CN104657553A - Similarity measurement hardware accelerating method based on rapid normalized cross correlation method - Google Patents

Abstract

The invention provides a similarity measurement hardware accelerating method based on a rapid normalized cross correlation method. The similarity measurement hardware accelerating method comprises the following steps of firstly establishing a mathematic model of correlation coefficients between a real-time image and a template image; then performing a hardware algorithm design by use of a two-layer production line, and calculating the correlation coefficients between the real-time image selected by a search window and the template image; after all the correlation coefficients between the real-time image selected by the search window and the template image are calculated, and finding out the maximal value of the correlation coefficients and the horizontal coordinate and longitudinal coordinate of the top left corner of the corresponding search window in a real-time image coordinate system, wherein the selected region corresponding to the search window is a matched region. According to the method, the hardware resource cost is reasonably controlled, 100M requirement is met under Xilinx Virtex5-XC5VFX100T, similarity matching calculation for a 80*64 real-time image and a 25*25 template image only needs 3.5ms, and thus the algorithm speed is greatly improved.

Description

A kind of hardware-accelerated method of similarity measure based on quick normalized crosscorrelation method

Technical field

The present invention relates to a kind of hardware-accelerated method, particularly a kind of hardware-accelerated method of similarity measure based on quick normalized crosscorrelation method, is applicable to tailor-made algorithm hardware circuit design field.

Background technology

The information processing technology based on images match has started to widely apply in space flight model of new generation, and the real-time of image procossing directly affects guidance precision.Adopted software simulating image processing algorithm to be more and more difficult to meet the requirement of real-time of space flight model based on general processor (as DSP) in the past.At present, the mode often adopting shortcut calculation to sacrifice arithmetic accuracy reduces algorithm operation time.Accelerated by hardware algorithm implementation algorithm to be the most effective means reducing algorithm operation time.

The matching process that what current image procossing adopted is based on half-tone information, this method matching precision is high but operand is larger.Image matching algorithm based on similarity measure is that wherein calculated amount is maximum, the longest algorithm consuming time.Similarity measurement criterion is the criterion calculating similarity degree between To Template figure and real-time figure.In numerous similarity judging basis, normalized crosscorrelation method (NCC) is considered to best.It is high that it has coupling degree of accuracy, the characteristics such as anti-ground unrest ability is strong, pattern distortion strong adaptability (when the distortion difference that image rotates around central point, relative convergent-divergent causes is little, its matching result also can be satisfactory).Based on above feature, normalized crosscorrelation method obtains extensively and effectively applies in matching problem.But normalized crosscorrelation method computational complexity is higher, and computing time is expended longer, the coupling application higher for requirement of real-time has certain limitation.Image matching algorithm many employings High Performance DSP now based on the similarity measure of normalized crosscorrelation method realizes.But the increase of the increase operand along with picture size, this scheme is difficult to the requirement of real-time meeting space flight model.For meeting real-time, even adopting and sacrificing the mode of arithmetic accuracy, based on similarity measure algorithm hardware-accelerated extremely urgent of normalized crosscorrelation method.

Summary of the invention

Technology of the present invention is dealt with problems and is: overcome the deficiencies in the prior art, provide a kind of hardware-accelerated method of similarity measure based on normalized crosscorrelation method, by Optimal improvements related coefficient algorithmic formula, two-layer the pipeline design is adopted in hardware calculates, conservative control hardware resource cost, state machine is adopted to carry out the calculating of related coefficient, and under Xilinx virtex5-XC5VFX100T, meet the requirement reaching 100M, greatly improve algorithm speed, meet the demand of tailor-made algorithm hardware circuit design to the full extent.

Technical solution of the present invention is: a kind of hardware-accelerated method of similarity measure based on quick normalized crosscorrelation method, and step is as follows:

(1) mathematical model of related coefficient between figure and Prototype drawing is in real time set up; Described mathematical model is by formula:

$\mathrm{\ρ}(p,q)=\frac{\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}(x+p,y+q)\·T(x,y)-\stackrel{\‾}{T}\·\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}(x+p,y+q)}{\sqrt{\left(\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}{(s+p,y+q)}^2\right)\·m\·n-{\left(\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}(x+p,y+q)\right)}^2}\·\sqrt{\frac{\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}T{(x,y)}^2}{m\·n}-{\stackrel{\‾}{T}}^2}}$

Provide, the related coefficient between the real-time figure that ρ (p, q) selectes for search window and Prototype drawing; P and q is respectively the horizontal ordinate of the search window upper left corner in real-time figure coordinate system and ordinate, T (x, y) be (x of coordinate points in Prototype drawing, y) image intensity value, S (x+p, y+q) is the image intensity value of corresponding point in real-time figure, described coordinate points (x, y) Prototype drawing coordinate system is arranged in represent search window in real-time figure cover the gray average in subgraph region, represent the gray average of Prototype drawing, m is Prototype drawing width, n is Prototype drawing height, described search window is the rectangular window all identical with Prototype drawing size, shape, the true origin of described real-time figure coordinate system is the upper left corner of real-time figure, the upper left corner is to the right real-time figure coordinate system X-axis positive dirction, the upper left corner is downwards real-time figure coordinate system Y-axis positive dirction, the true origin of described Prototype drawing coordinate system is the upper left corner of Prototype drawing, the upper left corner is to the right Prototype drawing coordinate system X-axis positive dirction, and the upper left corner is downwards Prototype drawing coordinate system Y-axis positive dirction;

(2) utilize hardware from storer, read figure and Prototype drawing in real time, according to the mathematical model in step (1), try to achieve respectively in mathematical model with

(3) after the calculating in step (2) completes, enter step (4), the related coefficient between the real-time figure utilizing the result calculating search window in the middle mathematical model of step (1) and step (2) to select and Prototype drawing; Meanwhile, utilize hardware from storer, read next width and scheme in real time and Prototype drawing, return step (2);

(4) related coefficient between the real-time figure utilizing finite state machine method calculating search window to select and Prototype drawing;

(5) after the Calculation of correlation factor between the real-time figure that selectes of all search windows and Prototype drawing completes, find out horizontal ordinate in real-time figure coordinate system of maximal value in related coefficient and the corresponding search window upper left corner and ordinate, the selection area that this search window is corresponding is target area.

Hardware in described step (2) is FPGA or ASIC.

Related coefficient between the real-time figure utilizing finite state machine method calculating search window to select in described step (4) and Prototype drawing, is specially:

Described hardware comprises two dividers and two square root extractors, and described state machine comprises 6 states, is respectively sta0 ~ sta5, and under each state, the concrete operations of hardware are:

Sta0: the result of calculation in backup-step (2), if first time related coefficient computing, prepares Prototype drawing two divisions with divisor and dividend, and it is enable to open two dividers, calculate figure summation in real time square and the product of real-time figure quadratic sum and m*n forward sta1 to, if not first time related coefficient, calculate in real time figure summation square and the product of figure quadratic sum and m*n in real time, forward sta2 to;

Sta1: calculate preservation data after calculating completes also jump to sta2;

Sta2: calculate intermediate variable by the result in sta1 and determine the evolution number in related coefficient computing, jump to sta3;

Sta3: calculate the molecule in related coefficient mathematical model by the intermediate variable in sta2; According to the related coefficient computing evolution number determined in sta2, if first time related coefficient computing, then open Prototype drawing and real-time figure extracting operation is enable, carry out Prototype drawing and real-time figure extracting operation; If not first time related coefficient computing, then open real-time figure extracting operation enable, carry out real-time figure extracting operation; Jump to sta4;

Sta4: after figure extracting operation terminates in real time, calculates the product of figure and Prototype drawing evolution result in real time; Open divider enable; Jump to sta5;

Sta5: carry out division calculation and obtain related coefficient, jump to sta0 after division arithmetic terminates, carries out the computing of next width real-time figure related coefficient.

State in described state machine adopts gray encoding, and concrete corresponding relation is: sta0 ~ 000, sta1 ~ 001, sta2 ~ 010, sta3 ~ 011, sta4 ~ 111, sta5 ~ 110.

The present invention's beneficial effect is compared with prior art:

(1) the present invention carries out Hardware distortion to quick normalized crosscorrelation method formula, makes it be more suitable for Hardware and realizes, and the formula after change not to need after reads image data calculating mean value reads image data again, decreases computing; And will scheme in real time to be separated with Prototype drawing computing, in whole Calculation of correlation factor process, Prototype drawing only needs to calculate once, adopts prime formula to need first computed image mean value, is carrying out likeness coefficient calculating, needs to read twice view data;

(2) the present invention adopts two-layer the pipeline design, operation efficiency is substantially increased by parallel computation, adopt the divider and square root extractor optimized, reduce cycle and hardware algorithm resource needed for likeness coefficient computing, and by adopting the high-bit width storer supporting non-alignment access, reduce the image reading time, reduce algorithm operation time;

(3) present invention employs finite state machine to control the multiplexing computing of divider, square root extractor, complete the calculating of formula of correlation coefficient, Appropriate application hardware resource, improves the efficiency of computing;

(4) realize based on image matching algorithm many employings High Performance DSP of the similarity measure of normalized crosscorrelation method at present, present invention achieves the Hardware Design of this algorithm, can be used for FPGA or ASIC to realize, substantially increase processing speed and the efficiency of this algorithm relative to DSP computing.

Accompanying drawing explanation

Fig. 1 is process flow diagram of the present invention;

Fig. 2 is Prototype drawing and scheme to mate schematic diagram in real time;

Fig. 3 is single match accumulation arithmetic pipelining schematic diagram;

Fig. 4 is that the two-layer streamline of related coefficient calculates schematic diagram;

Fig. 5 is that similarity measure state machine controls schematic diagram.

Embodiment

The content be not described in detail in instructions of the present invention belongs to the known technology of professional and technical personnel in the field.Below in conjunction with accompanying drawing, the specific embodiment of the present invention is further described in detail.

Similarity measurement refers to the similarity determining between feature to be matched with a kind of specific metric form, and whether its similarity obtained is the key foundation of images match, directly determine image and can correctly be mated.

Be illustrated in figure 1 process flow diagram of the present invention, as can be seen from Figure 1, a kind of hardware-accelerated method of similarity measure based on normalized crosscorrelation method that the present invention proposes, concrete steps are as follows:

(1) mathematical model of related coefficient between figure and Prototype drawing is in real time set up;

Similarity measure algorithm based on normalized crosscorrelation method is a kind of more classical statistical match algorithm, and its main processes of calculation is as follows: as shown in Figure 2, and the size of figure S is M × N in real time, and the size of Prototype drawing T is m × n.In real-time figure, yellow area represents Prototype drawing T on real-time figure S during translation, the subgraph region that search window covers.The coordinate position of (p, q) subgraph region top left corner apex in real-time figure S, then calculates the normalized correlation of subgraph and Prototype drawing by related function (formula 1).Images match is to scheming S from left to right in real time, from up to down carry out traversal search, then the related coefficient that all subgraph positions are corresponding is calculated, (M-m+1) * (N-n+1) secondary matching primitives, obtains the individual matching factor result of (M-m+1) * (N-n+1) altogether.The best match position calculated based on the similarity measure algorithm of normalized crosscorrelation method is the position corresponding to the maximum search subgraph of its related coefficient.

Related coefficient between the real-time figure that in formula 1, ρ (p, q) selectes for search window and Prototype drawing; P and q is respectively the horizontal ordinate of the search window upper left corner in real-time figure coordinate system and ordinate, T (x, y) be (x of coordinate points in Prototype drawing, y) image intensity value, S (x+p, y+q) is the image intensity value of corresponding point in real-time figure, described coordinate points (x, y) Prototype drawing coordinate system is arranged in represent search window in real-time figure cover the gray average in subgraph region, represent the gray average of Prototype drawing, m is Prototype drawing width, n is Prototype drawing height, described search window is the rectangular window all identical with Prototype drawing size, shape, the true origin of described real-time figure coordinate system is the upper left corner of real-time figure, the upper left corner is to the right real-time figure coordinate system X-axis positive dirction, the upper left corner is downwards real-time figure coordinate system Y-axis positive dirction, the true origin of described Prototype drawing coordinate system is the upper left corner of Prototype drawing, the upper left corner is to the right Prototype drawing coordinate system X-axis positive dirction, and the upper left corner is downwards Prototype drawing coordinate system Y-axis positive dirction;

$\mathrm{\ρ}(p,q)=\frac{\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}[S(s+p,y+q)-{\stackrel{\‾}{S}}_{p,q}][T(x,y)-\stackrel{\‾}{T}]}{{\left\{\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{[S(x+p,y+q)-{\stackrel{\‾}{S}}_{p,q}]}^2\right\}}^{1/2}{\left\{\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{[T(x,y)-\stackrel{\‾}{T}]}^2\right\}}^{1/2}}---\left(1\right)$

After being launched by formula mathematics, abbreviation can obtain:

$\mathrm{\ρ}(p,q)=\frac{\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}(s+p,y+q)\·T(x,y)-{\stackrel{\‾}{S}}_{p,q}\·\stackrel{\‾}{T}\·m\·n}{\sqrt{(\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}{(x+p,y+q)}^2-{{\stackrel{\‾}{S}}_{p,q}}^2\·m\·n)\·(\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}T{(x,y)}^2-{\stackrel{\‾}{T}}^2\·m\·n)}}---\left(2\right)$

In hardware operation, divide operations takes advantage of that operation is more time-consuming, expense resource far away than plus-minus, therefore reduce divide operations number of times as far as possible and can improve arithmetic speed.When each similarity matching calculates, Prototype drawing data are constant.Therefore divide operations forwarded on masterplate figure, such division only need calculate once as far as possible.Therefore formula is convertible as shown in formula (3), only needs the mean value of calculation template figure summation and Prototype drawing quadratic sum, no longer calculates real-time figure sum average value.

$\mathrm{\ρ}(p,q)=\frac{\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}(x+p,y+q)\·T(x,y)-\stackrel{\‾}{T}\·\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}(x+p,y+q)}{\sqrt{(\left(\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}{(x+p,y+q)}^2\right)\·m\·n-{\left(\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}(x+p,y+q)\right)}^2)\·(\frac{\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}T{(x,y)}^2}{m\·n}-{\stackrel{\‾}{T}}^2)}}---\left(3\right)$

Extracting operation generally needs multiple cycle, and the extracting operation time is directly proportional to square root extractor bit wide.Formula denominator is carried out Prototype drawing be separated with scheming in real time, such Prototype drawing part evolution only need calculate once, after can be used as constant participate in computing, square root extractor bit wide can reduce greatly, shortens the extracting operation time.Final application of formula conversion is as follows:

$\mathrm{\ρ}(p,q)=\frac{\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}(x+p,y+q)\·T(x,y)-\stackrel{\‾}{T}\·\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}(x+p,y+q)}{\sqrt{\left(\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}{(s+p,y+q)}^2\right)\·m\·n-{\left(\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}(x+p,y+q)\right)}^2}\·\sqrt{\frac{\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}T{(x,y)}^2}{m\·n}-{\stackrel{\‾}{T}}^2}}---\left(4\right)$

(2) utilize hardware from storer, read figure and Prototype drawing in real time, according to the mathematical model in step (1), try to achieve respectively in mathematical model with described hardware is FPGA or ASIC, selects Xilinx virtex5-XC5VFX100T in the present embodiment.

Calculate the employing of these cumulative datas and get location calculating, digital independent, the design of accumulation calculating three class pipeline.Prototype drawing and in real time figure read m*n gray scale respectively, are designed, once read multiple gray scale, can reduce single match cumulative operational time by memory optimization.The present invention once reads 4 gray scales, and by optimizing accumulation computing, single match cumulative operational time can be reduced to close to 1/4th before.Single match accumulation arithmetic pipelining schematic diagram is concrete as shown in Figure 3;

(3) after the calculating in step (2) completes, enter step (4), the related coefficient between the real-time figure utilizing the result calculating search window in the middle mathematical model of step (1) and step (2) to select and Prototype drawing; Meanwhile, utilize hardware from storer, read next width and scheme in real time and Prototype drawing, return step (2), it is concrete as shown in Figure 4 that the two-layer streamline of related coefficient calculates schematic diagram;

Single match starts to calculate related coefficient according to formula (4), meanwhile, carries out next coupling figure accumulation computing after obtaining the accumulated value such as mean value, summation, quadratic sum, sum of products of Calculation of correlation factor needs.Because view data is comparatively large, general single match is accumulated as the critical path of streamline.Designed by memory optimization, reduce single match cumulative operational time, and in the little situation of view data, the Calculation of correlation factor time may exceed single match cumulative operational time, becomes the critical path of streamline.

(4) related coefficient between the real-time figure utilizing finite state machine method calculating search window to select and Prototype drawing; Described state conversion process is concrete as shown in Figure 5, and as can be seen from Figure 5, the finite state machine method in the present invention is specially:

Described hardware comprises two dividers and two square root extractors, and described state machine comprises 6 states, is respectively sta0 ~ sta5, and under each state, the concrete operations of hardware are:

Sta0: the result of calculation in backup-step (2), if first time related coefficient computing, prepares Prototype drawing two divisions with divisor and dividend, and it is enable to open two dividers, calculate figure summation in real time square and the product of real-time figure quadratic sum and m*n forward sta1 to, if not first time related coefficient, calculate in real time figure summation square and the product of figure quadratic sum and m*n in real time, forward sta2 to;

Sta1: calculate preservation data after calculating completes also jump to sta2;

Sta2: calculate intermediate variable by the result in sta1 and determine the evolution number in related coefficient computing, jump to sta3;

Sta3: calculate the molecule in related coefficient mathematical model by the intermediate variable in sta2; According to the related coefficient computing evolution number determined in sta2, if first time related coefficient computing, then open Prototype drawing and real-time figure extracting operation is enable, carry out Prototype drawing and real-time figure extracting operation; If not first time related coefficient computing, then open real-time figure extracting operation enable, carry out real-time figure extracting operation; Jump to sta4;

Sta4: after figure extracting operation terminates in real time, calculates the product of figure and Prototype drawing evolution result in real time; Open divider enable; Jump to sta5;

Sta5: carry out division calculation and obtain related coefficient, jump to sta0 after division arithmetic terminates, carries out the computing of next width real-time figure related coefficient.

State in described state machine adopts gray encoding, and concrete corresponding relation is: sta0 ~ 000, sta1 ~ 001, sta2 ~ 010, sta3 ~ 011, sta4 ~ 111, sta5 ~ 110.

(5) after the Calculation of correlation factor between the real-time figure that selectes of all search windows and Prototype drawing completes, find out horizontal ordinate in real-time figure coordinate system of maximal value in related coefficient and the corresponding search window upper left corner and ordinate, the selection area that this search window is corresponding is target area.

The present invention FPGA under Xilinx virtex5-XC5VFX100T realizes, and meets the requirement reaching 100M.Similarity matching computing is carried out to the Prototype drawing of size to be the real-time figure of 80*64 and size be 25*25 and only needs 3.5ms, improve two orders of magnitude relative to travelling speed in TI C6416 series DSP.

Claims (4)

1., based on the hardware-accelerated method of similarity measure of quick normalized crosscorrelation method, it is characterized in that step is as follows:

(1) mathematical model of related coefficient between figure and Prototype drawing is in real time set up; Described mathematical model is by formula:

$\mathrm{\ρ}(p,q)=\frac{\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}(x+p,y+q)\·T(x,y)-\stackrel{\‾}{T}\·\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}(x+p,y+q)}{\sqrt{\left(\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}{(x+p,y+q)}^2\right)\·m\·n-{\left(\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}(x+p,y+q)\right)}^2}\sqrt{\frac{\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}T{(x,y)}^2}{m\·n}{\stackrel{\‾}{T}}^2}}$

Provide, the related coefficient between the real-time figure that ρ (p, q) selectes for search window and Prototype drawing; P and q is respectively the horizontal ordinate of the search window upper left corner in real-time figure coordinate system and ordinate, T (x, y) be (x of coordinate points in Prototype drawing, y) image intensity value, S (x+p, y+q) is the image intensity value of corresponding point in real-time figure, described coordinate points (x, y) Prototype drawing coordinate system is arranged in represent search window in real-time figure cover the gray average in subgraph region, represent the gray average of Prototype drawing, m is Prototype drawing width, n is Prototype drawing height, described search window is the rectangular window all identical with Prototype drawing size, shape, the true origin of described real-time figure coordinate system is the upper left corner of real-time figure, the upper left corner is to the right real-time figure coordinate system X-axis positive dirction, the upper left corner is downwards real-time figure coordinate system Y-axis positive dirction, the true origin of described Prototype drawing coordinate system is the upper left corner of Prototype drawing, the upper left corner is to the right Prototype drawing coordinate system X-axis positive dirction, and the upper left corner is downwards Prototype drawing coordinate system Y-axis positive dirction;

(2) utilize hardware from storer, read figure and Prototype drawing in real time, according to the mathematical model in step (1), try to achieve respectively in mathematical model $\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}F(x,y),$ $\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}(x+p,y+q),$ $\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}{(x+p,y+q)}^2,$ $\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}F{(x,y)}^2$ With $\underset{x=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{n-1}{\mathrm{\Σ}}}{S}_{p,q}(x+p,y+q)\·F(x,y)$

(3) after the calculating in step (2) completes, enter step (4), the related coefficient between the real-time figure utilizing the result calculating search window in the middle mathematical model of step (1) and step (2) to select and Prototype drawing; Meanwhile, utilize hardware from storer, read next width and scheme in real time and Prototype drawing, return step (2);

(4) related coefficient between the real-time figure utilizing finite state machine method calculating search window to select and Prototype drawing;

(5) after the Calculation of correlation factor between the real-time figure that selectes of all search windows and Prototype drawing completes, find out horizontal ordinate in real-time figure coordinate system of maximal value in related coefficient and the corresponding search window upper left corner and ordinate, the selection area that this search window is corresponding is target area.

2., according to a kind of hardware-accelerated method of similarity measure based on quick normalized crosscorrelation method in claim 1, it is characterized in that: the hardware in described step (2) is FPGA or ASIC.

3. according to a kind of hardware-accelerated method of similarity measure based on quick normalized crosscorrelation method in claim 1, it is characterized in that: the related coefficient between the real-time figure utilizing finite state machine method calculating search window to select in described step (4) and Prototype drawing, is specially:

Described hardware comprises two dividers and two square root extractors, and described state machine comprises 6 states, is respectively sta0 ~ sta5, and under each state, the concrete operations of hardware are:

Sta0: the result of calculation in backup-step (2), if first time related coefficient computing, prepares Prototype drawing two divisions with divisor and dividend, and it is enable to open two dividers, calculate figure summation in real time square and the product of real-time figure quadratic sum and m*n forward sta1 to, if not first time related coefficient, calculate in real time figure summation square and the product of figure quadratic sum and m*n in real time, forward sta2 to;

Sta1: calculate preservation data after calculating completes also jump to sta2;

Sta2: calculate intermediate variable by the result in sta1 and determine the evolution number in related coefficient computing, jump to sta3;

Sta3: calculate the molecule in related coefficient mathematical model by the intermediate variable in sta2; According to the related coefficient computing evolution number determined in sta2, if first time related coefficient computing, then open Prototype drawing and real-time figure extracting operation is enable, carry out Prototype drawing and real-time figure extracting operation; If not first time related coefficient computing, then open real-time figure extracting operation enable, carry out real-time figure extracting operation; Jump to sta4;

Sta4: after figure extracting operation terminates in real time, calculates the product of figure and Prototype drawing evolution result in real time; Open divider enable; Jump to sta5;

Sta5: carry out division calculation and obtain related coefficient, jump to sta0 after division arithmetic terminates, carries out the computing of next width real-time figure related coefficient.

4. according to a kind of hardware-accelerated method of similarity measure based on quick normalized crosscorrelation method in claim 3, it is characterized in that: the state in described state machine adopts gray encoding, concrete corresponding relation is: sta0 ~ 000, sta1 ~ 001, sta2 ~ 010, sta3 ~ 011, sta4 ~ 111, sta5 ~ 110.