Line storage group 10 is according to the input characteristics of serial digital video code stream; Utilize current input pixel; According to the different in kind of processing digital images, line storage group 10 adopts 2n line storage to accomplish the gray value of 2n+1 line of numbers video image or obtaining of the luminance component I value in the HSI color space.
See Fig. 1, lock phase/shift circuit group 11 is utilized current input pixel according to the input characteristics of serial digital video code stream, and according to the different in kind of the digital picture of handling, lock phase/shift circuit group 11 adopts 3n altogether
2+ 3n d type flip flop carries out a time-delay through the luminance component I to gray level image or digital color image and produces required (2n+1) * (2n+1) pel array of calculating digital image fractional order calculus; (2n+1) * (2n+1) the 1st of pel array the row adopts 2n d type flip flop, and the 2nd row adopts 2n-1 d type flip flop, all is to subtract n+1 d type flip flop of the capable employing of one, the n line by line until the capable every row of n adopts the number of d type flip flop; (2n+1) * (2n+1) 2n d type flip flop of the capable employing of the n+1 of pel array; (2n+1) * (2n+1) n+1 d type flip flop of the capable employing of the n+2 of pel array, n+2 d type flip flop of the capable employing of n+3 all is to add 2n d type flip flop of the capable employing of one, the 2n+1 line by line until the capable every row of 2n+1 adopts the number of d type flip flop.
See Fig. 1; Fractional calculus mask convolution circuit 12 is circuit blocks of realizing the fractional calculus most critical of digital picture in all forming circuit parts of fractional calculus filter of digital picture of high precision computation of the present invention, also is the core content of fractional calculus filter new departure of the present invention's digital picture of proposing this high precision computation.For the circuit that clearly demonstrates fractional calculus mask convolution circuit 12 constitutes, be necessary that elder generation carries out following brief description to the operation rule of fractional calculus mask convolution circuit:
Estimate down at Euclidean, what fractional calculus was the most frequently used is Gr ü mwald-Letnikov definition:
Wherein, the duration of signal s (x) is x ∈ [a, x].V can be any plural number, and it is any real number (comprising mark) that the present invention gets v.
expression is based on the fractional order differential operator of Gr ü mwald-Letnikov definition.When v is that
is the fractional order integration operator when bearing real number; When v is zero;
is the all-pass wave filtering operator, neither also integration not of differential; When v was arithmetic number,
was the fractional order differential operator.The Gr ü mwald-Letnikov of fractional calculus is defined in the integer step that Euclidean estimates down integer rank calculus and is generalized to the mark step-length, thereby the integer rank of calculus are generalized to fractional order.The calculating of the Gr ü mwald-Letnikov definition of fractional calculus is simple and easy to do; It only need be relevant with signal s (x) self the discrete sampling value of
, and derivative and the integrated value of undesired signal s (x).
Be without loss of generality, the present invention makes a=0, and the duration x ∈ [0, x] of signal s (x) is carried out the N five equilibrium, total N+1 node.The value of this N+1 cause and effect pixel is respectively s
N≡ s (0), s
N-1≡ s (x/N) ..., s
k≡ s (x-kx/N) ..., s
0≡ s (x).Wherein, k gets the integer between 0~N.Because Digital Image Processing is stored in the medium and carries out, in addition, owing to the border of image can be expanded it through the mode of cycle developing, so we can also handle the non-causal pixel in the digital picture.The value of this N+1 non-causal pixel is respectively s
0≡ s (x), s
-1≡ s (x+x/N) ..., s
-k≡ s (x+kx/N) ..., s
-N≡ s (2x).Gr ü mwald-Letnikov definition based on fractional calculus; When enough big of N; Can remove limit symbol, so can derive to such an extent that
sets up.Wherein,
expression Gamma function.In order to improve convergence rate and convergence precision; The present invention
is improved to
so; Except v=0; ± 2; ± 4; Outside, also introduced the signal value of signal s (x) in
at non-node place.Three adjacent segments point values of the present invention's number of winning the confidence s (x)
means of Lagrange three point interpolation formula, so can derive:
Then; The present invention makes
that signal s (x) is carried out the mark interpolation, can derive:
And then can derive:
Be an approximate formula, it is summed up as simple multiplication and add operation with the fractional calculus of signal s (x), and it has also relaxed the condition restriction of N → ∞ in the fractional calculus Gr ü mwald-Letnikov definition simultaneously.
Because what computer or digital filter were handled is digital quantity, its value is limited; In addition; Because the gray value of picture signal or the maximum variable quantity of luminance component I value are limited; It is between two neighbors that the gray value of digital picture or luminance component I value change the beeline that takes place; So two-dimensional digital image s (x, y) duration on x and y direction of principal axis (size of image array) possibly be that unit measures with the pixel only, the minimum range between the neighbor only possibly be a pixel.(x, branches such as minimum y) possibly be pixel units only to s at interval.If s (x; Y) duration of variable x and y is respectively [0 in; X] and [0; Y]; Then the branches such as unit on x and y direction of principal axis be respectively at interval
and
its maximum isodisperse be respectively
and
visible; For Digital Image Processing; Even fractional calculus means of mask dimensions number is big to equaling digital picture s (x; Y) size number (is n+2=N=min (L; H)); Possibly be that the maximum of its fractional calculus analytic value is approached only also, and can not be equal to the analytic value of its fractional calculus.When k=n≤N-1, the s that can derive (x, y) the approximate backward difference of the preceding n+2 item of the inclined to one side calculus of fractional order is respectively on x and y reference axis negative direction:
With
See Fig. 1; Because in digital picture; Has very big correlation between neighborhood interior pixel and the pixel; In order to strengthen the anti-image rotatory of fractional calculus
mask convolution circuit 12; Make it can on 8 symmetry directions, accomplish (x respectively to image s; Y) fractional calculus computing; (the present invention constructs the fractional calculus mask (expression with
on x axle negative direction respectively for x, y) the approximate backward difference of the preceding n+2 item of the inclined to one side calculus of fractional order on x and y reference axis negative direction to utilize s; See Fig. 2), the fractional calculus mask on the y axle negative direction (Fig. 4 is seen in expression with
).In addition, the fractional calculus mask on the x axle positive direction (is used
Fig. 3 is seen in expression), the fractional calculus mask on the y axle positive direction (uses
Fig. 5 is seen in expression), a left side down the fractional calculus mask on the diagonal (use W
LDDFig. 6 is seen in expression), the fractional calculus mask on the upper right diagonal (uses W
RUDFig. 7 is seen in expression), the fractional calculus mask on the upper left diagonal (uses W
LUDFig. 8 is seen in expression), the fractional calculus mask on the diagonal line direction (uses W
RDDFig. 9 is seen in expression) with
With
Aufbauprinciple and method similar, repeat no more here.The present invention utilizes the fractional calculus mask on above-mentioned 8 directions can calculate s (x, y) approximation of the v rank fractional calculus on diagonal, upper right diagonal, upper left diagonal and 8 directions of diagonal line under x axle negative direction, x axle positive direction, y axle negative direction, y axle positive direction, the left side respectively.For computing easy, the present invention with s (x, y) in the approximation of the maximum v rank fractional calculus of above-mentioned 8 direction patrix values as s (x, the approximation of v rank fractional calculus y).
See Fig. 2, Fig. 3, Fig. 4, Fig. 5, Fig. 6, Fig. 7, Fig. 8 and Fig. 9,
Be to cover s on the non-causal pixel
-1The fractional calculus mask numerical value of=s (x+x/N).
Be to cover point-of-interest s
0Fractional calculus mask numerical value on the=s (x).When k → n=1, the present invention can construct 3 * 3 fractional calculus mask; When k → n=3, the present invention can construct 5 * 5 fractional calculus mask; When k → n=2m-1, the present invention can construct the fractional calculus mask of (n+2) * (n+2)=(2m+1) * (2m+1), and wherein n gets odd number, and m gets natural number.Fractional calculus mask numerical value of the present invention is respectively:
(n+2) * (n+2) of the present invention fractional calculus mask is a sparse matrix, and its mask coefficient is a n+2 nonzero value, and they all are the functions of fractional order differential order v.Can prove that said n+2 a fractional calculus mask coefficient sum is not equal to zero, this is one of image fractional calculus mask and the remarkable difference on characteristic of image integer rank calculus mask.
Because the Digital Image Processing of computer or digital filter is directly to be treated to the basis to discrete pixel; So the numerical operation rule of fractional calculus mask adopts the airspace filter scheme of fractional calculus mask convolution too, the mode of this space filtering be exactly in pending digital picture pointwise move and corresponding long-pending and fractional calculus mask.The operation rule of fractional calculus mask convolution circuit 12 is airspace filters that the scheme of employing fractional calculus mask convolution realizes digital image fractional order calculus, is fit to realize data image signal is carried out handled with hardware circuit.Fractional calculus mask convolution circuit 12 to the operation rule of gray level image and digital color image is:
A. fractional calculus
mask convolution circuit 12 is to the operation rule of gray level image.The gray level image s of L * H (x, y) in, on above-mentioned 8 directions, carry out convolutional filtering with the fractional calculus mask of (n+2) * (n+2)=(2m+1) * (2m+1) of the present invention,
W
LDD, W
RUD, W
LUDAnd W
RDDThe numerical operation rule be respectively:
B. fractional calculus mask convolution circuit 12 is to the operation rule of digital color image.Because digital color image s (x; Y) there is correlation between each component of the RGB of each pixel in; And its value generally is limited between [0,255], therefore; Fractional calculus can destroy the correlation of R, G, three components of B, and color distortion possibly appear in the coloured image of handling through fractional calculus.For these reasons, the present invention only handles the fractional calculus of the luminance component I value of digital color image in the HSI color space.Fractional calculus mask convolution circuit 12 is identical with its operation rule and parameter to gray level image to the operation rule of the luminance component I value of digital color image.
Specify the circuit structure of fractional calculus mask convolution circuit 12 below: see Fig. 1, fractional calculus mask convolution circuit 12 is made up of the specific first algorithm unit circuit, 1 to the 8th algorithm unit circuit 8 of 8 parallel computations; The first algorithm unit circuit, 1 to the 8th algorithm unit circuit 8 calculates the approximation of the gray value of each pixel of digital video image or the v rank fractional calculus of the luminance component I value in the HSI color space on diagonal, upper right diagonal, upper left diagonal and 8 directions of diagonal line under x axle negative direction, x axle positive direction, y axle negative direction, y axle positive direction, the left side respectively.
See Fig. 1 and Figure 10, each algorithm unit circuit is made up of the n+2 identical with fractional calculus mask size number first multiplier to the, seven
multipliers 15~21 and an
adder 22; The non-zero weights of this n+2 multiplier are respectively in order:
,
With
The output valve feed-in
maximum comparator 13 of
adder 22.
See Fig. 1, fractional calculus mask convolution circuit 12 is made up of the specific first algorithm unit circuit, 1 to the 8th algorithm unit circuit 8 of following 8 parallel computations:
The first algorithm unit circuit, 1 calculating pixel S
xThe approximation of (k+n (H+1)) inclined to one side calculus of v rank fractional order on x axle negative direction; Pixel S
x(gray value of k+n (H+1)+H) or luminance component I value and the weights in the HSI color space
Difference feed-in
first multiplier 15, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)) or luminance component I value and the weights in the HSI color space
Difference feed-in
second multiplier 16, back feed-in
adder 22 multiplies each other; Pixel S
x(gray value of k+n (H+1)-H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
3rd multiplier 17, back feed-in
adder 22 multiplies each other; By that analogy, if 1≤i≤n+2, pixel S
x(gray value of k+n (H+1)-iH) or luminance component I value and the weights in the HSI color space
Difference feed-in the
4th multiplier 18, back feed-in
adder 22 multiplies each other; By that analogy, pixel S
xThe gray value of (k+n (H+1)-(n-2) H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
5th multiplier 19, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)-(n-1) H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
6th multiplier 20, back feed-in
adder 22 multiplies each other; Pixel S
x(gray value of k+n (H+1)-nH) or luminance component I value and the weights in the HSI color space
Difference feed-in the
7th multiplier 21, back feed-in
adder 22 multiplies each other.
The second algorithm unit circuit, 2 calculating pixel S
xThe approximation of (k+n (H+1)) inclined to one side calculus of v rank fractional order on x axle positive direction; Pixel S
x(gray value of k+n (H+1)-H) or luminance component I value and the weights in the HSI color space
Difference feed-in
first multiplier 15, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)) or luminance component I value and the weights in the HSI color space
Difference feed-in
second multiplier 16, back feed-in
adder 22 multiplies each other; Pixel S
x(gray value of k+n (H+1)+H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
3rd multiplier 17, back feed-in
adder 22 multiplies each other; By that analogy, if 1≤i≤n+2, pixel S
x(gray value of k+n (H+1)+iH) or luminance component I value and the weights in the HSI color space
Difference feed-in the
4th multiplier 18, back feed-in
adder 22 multiplies each other; By that analogy, pixel S
xThe gray value of (k+n (H+1)+(n-2) H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
5th multiplier 19, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)+(n-1) H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
6th multiplier 20, back feed-in
adder 22 multiplies each other; Pixel S
x(gray value of k+n (H+1)+nH) or luminance component I value and the weights in the HSI color space
Difference feed-in the
7th multiplier 21, back feed-in
adder 22 multiplies each other.
Algorithm element circuit 3 calculating pixel S
xThe approximation of (k+n (H+1)) inclined to one side calculus of v rank fractional order on y axle negative direction; Pixel S
xThe gray value of (k+n (H+1)+1) or luminance component I value and the weights in the HSI color space
Difference feed-in
first multiplier 15, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)) or luminance component I value and the weights in the HSI color space
Difference feed-in
second multiplier 16, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)-1) or luminance component I value and the weights in the HSI color space
Difference feed-in the
3rd multiplier 17, back feed-in
adder 22 multiplies each other; By that analogy, if 1≤i≤n+2, pixel S
x(gray value of k+n (H+1)-i) or luminance component I value and the weights in the HSI color space
Difference feed-in the
4th multiplier 18, back feed-in
adder 22 multiplies each other; By that analogy, pixel S
xThe gray value of (k+n (H+1)-(n-2)) or luminance component I value and the weights in the HSI color space
Difference feed-in the
5th multiplier 19, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)-(n-1)) or luminance component I value and the weights in the HSI color space
Difference feed-in the
6th multiplier 20, back feed-in
adder 22 multiplies each other; Pixel S
x(gray value of k+n (H+1)-n) or luminance component I value and the weights in the HSI color space
Difference feed-in the
7th multiplier 21, back feed-in
adder 22 multiplies each other.
The 4th
algorithm unit circuit 4 calculating pixel S
xThe approximation of (k+n (H+1)) inclined to one side calculus of v rank fractional order on y axle positive direction; Pixel S
xThe gray value of (k+n (H+1)-1) or luminance component I value and the weights in the HSI color space
Difference feed-in
first multiplier 15, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)) or luminance component I value and the weights in the HSI color space
Difference feed-in
second multiplier 16, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)+1) or luminance component I value and the weights in the HSI color space
Difference feed-in the
3rd multiplier 17, back feed-in
adder 22 multiplies each other; By that analogy, if 1≤i≤n+2, pixel S
x(gray value of k+n (H+1)+i) or luminance component I value and the weights in the HSI color space
Difference feed-in the
4th multiplier 18, back feed-in
adder 22 multiplies each other; By that analogy, pixel S
xThe gray value of (k+n (H+1)+(n-2)) or luminance component I value and the weights in the HSI color space
Difference feed-in the
5th multiplier 19, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)+(n-1)) or luminance component I value and the weights in the HSI color space
Difference feed-in the
6th multiplier 20, back feed-in
adder 22 multiplies each other; Pixel S
x(gray value of k+n (H+1)+n) or luminance component I value and the weights in the HSI color space
Difference feed-in the
7th multiplier 21, back feed-in
adder 22 multiplies each other.
The 5th algorithm unit circuit 5 calculating pixel S
xThe approximation of (k+n (H+1)) inclined to one side calculus of v rank fractional order under a left side on the diagonal; Pixel S
x(gray value of k+n (H+1)+1-H) or luminance component I value and the weights in the HSI color space
Difference feed-in
first multiplier 15, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)) or luminance component I value and the weights in the HSI color space
Difference feed-in
second multiplier 16, back feed-in
adder 22 multiplies each other; Pixel S
x(gray value of k+n (H+1)-1+H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
3rd multiplier 17, back feed-in
adder 22 multiplies each other; By that analogy, if 1≤i≤n+2, pixel S
x(gray value of k+n (H+1)-i+iH) or luminance component I value and the weights in the HSI color space
Difference feed-in the
4th multiplier 18, back feed-in
adder 22 multiplies each other; By that analogy, pixel S
xThe gray value of (k+n (H+1)-(n-2)+(n-2) H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
5th multiplier 19, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)-(n-1)+(n-1) H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
6th multiplier 20, back feed-in
adder 22 multiplies each other; Pixel S
x(gray value of k+n (H+1)-n+nH) or luminance component I value and the weights in the HSI color space
Difference feed-in the
7th multiplier 21, back feed-in
adder 22 multiplies each other.
The 6th
algorithm unit circuit 6 calculating pixel S
xThe approximation of (k+n (H+1)) inclined to one side calculus of v rank fractional order on upper right diagonal; Pixel S
x(gray value of k+n (H+1)-1+H) or luminance component I value and the weights in the HSI color space
Difference feed-in
first multiplier 15, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)) or luminance component I value and the weights in the HSI color space
Difference feed-in
second multiplier 16, back feed-in
adder 22 multiplies each other; Pixel S
x(gray value of k+n (H+1)+1-H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
3rd multiplier 17, back feed-in
adder 22 multiplies each other; By that analogy, if 1≤i≤n+2, pixel S
x(gray value of k+n (H+1)+i-iH) or luminance component I value and the weights in the HSI color space
Difference feed-in the
4th multiplier 18, back feed-in
adder 22 multiplies each other; By that analogy, pixel S
xThe gray value of (k+n (H+1)+(n-2)-(n-2) H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
5th multiplier 19, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)+(n-1)-(n-1) H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
6th multiplier 20, back feed-in
adder 22 multiplies each other; Pixel S
x(gray value of k+n (H+1)+n-nH) or luminance component I value and the weights in the HSI color space
Difference feed-in the
7th multiplier 21, back feed-in
adder 22 multiplies each other.
The 7th algorithm unit circuit 7 calculating pixel S
xThe approximation of (k+n (H+1)) inclined to one side calculus of v rank fractional order on upper left diagonal; Pixel S
x(gray value of k+n (H+1)+1+H) or luminance component I value and the weights in the HSI color space
Difference feed-in
first multiplier 15, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)) or luminance component I value and the weights in the HSI color space
Difference feed-in
second multiplier 16, back feed-in
adder 22 multiplies each other; Pixel S
x(gray value of k+n (H+1)-1-H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
3rd multiplier 17, back feed-in
adder 22 multiplies each other; By that analogy, if 1≤i≤n+2, pixel S
x(gray value of k+n (H+1)-i-iH) or luminance component I value and the weights in the HSI color space
Difference feed-in the
4th multiplier 18, back feed-in
adder 22 multiplies each other; By that analogy, pixel S
xThe gray value of (k+n (H+1)-(n-2)-(n-2) H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
5th multiplier 19, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)-(n-1)-(n-1) H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
6th multiplier 20, back feed-in
adder 22 multiplies each other; Pixel S
x(gray value of k+n (H+1)-n-nH) or luminance component I value and the weights in the HSI color space
Difference feed-in the
7th multiplier 21, back feed-in
adder 22 multiplies each other.
The 8th
algorithm unit circuit 8 calculating pixel S
xThe approximation of (k+n (H+1)) inclined to one side calculus of v rank fractional order on diagonal line direction; Pixel S
x(gray value of k+n (H+1)-1-H) or luminance component I value and the weights in the HSI color space
Difference feed-in
first multiplier 15, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)) or luminance component I value and the weights in the HSI color space
Difference feed-in
second multiplier 16, back feed-in
adder 22 multiplies each other; Pixel S
x(gray value of k+n (H+1)+1+H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
3rd multiplier 17, back feed-in
adder 22 multiplies each other; By that analogy, if 1≤i≤n+2, pixel S
x(gray value of k+n (H+1)+i+iH) or luminance component I value and the weights in the HSI color space
Difference feed-in the
4th multiplier 18, back feed-in
adder 22 multiplies each other; By that analogy, pixel S
xThe gray value of (k+n (H+1)+(n-2)+(n-2) H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
5th multiplier 19, back feed-in
adder 22 multiplies each other; Pixel S
xThe gray value of (k+n (H+1)+(n-1)+(n-1) H) or luminance component I value and the weights in the HSI color space
Difference feed-in the
6th multiplier 20, back feed-in
adder 22 multiplies each other; Pixel S
x(gray value of k+n (H+1)+n+nH) or luminance component I value and the weights in the HSI color space
Difference feed-in the
7th multiplier 21, back feed-in
adder 22 multiplies each other.
See Fig. 1, maximum comparator 13 value that mould value in the first algorithm unit circuit, 1 to the 8th algorithm unit circuit, 8 output valves of fractional calculus mask convolution circuit 12 is maximum is as pixel S
xThe approximation of the v rank fractional calculus of (k+n (H+1)).Maximum comparator 13 has 8 tunnel inputs, 1 tunnel output, and the output valve of the feed-in first algorithm unit circuit 1 to the 8th algorithm unit circuit 8 is exported the maximum value of mould value in above-mentioned 8 feed-in values respectively.
New departure below in conjunction with the fractional calculus filter of the digital picture of accompanying drawing and the high precision computation of the present invention of circuit arrangement example in detail: