Summary of the invention
The invention discloses a wide range of high definition rapid registering method and device of a kind of ophthalmology OCT image, its object is to
During several width ophthalmology OCT images are synthesized a panel height clear OCT image, any one figure and benchmark to be matched are used
Figure optimizes registration process in the vertical direction and the horizontal direction on time.
The technical solution of method of the present invention is as follows:
A kind of a wide range of high definition rapid registering method of ophthalmology OCT image, includes the following steps:
S1: that strongest width of signal-to-noise ratio is chosen as base from the n width OCT image that ophthalmology OCT device scanning human eye obtains
Quasi- figure, is registrated with the reference map respectively using remaining n-1 width OCT image as figure to be matched, obtains m width registration OCT figure
Picture;
S2: m width registration OCT image is synthesized into the clear OCT image of a panel height;
Wherein, m≤n-1.
Further: S11, optional i-th figure to be matched and the reference map are matched from described n-1 figure to be matched
Standard obtains being registrated OCT image described in a width;
S12, in method same with step S11, remaining n-2 figure to be matched is matched with the reference map respectively
Standard obtains remaining m-1 width registration OCT image;
Wherein, 1≤i≤n-2.
Further, step S11 specifically comprises the following steps:
S11A: i-th figure to be matched and the reference map are defined match for the first time and obtain maximum correlation when institute on time
Stating the location of i-th figure to be matched is p1_max, when obtaining time big correlation described in the location of i-th figure to be matched
For p1_sec;If | p1_max | more than the effective range R_limit of i-th map migration to be matched, abandons the i-th width and wait for
Figure;If p1_max is 0 and p1_sec is+R or-R, S11B is thened follow the steps;Wherein ,+R indicate it is described i-th it is to be matched
Figure moves right R pixel;- R is that i-th figure to be matched is moved to the left R pixel, R > R_limit;R is described i-th
The maximum moving range of figure to be matched;R_limit is related with the picture traverse of i-th figure to be matched;
S11B: define p=(p1_max+p1_sec)/2, after i-th figure to be matched is translated with p and the reference map into
Second of registration of row, becomes the i-th width and translates figure to be matched;Described in when will obtain maximum correlation in second of registration process
Position where i width translates figure to be matched is defined as p2_max, when obtaining time big correlation described in the i-th width translate figure institute to be matched
Position be defined as p2_sec, if | p2_max | more than R_limit, abandon i-th width and translate figure to be matched;Otherwise,
Execute step S11C;
S11C: if | p2_max-p2_sec | < translation precision, then it is assumed that i-th width translates the translation precision of figure to be matched
It meets the requirements, is transferred to step S11D;Otherwise, i-th width is translated figure to be matched and the reference map again by return step S11B
K registration is carried out, executes step S11D after translation precision is met the requirements;
When S11D: translating figure to be matched for i-th width and obtains maximum correlation after being registrated for the reference map progress k times
Position where i-th width translates figure to be matched is defined as pk_max, and i-th width is translated figure to be matched and the benchmark
Figure carries out vertical registration, obtains the registration OCT figure.
Further: in step S11A, the p1_max and the p1_sec are obtained as follows:
S11A1: by i-th figure to be matched in initial position and the reference map carry out vertical registration, obtain just
Beginning registration result align0;The initial relevance of initial registration result align0 and reference map is calculated according to evaluation function
value0;Wherein, the initial position the location of refers to when i-th figure to be matched does not translate;
S11A2: by i-th figure to be matched to R pixel of right translation, obtaining the i-th width and move to right figure to be matched, will be described
I-th width moves to right figure to be matched and the reference map carries out vertical registration, obtains the first registration result align1;According to evaluation function
Calculate the first correlation value1 of the first registration result align1 and the reference map;
S11A3: by i-th figure to be matched to R pixel of left, obtaining the i-th width and move to left figure to be matched, will be described
I-th width moves to left figure to be matched and the reference map carries out vertical registration, obtains the second registration result align2;According to evaluation function
Calculate the second correlation value2 of the second registration result align2 and the reference map;
S11A4: comparing the size of initial relevance value0, the first correlation value1 and the second correlation value2,
Maximum value is assigned to maximum correlation value_max, second largest value is assigned to time big correlation value_sec, and will and it is maximum
The corresponding translational displacement of correlation value_max is assigned to the p1_max, will assign with the secondary big corresponding translational displacement of correlation
It is worth to the p1_sec;
Wherein, step S11A2 can be exchanged with the sequence of step S11A3.
Further, the evaluation function are as follows: value=sum (align.*ref).
Further, i-th figure to be matched carries out vertical registration in initial position and the reference map, refers to first
Each column of i-th figure to be matched and the reference map are carried out to the registration of vertical direction when beginning position.
A kind of a wide range of high definition rapid registering device of ophthalmology OCT image is also disclosed in the present invention, comprising:
OCT image registration module, is used for: choosing noise from the n width OCT image that ophthalmology OCT device scanning human eye obtains
Than a strongest width as reference map, match respectively with the reference map using remaining n-1 width OCT image as figure to be matched
Standard obtains m width registration OCT image;
High definition OCT image synthesis module, is used for: m width registration OCT image is synthesized the clear OCT image of a panel height.
Advantageous effects of the invention: it from several OCT images that ophthalmology OCT device scanning human eye obtains, chooses
The strongest width of signal-to-noise ratio is registrated with reference map as reference map using remaining OCT image as figure to be matched respectively.Its
In, i-th is arbitrarily selected from several figures to be matched, is registrated with reference map, method for registering is as follows: first by i-th
Figure to be matched and reference map carry out vertical registration, then utilize the multiple horizontal registration of dichotomy progress with reference map again.I-th
To be matched figure and reference map carry out in first time registration process, if i-th figure and reference map to be matched with obtaining maximum on time
The absolute value of the location of i-th figure to be matched p1_max described in when correlation value_max | p1_max | more than the i-th width
The effective range R of map migration to be matched then weeds out i-th figure to be matched, and computer automatically selects next figure to be matched
It is registrated with reference map;If match in second and later each time and reference map of i-th figure to be matched obtains most on time
The location of i-th figure to be matched described in when big correlation value_max pk_max is 0, then i-th figure to be matched and
Reference map be iterated in later multiple horizontal registration using dichotomy, until translation precision meets the requirements;When meeting
After translating required precision, then by i-th figure and reference map progress vertical registration to be matched, obtain width registration OCT figure.With with
The same method that i-th figure to be matched is registrated with reference map matches remaining figure and reference map to be matched, if obtaining
Dry registration OCT figure;Finally, all registration OCT figures are synthesized a high-definition image.Therefore side announced through the invention
Method and device, reduce cycle-index, improve the speed and precision of algorithm.
Specific embodiment
In order to which technical problems, technical solutions and advantages to be solved are more clearly understood, tie below
Accompanying drawings and embodiments are closed, the present invention will be described in further detail.It should be appreciated that specific embodiment described herein is only used
To explain the present invention, it is not intended to limit the present invention.
With reference to Fig. 1, Fig. 1 is the total flow chart of the present invention, comprising:
S1: the strongest width of signal-to-noise ratio is chosen as base from the n width OCT image that ophthalmology OCT device scanning human eye obtains
Quasi- figure, is registrated with reference map respectively using remaining n-1 width OCT image as figure to be matched, obtains m width registration OCT image;
S2: m width registration OCT image is superposed to width synthesis high definition OCT image.
For the ophthalmology OCT device mentioned in step S1, with reference to Fig. 2, comprising: light source 100, spectral module 200, reference arm
Module 300, sample arm module 500, detecting module 600 and control system 700.The working principle of ophthalmology OCT device is as follows: light source
100 light issued are transferred to spectral module 200, and the light received is divided into reference light and detection light by spectral module 200, wherein joining
It examines light and passes to reference arm module 300, detection light passes to sample arm module 500.The reference that reference arm module 300 will receive
Light transfers back to spectral module 200 after reflection;It detects light to collimate by collimating mirror 400, scans eyes through sample arm module 500
800;Detection light forms signal light after the tissue scatter in eyes 800, then through sample arm module 500 and collimating mirror 400, returns
Spectral module 200, signal light and reference light form interference light after interfering at spectral module 200.Detecting module 600 is by the interference
Optical transport to control system 700, control system 700 handles the interference light, obtains the OCT signal on eyeground, then by control system
The display device (such as display screen) of 700 (computers) of system shows the OCT tomograph of scanned human eye (see Fig. 3).Wherein, it refers to
Arm module 300 is built-in with reference mirror 301, and spectral module 200 is divided obtained reference light and returns to after the reflection of reference mirror 301
In spectral module 200.It should be noted that in general, collimating mirror 400 should belong to a part of sample arm module 500,
Only in the technical scheme, by collimating mirror 400 independently of sample arm module 500, but this has no effect on the work of OCT system
Make principle.
With reference to Fig. 4, when scanning human eye using OCT device, after collecting n width OCT image as shown in Figure 3, from this n width
The strongest width of signal-to-noise ratio is chosen in OCT image as reference map, regard remaining n-1 width OCT figure as figure to be matched, point
It is not registrated one by one with reference map, obtains m width registration OCT figure.Due to this n-1 figure to be matched each width and reference map
It is registrated during obtaining registration OCT figure, can be repeatedly registrated with reference map, it, may in multiple registration process
There are small part figure to be matched and reference map to obtain the location of maximum correlation epoch registration figure pk_ in certain primary registration
The absolute value of max | pk_max | more than the effective range R_limit of offset.For | pk_max | more than the effective range of offset
The figure to be matched of R_limit, then weed out them.Here the picture traverse of R_limit and i-th figure to be matched is related, k generation
The registration number of table figure to be matched and reference map.Therefore the width number m of the registration OCT figure finally obtained may be less than n-1.When
So, the most ideal situation is that, each width of n-1 figure to be matched carries out in registration process several times with reference map, and matching will definitely
Pk_max is then obtained in the effective range R_limit of map migration to be matched the location of when the maximum correlation arrived
The width number m for being registrated OCT figure is equal to n-1.Therefore, the value range of m is m≤n-1.
Specifically, step S1 expands into following several steps:
S11, optional i-th width and the reference map are registrated from described n-1 figure to be matched, obtain matching described in a width
Quasi- OCT image;
S12, in method same with step S11, remaining n-2 figure to be matched is matched with the reference map respectively
Standard obtains remaining m-1 width registration OCT image;
Wherein, 1≤i≤n-2.
That is, any i-th figure (1≤i≤n-2) to be matched and reference map can be matched by step S11
Standard obtains width registration OCT image.Then with same method, respectively and benchmark by remaining remaining n-3 figure to be matched
Figure is registrated, and m-1 width registration OCT image is obtained.
Below step S11 is unfolded to describe, how i-th figure and reference map to be matched are registrated by specific introduce, and obtain
OCT image is registrated to a width.
Step S11 is specifically decomposed into the following steps:
Step S11A: i-th figure and reference map to be matched are subjected to first time registration, if obtained most in registration process
The location of i-th figure to be matched is defined as p1_max when big correlation;When time big correlation that will be obtained i-th it is to be matched
The location of figure is defined as p1_sec.If the absolute value of p1_max | p1_max | more than the effective range R_ of offset
Limit then abandons i-th figure to be matched;If p1_max is 0 and p1_sec is+R or-R, following step is executed
S11B.Wherein ,+R indicates that i-th figure to be matched moves right R pixel;- R indicates that i-th figure to be matched is moved to the left R picture
Element, R > R_limit;R is i-th figure to be matched for the first time with punctual maximum moving range.In this step, it is specified that i-th
The effective range R_limit of the offset of figure to be matched is in order to which i-th figure and reference map to be matched are carried out first time registration
When, if the absolute value of p1_max | p1_max | the effective range R_limit of the map migration to be matched more than i-th just rejects it
Fall, is no longer participate in the registration again that next and reference map carries out.Computer then looks for next figure and reference map to be matched
It is registrated, is conducive to improve registration efficiency in this way.But if i-th figure to be matched is being registrated with reference map in first time
P1_max is 0 in the process and p1_sec is+R or-R, then illustrates that i-th figure to be matched is not above the effective range of offset
R_limit, then this figure to be matched, which is given, retains, and carries out the movement of (i.e. step S11B) in next step.
Step S11B: defining p=(p1_max+p1_sec)/2, by i-th figure to be matched to be displaced p translation (if p is
Just, then to right translation;P is negative, then to left) and be registrated for second of reference map progress, it obtains the i-th width and translates figure to be matched,
I-th width is translated into figure and reference map to be matched and carries out second of registration, when by obtaining maximum correlation in second of registration process
I-th width translates the location of figure to be matched and is defined as p2_max, obtains the i-th width when secondary big correlation and translates locating for figure to be matched
Position be defined as p2_sec.By the absolute value of p2_max | p2_max | it is compared with R_limit: if | p2_max | it is more than offset
Effective range R_limit, then abandon the i-th width and translate figure to be matched;If p2_max is 0, S11C is thened follow the steps.Therefore, exist
In this step, the i-th width translates figure to be matched and may be removed, and may give and retain, to carry out the operation of next step.It needs
Illustrate, the i-th width translate figure to be matched be by i-th figure to be matched in step S11A be displaced after P translation and reference map into
It is obtained after second of registration of row, it and i-th figure to be matched are not same width figures.In fact, any figure to be matched is only
Will be after translation, image when obtained image all and does not translate is not identical, so can change in address.It needs
It is bright, it is carried out in second and second each time later registration the i-th width to be translated to figure and reference map to be matched, matching will definitely
It will not be+R or-R that the i-th width translates the location of figure to be matched pk_sec when the secondary big correlation arrived again, can only ratio+R
Or-R is small.
Step S11C: if | p2_max-p2_sec | < translation precision, then it is assumed that the i-th width translates the translation precision of figure to be matched
It meets the requirements, is transferred to step S11D;Otherwise, return step S11B is repeatedly recycled between step S11B and step S11C,
The i-th width is translated into figure to be matched and the reference map is repeatedly registrated, until translation precision is met the requirements, subsequently into step
Rapid S11D.Translation precision described in present patent application is preferably the distance of 1 pixel.
Step S11D: i-th width is translated into figure to be matched and the reference map carries out k times obtaining maximal correlation after being registrated
Property when described in the i-th width translate figure to be matched where position be defined as pk_max, i-th width is translated into figure to be matched and institute
It states reference map and carries out vertical registration, obtain the registration OCT figure.
For the p1_max and p1_sec mentioned in step S11A, find out as follows:
Step S11A1: by i-th figure to be matched in initial position and reference map carry out vertical registration, initially matched
Quasi- result align0;Go out the initial relevance value0 of initial registration result align0 and reference map according to cost function calculation;
Wherein, the initial position the location of refers to when i-th figure to be matched does not translate.Define maximum correlation
Value_max=initial relevance value0 then defines it since i-th figure to be matched does not move in initial position
Initial maximum position p_max=0.
Step S11A2: it by i-th figure to be matched to R pixel of right translation, obtains the i-th width and moves to right figure to be matched, by i-th
Width moves to right figure and reference map to be matched and carries out vertical registration, obtains the first registration result align1;Gone out according to cost function calculation
The first correlation value1 of first registration result align1 and reference map;
Step S11A3: by i-th figure to be matched to-R pixels of left, obtaining the i-th width and move to left figure to be matched,
I-th width is moved to left into figure to be matched and the reference map carries out vertical registration, obtains the second registration result align2;According to generation
Valence function calculates the second correlation value2 of the second registration result align2 and the reference map;
Step S11A4: compare initial relevance value0, the first correlation value1, the second correlation value2 it is big
It is small, maximum value is assigned to maximum correlation value_max, second largest value is assigned to time big correlation value_sec, and will with most
The big corresponding translational displacement of correlation value_max is assigned to p1_max, will translate accordingly with secondary big correlation value_sec
Displacement is assigned to p1_sec.
Wherein, the sequence of step S11A2 and step S11A3 can exchange.
In step S11A1- step S11A3, cost function is preferably value=sum (align.*ref), wherein
Align refers to that any one figure and reference map to be matched carry out the result obtained after vertical registration;Ref indicates reference map.
Finally, Fig. 1 is returned again to, according to step S2: m width registration OCT image is synthesized the clear OCT image of a panel height (see figure
4)。
The method announced through the invention is carried out with punctual when by any one figure and reference map to be matched: in Vertical Square
To frequency domain registration method is used, i.e., each of each column of figure to be matched and reference map is listed in vertical direction and matched respectively
It is quasi-;In the horizontal direction, it is iterated using dichotomy, reduces cycle-index, improve the speed and precision of algorithm.
With reference to Fig. 6, a kind of a wide range of high definition rapid registering device of ophthalmology OCT image is also disclosed in the present invention, comprising:
OCT image registration module, is used for: it is strongest to choose signal-to-noise ratio from the n width OCT image that ophthalmology OCT device scanning human eye obtains
One width is registrated respectively with the reference map using remaining n-1 width OCT image as figure to be matched as reference map, obtains m width
It is registrated OCT image;High definition OCT image synthesis module, is used for: m width registration OCT image being synthesized the clear OCT of a panel height and is schemed
Picture.
Specifically, in front by the method for any i-th of n-1 width OCT image figure and reference map to be matched registration
It described, was not repeated herein.
The a wide range of high definition rapid registering device of ophthalmology OCT image announced through the invention, when by any one to
Matching figure and reference map are carried out with punctual: in vertical direction using frequency domain registration method, i.e., by each column and base of figure to be matched
Each vertical direction that is listed in of quasi- figure is registrated respectively;In the horizontal direction, it is iterated using dichotomy, reduces circulation
Number improves the speed and precision of algorithm.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.