CN104408696A - Image splicing method aiming at side-scan sonar imaging features - Google Patents
Image splicing method aiming at side-scan sonar imaging features Download PDFInfo
- Publication number
- CN104408696A CN104408696A CN201410468470.0A CN201410468470A CN104408696A CN 104408696 A CN104408696 A CN 104408696A CN 201410468470 A CN201410468470 A CN 201410468470A CN 104408696 A CN104408696 A CN 104408696A
- Authority
- CN
- China
- Prior art keywords
- image
- scale
- harris
- metric space
- angle point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Abstract
The invention discloses an image splicing method aiming at side-scan sonar imaging features. The method aims at solving the problems of low resolution, serious noise pollution, ghosting, Doppler effects and the like in the side-scan sonar imaging. According to the method, the study is carried out on the basis of a splicing problem of image feature points; the scale-space-theory-fusing Harris corner detection and SIFT feature description on corners are mainly studied; and further, the image registration is realized. In an image fusion stage, after the image conversion relationship corresponding, affine transformation and weighted average methods are sequentially adopted for completing the image fusion in the space and pixels, and a complete spliced image is obtained.
Description
Technical field
What the present invention relates to is a kind of image mosaic technology, particularly be a kind of image split-joint method for side-scan sonar imaging characteristics.
Background technology
Sonar detection technology is the key means in marine charting field.Side-scan sonar transmits sound waves on sea bed and accepts reflective sound wave successively again, utilizes the principle that different submarine geologies reflects different intensity of acoustic wave to draw out seafloor topography.But the band of the just description seafloor topography that side-scan sonar obtains, if will obtain complete bottom relief map, needs each adjacent band to carry out splicing, draws out the sonar spliced map in whole region.Namely the content that this paper will be studied is the splicing of side-scanning sonar image, is sequentially spliced mutually by the sonar image in measurement range, draws out the sonar spliced map in gamut, thus obtains topomap under complete sea.
The focus of current scholar's research that to take images match as the image mosaic technology of key be.The Richard professor of Microsoft Research proposed a kind of based drive Panorama Mosaic model in 1996, adopt L-M iterative nonlinear Method for minimization, by the geometric transform relation such as translation, rotation, convergent-divergent between this model determination image, thus determine the registration transformation relation between image, the method effect is better, fast convergence rate, the to be spliced image of great majority through space geometry conversion can be processed, so be acknowledged as the classic algorithm in image mosaic field, therefore Richard also becomes the founder in image mosaic field.
Summary of the invention
A kind of image split-joint method for side-scan sonar imaging characteristics of the present invention, object is the problem such as resolution in side-scan sonar imaging is low, noise pollution serious, ghost image, Doppler effect in order to solve.
The object of the present invention is achieved like this: a kind of step of the image split-joint method for side-scan sonar imaging characteristics is such:
1, Image semantic classification, and unique point is proposed.For the shortcoming of Harris angle point to dimensional variation sensitivity, propose the multiple dimensioned Harris Corner Detection Algorithm merged with Scale-space theory, through the detection of the template of two groups of change of scale, the Harris angle point detected more accurately and have scale invariant feature, also lays the first stone for hereinafter SIFT describes simultaneously.
2, at image characteristic point detection-phase, on the basis of dimensional variation sensitivity, the improvement of multiple scale detecting has been carried out to Harris at Harris operator, made it adapt to multiple dimensioned change.
3, images match.From utilizing gradation of image information and characteristics of image two aspect to carry out the introduction of images match, describe based on pixel difference quadratic sum (SSD) and the matching process based on cross-correlation (CC), paper adopts the matching process based on characteristics of image simultaneously.On the basis of the Harris angle point extracted before, SIFT feature is utilized to re-start description to angle point, generate SIFT feature vector descriptor, then utilize Euclidean distance as the basis for estimation of coupling, thus determine the transformation relation of benchmark image and image to be matched.
4, image co-registration.After matching relationship correspondence, determine the spatial transform relation between image to be fused and benchmark image by affine change; Determine that two image overlapping region pixel values must be determined gradually being gone out method by being fade-in in method of weighted mean, and the pixel of overlapping region is seamlessly transitted, without obvious gap, finally complete the fusion of image.
The present invention also comprises:
The Harris Corner Detection Algorithm improved is achieved in that building different metric spaces is exactly to detect that image has the unique point of scale invariability, and paper adopts Gaussian function to set up metric space:
G(
,
)=
,
If I (
) be the pixel of input picture, G (
,
) be the Gaussian function of changeable scale, the metric space of image be defined as L (
,
), expression formula is as follows:
L(
,
)=I(
)^G(
,
),
Wherein, ^ represents convolution algorithm, represents metric space parameter, greatly, then and the roughly feature of metric space Description Image; Little, then the minutia of metric space Description Image, along with change, just set up the metric space that one group of yardstick is different.First select herein
gaussian function when=0.4 carries out Gaussian convolution to former figure, and it can be used as the first width detected image of first group of metric space, then increases until generate the 5th width detected image successively.When establishment second group of metric space, that sub-picture of centre getting one group of metric space carry out 1/2nd samplings (by this figure length and be widely all reduced to original 1/2nd) as the piece image of second group of metric space, then carry out same aforesaid operations and equally generate 5 width images, obtain second group of metric space.
The yardstick level in two groups of different scale spaces of above-mentioned foundation carries out Harris Corner Detection respectively, its step show greatly before Harris Corner Detection similar, just launch respectively in different scale space, concrete steps are as follows:
Calculate the direction gradient under different scale and Gaussian convolution, obtain the auto-correlation Metzler matrix under different scale:
M(
)=
,
Then calculate on metric space Harris angle point response function R (
):
R(
)=det(M(
)
k×
(M(
)),
In formula, k is weights coefficient, usually gets 0.04-0.06, and tr is rank of matrix, when R (
) be greater than set threshold value and around in 3 × 3 fields for maximal value time, then determine that this point is angle point.
In two groups of metric spaces, along small scale direction to the pseudo-angle point of large scale direction filtering.Suppose a Harris angle point to be detected when yardstick; A Harris angle point detected when yardstick, until out to out, judge whether Harris angle point exists at the same coordinates regional of each yardstick (3 × 3 template scope) of this group metric space, if exist, this angle point is set to unique point; If do not exist, then illustrate that this point may be isolated pixel point, noise or edge, reject this angle point.
Accompanying drawing explanation
Fig. 1 is the process flow diagram of image mosaic system.
Embodiment
Below in conjunction with accompanying drawing citing, the present invention is described in more detail:
Embodiment 1
Composition graphs 1, Fig. 1 is the process flow diagram of image mosaic system.The present invention studies at the Bonding Problem based on image characteristic point, mainly have studied the Harris Corner Detection of fusion Scale-space theory and describes the SIFT feature of angle point and then realize image registration.In the image co-registration stage, after inter-image transformations relation correspondence, successively adopt affined transformation and method of weighted mean to complete the fusion of image on space and pixel, obtain a complete stitching image.
(1) at image pre-processing phase, for the shortcoming of Harris angle point to dimensional variation sensitivity, propose the multiple dimensioned Harris Corner Detection Algorithm merged with Scale-space theory, through the detection of the template of two groups of change of scale, the Harris angle point detected more accurately and have scale invariant feature, also lays the first stone for hereinafter SIFT describes simultaneously.
(2) at image characteristic point detection-phase, on the basis of dimensional variation sensitivity, the improvement of multiple scale detecting has been carried out to Harris at Harris operator, made it adapt to multiple dimensioned change.
(3) in the images match stage, from utilizing gradation of image information and characteristics of image two aspect to carry out the introduction of images match, describe based on pixel difference quadratic sum (SSD) and the matching process based on cross-correlation (CC), paper adopts the matching process based on characteristics of image simultaneously.On the basis of the Harris angle point extracted before, SIFT feature is utilized to re-start description to angle point, generate SIFT feature vector descriptor, then utilize Euclidean distance as the basis for estimation of coupling, thus determine the transformation relation of benchmark image and image to be matched.
(4) in image co-registration, after matching relationship correspondence, the spatial transform relation between image to be fused and benchmark image is determined by affine change; Determine that two image overlapping region pixel values must be determined gradually being gone out method by being fade-in in method of weighted mean, and the pixel of overlapping region is seamlessly transitted, without obvious gap, finally complete the fusion of image.
Embodiment 2
A kind of image split-joint method for side-scan sonar imaging characteristics, the Harris Corner Detection Algorithm improved is achieved in that building different metric spaces is exactly to detect that image has the unique point of scale invariability, adopts Gaussian function to set up metric space:
G(
,
)=
,
If I (
) be the pixel of input picture, G (
,
) be the Gaussian function of changeable scale, the metric space of image be defined as L (
,
), expression formula is as follows:
L(
,
)=I(
)^G(
,
),
Wherein, ^ represents convolution algorithm, represents metric space parameter, greatly, then and the roughly feature of metric space Description Image; Little, then the minutia of metric space Description Image, along with change, just set up the metric space that one group of yardstick is different; First select
gaussian function when=0.4 carries out Gaussian convolution to former figure, and it can be used as the first width detected image of first group of metric space, then increase successively until generate the 5th width detected image, when establishment second group of metric space, that sub-picture of centre getting one group of metric space carry out 1/2nd samplings (by this figure length and be widely all reduced to original 1/2nd) as the piece image of second group of metric space, then carry out same aforesaid operations and equally generate 5 width images, obtain second group of metric space;
The yardstick level in two groups of different scale spaces of above-mentioned foundation carries out Harris Corner Detection respectively, its step show greatly before Harris Corner Detection similar, just launch respectively in different scale space, concrete steps are as follows:
1, calculate the direction gradient under different scale and Gaussian convolution, obtain the auto-correlation Metzler matrix under different scale:
M(
)=
,
Then calculate on metric space Harris angle point response function R (
):
R(
)=det(M(
)
k×
(M(
)),
In formula, k is weights coefficient, usually gets 0.04-0.06, and tr is rank of matrix, when R (
) be greater than set threshold value and around in 3 × 3 fields for maximal value time, then determine that this point is angle point;
2, in two groups of metric spaces, along small scale direction to the pseudo-angle point of large scale direction filtering; Suppose at yardstick
in time, detects
individual Harris angle point; At yardstick
in time, detects
individual Harris angle point, until out to out
, judge whether Harris angle point exists at the same coordinates regional of each yardstick (3 × 3 template scope) of this group metric space, if exist, this angle point is set to unique point; If do not exist, then illustrate that this point may be isolated pixel point, noise or edge, reject this angle point.
Claims (2)
1. for an image split-joint method for side-scan sonar imaging characteristics, it is characterized in that: a kind of step of the image split-joint method for side-scan sonar imaging characteristics is:
1, Image semantic classification, and unique point is proposed: for the shortcoming of Harris angle point to dimensional variation sensitivity, the multiple dimensioned Harris Corner Detection Algorithm merged with Scale-space theory is proposed, through the detection of the template of two groups of change of scale, detect that Harris angle point more accurately and have scale invariant feature;
2, image characteristic point detects: at Harris operator to the improvement basis of dimensional variation sensitivity being carried out multiple scale detecting to Harris;
3, images match: from utilizing gradation of image information and characteristics of image two aspect to carry out the introduction of images match, introduces based on pixel difference quadratic sum and the matching process based on cross-correlation, adopts the matching process based on characteristics of image simultaneously; On the basis of the Harris angle point extracted before, SIFT feature is utilized to re-start description to angle point, generate SIFT feature vector descriptor, then utilize Euclidean distance as the basis for estimation of coupling, thus determine the transformation relation of benchmark image and image to be matched;
4, image co-registration: after matching relationship correspondence, determines the spatial transform relation between image to be fused and benchmark image by affine change; Determine that two image overlapping region pixel values must be determined gradually being gone out method by being fade-in in method of weighted mean, and the pixel of overlapping region is seamlessly transitted, complete the fusion of image.
2. a kind of image split-joint method for side-scan sonar imaging characteristics according to claim 1, it is characterized in that: the Harris Corner Detection Algorithm of improvement is achieved in that building different metric spaces is exactly to detect that image has the unique point of scale invariability, adopts Gaussian function to set up metric space:
G(
,
)=
,
If I (
) be the pixel of input picture, G (
,
) be the Gaussian function of changeable scale, the metric space of image be defined as L (
,
), expression formula is as follows:
L(
,
)=I(
)^G(
,
),
Wherein, ^ represents convolution algorithm,
represent metric space parameter,
greatly, then the roughly feature of metric space Description Image;
little, then the minutia of metric space Description Image, along with
change, just set up the metric space that one group of yardstick is different; First select
gaussian function when=0.4 carries out Gaussian convolution to former figure, and it can be used as the first width detected image of first group of metric space, then increases successively
until generate the 5th width detected image, when establishment second group of metric space, that sub-picture of centre getting one group of metric space carry out 1/2nd samplings (by this figure length and be widely all reduced to original 1/2nd) as the piece image of second group of metric space, then carry out same aforesaid operations and equally generate 5 width images, obtain second group of metric space;
The yardstick level in two groups of different scale spaces of above-mentioned foundation carries out Harris Corner Detection respectively, its step show greatly before Harris Corner Detection similar, just launch respectively in different scale space, concrete steps are as follows:
1, calculate the direction gradient under different scale and Gaussian convolution, obtain the auto-correlation Metzler matrix under different scale:
M(
)=
,
Then at metric space
upper calculating Harris angle point response function R (
):
R(
)=det(M(
)
k×
(M(
)),
In formula, k is weights coefficient, usually gets 0.04-0.06; Tr is rank of matrix, when R (
) be greater than set threshold value and around in 3 × 3 fields for maximal value time, then determine that this point is angle point;
2, in two groups of metric spaces, along small scale direction to the pseudo-angle point of large scale direction filtering; Suppose at yardstick
in time, detects
individual Harris angle point; At yardstick
in time, detects
individual Harris angle point, until out to out
, judge whether Harris angle point exists at the same coordinates regional of each yardstick (3 × 3 template scope) of this group metric space, if exist, this angle point is set to unique point; If do not exist, then illustrate that this point may be isolated pixel point, noise or edge, reject this angle point.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410468470.0A CN104408696A (en) | 2014-09-16 | 2014-09-16 | Image splicing method aiming at side-scan sonar imaging features |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410468470.0A CN104408696A (en) | 2014-09-16 | 2014-09-16 | Image splicing method aiming at side-scan sonar imaging features |
Publications (1)
Publication Number | Publication Date |
---|---|
CN104408696A true CN104408696A (en) | 2015-03-11 |
Family
ID=52646325
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410468470.0A Pending CN104408696A (en) | 2014-09-16 | 2014-09-16 | Image splicing method aiming at side-scan sonar imaging features |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104408696A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104392428A (en) * | 2014-12-10 | 2015-03-04 | 黑龙江真美广播通讯器材有限公司 | Splicing system for side-scan sonar images |
CN105447864A (en) * | 2015-11-20 | 2016-03-30 | 小米科技有限责任公司 | Image processing method, device and terminal |
CN105718852A (en) * | 2015-11-24 | 2016-06-29 | 深圳芯启航科技有限公司 | Fingerprint image processing method and apparatus |
CN105869193A (en) * | 2016-03-31 | 2016-08-17 | 哈尔滨工程大学 | UUV-based side-scan sonar image auxiliary interpreting method |
CN105891836A (en) * | 2016-04-01 | 2016-08-24 | 中国船舶重工集团公司第七〇五研究所 | Secondary echo suppression and geomorphologic map fusion method based on sounding side-scan sonar |
CN106596590A (en) * | 2016-12-24 | 2017-04-26 | 大连日佳电子有限公司 | Tray IC detection method |
CN109031319A (en) * | 2018-07-26 | 2018-12-18 | 江苏科技大学 | A kind of side-scanning sonar image splicing system and its method |
CN110781924A (en) * | 2019-09-29 | 2020-02-11 | 哈尔滨工程大学 | Side-scan sonar image feature extraction method based on full convolution neural network |
CN112164051A (en) * | 2020-09-29 | 2021-01-01 | 中国船舶重工集团公司第七二四研究所 | Radar antenna area array liquid leakage detection device and method based on image analysis |
CN113096171A (en) * | 2021-03-01 | 2021-07-09 | 中国人民解放军海军大连舰艇学院 | Multi-scale iterative self-adaptive registration method for multi-beam and side-scan sonar images |
-
2014
- 2014-09-16 CN CN201410468470.0A patent/CN104408696A/en active Pending
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104392428A (en) * | 2014-12-10 | 2015-03-04 | 黑龙江真美广播通讯器材有限公司 | Splicing system for side-scan sonar images |
CN105447864A (en) * | 2015-11-20 | 2016-03-30 | 小米科技有限责任公司 | Image processing method, device and terminal |
CN105447864B (en) * | 2015-11-20 | 2018-07-27 | 小米科技有限责任公司 | Processing method, device and the terminal of image |
CN105718852A (en) * | 2015-11-24 | 2016-06-29 | 深圳芯启航科技有限公司 | Fingerprint image processing method and apparatus |
CN105869193A (en) * | 2016-03-31 | 2016-08-17 | 哈尔滨工程大学 | UUV-based side-scan sonar image auxiliary interpreting method |
CN105869193B (en) * | 2016-03-31 | 2018-09-28 | 哈尔滨工程大学 | Side-scanning sonar image auxiliary interpretation method based on UUV |
CN105891836A (en) * | 2016-04-01 | 2016-08-24 | 中国船舶重工集团公司第七〇五研究所 | Secondary echo suppression and geomorphologic map fusion method based on sounding side-scan sonar |
CN106596590A (en) * | 2016-12-24 | 2017-04-26 | 大连日佳电子有限公司 | Tray IC detection method |
CN109031319A (en) * | 2018-07-26 | 2018-12-18 | 江苏科技大学 | A kind of side-scanning sonar image splicing system and its method |
CN110781924A (en) * | 2019-09-29 | 2020-02-11 | 哈尔滨工程大学 | Side-scan sonar image feature extraction method based on full convolution neural network |
CN112164051A (en) * | 2020-09-29 | 2021-01-01 | 中国船舶重工集团公司第七二四研究所 | Radar antenna area array liquid leakage detection device and method based on image analysis |
CN113096171A (en) * | 2021-03-01 | 2021-07-09 | 中国人民解放军海军大连舰艇学院 | Multi-scale iterative self-adaptive registration method for multi-beam and side-scan sonar images |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104408696A (en) | Image splicing method aiming at side-scan sonar imaging features | |
CN101901481B (en) | Image mosaic method | |
CN107103277B (en) | Gait recognition method based on depth camera and 3D convolutional neural network | |
CN101789122B (en) | Method and system for correcting distorted document image | |
CN105182350B (en) | A kind of multibeam sonar object detection method of application signature tracking | |
CN107111598B (en) | Optical flow imaging system and method using ultrasound depth sensing | |
CN104200461B (en) | The remote sensing image registration method of block and sift features is selected based on mutual information image | |
KR100986809B1 (en) | The Method of Automatic Geometric Correction for Multi-resolution Satellite Images using Scale Invariant Feature Transform | |
CN102075686B (en) | Robust real-time on-line camera tracking method | |
JP6976733B2 (en) | Image processing equipment, image processing methods, and programs | |
CN104392428A (en) | Splicing system for side-scan sonar images | |
CN105654423B (en) | Remote sensing image registration method based on region | |
CN111160291B (en) | Human eye detection method based on depth information and CNN | |
CN103279952A (en) | Target tracking method and device | |
CN104156965A (en) | Automatic fast mine monitoring image stitching method | |
Cheng et al. | Omnidirectional depth extension networks | |
CN109376641A (en) | A kind of moving vehicle detection method based on unmanned plane video | |
Xia et al. | Globally consistent alignment for planar mosaicking via topology analysis | |
CN103793894A (en) | Cloud model cellular automata corner detection-based substation remote viewing image splicing method | |
Redzuwan et al. | Affine versus projective transformation for SIFT and RANSAC image matching methods | |
CN109166075A (en) | One kind being directed to small overlapping region image split-joint method | |
Chen et al. | Image stitching algorithm research based on OpenCV | |
CN108961182B (en) | Vertical direction vanishing point detection method and video correction method for video image | |
Moroni et al. | Underwater scene understanding by optical and acoustic data integration | |
Wang et al. | A real-time correction and stitching algorithm for underwater fisheye images |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20150311 |