CN105405099A - Underwater image super-resolution reconstruction method based on point spread function - Google Patents

Abstract

The invention discloses an underwater image super-resolution reconstruction method based on a point spread function, and resolution of an underwater image can be improved. The method comprises the following steps: firstly obtaining continuous-frame laser range gating images, taking a first frame as a reference frame, taking the reference frame as a recovery image, and taking a second frame image as an observation image; performing blind deconvolution processing aiming at the recovery image and the observation image, obtaining a point spread function of a current frame, and then updating the recovery image by utilizing the point spread function; and, in a similar fashion, updating a next frame of the image to the observation image in turn, performing the blind deconvolution processing on the updated recovery image and the observation image to obtain the point spread function and continuously update the recovery image until the last recover image is obtained, performing the blind deconvolution processing on the recovery image and the reference frame, and obtaining a final point spread function for replacing a degradation model in a POCS algorithm. The POCS algorithm is improved, and image resolution is improved.

Description

A kind of underwater picture super resolution ratio reconstruction method based on point spread function

Technical field

The present invention relates to underwater picture super-resolution rebuilding field, particularly relate to a kind of underwater picture super resolution ratio reconstruction method based on point spread function.

Background technology

In recent years, along with the mankind's deepening continuously to marine resources development, underweater vision has become the indispensable ingredient in the fields such as sub-sea floor targets detection, ocean geography engineering and ocean military affairs.Under water in environment, because aqueous medium causes light intensity attenuation, front and back to factor impacts such as scattered noises, cause image quality poor, details is fuzzy, is difficult to obtain accurate information.Therefore can be processed image by the method for super-resolution rebuilding, to obtain the higher image of more clear resolution.

Underwater laser range-gated imaging technique is one of Underwater Imaging technology of comparatively maturation instantly, has good effect for the image quality improving underwater imaging system.The principle of technique of laser range gated imaging as shown in Figure 2.The optical window of water body is roughly at 480 ~ 550nm, so underground distance gated imaging mode is often using Nd:YAG solid pulse laser as radiation source, and T ₀moment Laser emission incident light illumination target object, because the storbing gate on ICCD detector is closed, a large amount of rear orientation lights of water body can not enter detector, detection complete object required time Δ T, T ₁moment target reflecting light arrives detector, and storbing gate opens the time of Δ T, and ICCD is to detection of a target complete imaging, and maskable falls most water body rear orientation light thus, increases the operating distance of Underwater Imaging system.The application of underground distance gated imaging technology, can improve 3 ~ 5 times by the detection range of Underwater Imaging equipment.

Projections onto convex sets (ProjectionsOntoConvexSets, POCS) has consequence in the algorithm of numerous super-resolution rebuilding.In POCS algorithm, degenrate function is important priori.But under different waters, different illumination conditions, the degree of degeneration of underwater picture is widely different, and degenrate function has uncertainty.At present for the research of the still rarely seen systematic quantification of degradation model of underwater picture.

Summary of the invention

In view of this, the invention provides a kind of underwater picture super resolution ratio reconstruction method based on point spread function, the degenrate function of sequential frame image under water can be solved, i.e. point spread function, apply in POCS algorithm, thus improve the resolution of underwater picture.

In order to achieve the above object, technical scheme of the present invention is for comprising following step:

S1: adopt the method for technique of laser range gated to take one group of sequential frame image to target, using the first frame in sequential frame image as with reference to frame, with reference to frame as restored image, using the second two field picture as observed image.

S2: carry out blind deconvolution process for restored image and observed image, obtains the point spread function of present frame, then uses the point spread function of present frame to upgrade restored image.

S3: observed image is updated to the 3rd two field picture, carries out blind deconvolution process for the restored image after renewal and observed image, tries to achieve the point spread function of present frame, then use the point spread function of present frame to upgrade restored image.

S4: the rest may be inferred, finally obtains with last frame image as the restored image after renewal during observed image; Restored image after this renewal and reference frame are done the process of blind deconvolution, try to achieve final point spread function.

S5: brought into by final point spread function in projections onto convex sets POCS algorithm, replaces the degradation model of image in projections onto convex sets POCS algorithm, improves POCS algorithm, improve the resolution of underwater picture.

Further, the process of blind deconvolution comprises: the point spread function solving t two field picture and restored image $f_t={\mathrm{argmin}}_{f_t\≥0}\left|\right|y_t-F_t{x}_{t-1}|{|}^2;$ Wherein restored image x _titerative formula be:

F in formula _trepresent the point spread function that t two field picture and restored image solve; F _tfor only and f _trelevant function, f _t* x _t=F _tx _t, * represents convolution algorithm; In ⊙ representing matrix, corresponding element is multiplied, || || for norm solves symbol; T represents t frame, the value 2 ~ N of t; x _tfor restored image, x _tinitial value x ₁for reference frame; y _tfor observed image, y _tinitial value y ₂it is the second two field picture; By x ₁, y ₂bring in formula 1, solve and obtain f ₂; By the f tried to achieve ₂and x ₁, y ₂bring in formula 2 and obtain x ₂, by that analogy, obtain restored image x in turn ₂~ x _n.

Beneficial effect:

A kind of underwater picture super resolution ratio reconstruction method based on point spread function provided by the present invention, can by processing sequential frame image under water, the method of loop iteration is used to estimate sequential frame image point spread function under water, thus obtained point spread function can reflect the image degradation degree of successive frame more truly, this point spread function is applied in POCS algorithm, POCS algorithm can be improved, improve the resolution of underwater picture.

Accompanying drawing explanation

The method flow diagram of Fig. 1 to be the present invention be embodiment;

Fig. 2 is underwater laser Range-gated Imager system diagram.

Embodiment

To develop simultaneously embodiment below in conjunction with accompanying drawing, describe the present invention.

Fig. 1 is this method process flow diagram:

Step S1: adopt the method for technique of laser range gated to take one group of sequential frame image to target, using the first frame in sequential frame image as with reference to frame, with reference to frame as restored image, using the second two field picture as observed image.

Step S2: using the second two field picture as observed image, carries out blind deconvolution process for restored image and observed image, obtains the point spread function of present frame, then uses the point spread function of present frame to upgrade restored image.

Blind deconvolution adopts following formula 1 and formula 2 to calculate;

Formula 1: $f_t={\mathrm{argmin}}_{f_t\≥0}\left|\right|y_t-F_t{x}_{t-1}|{|}^2;$

Formula 1:

F in formula _trepresent the point spread function that t two field picture and restored image solve; F _tfor only and f _trelevant function, f _t* x _t=F _tx _t, * represents convolution algorithm; In ⊙ representing matrix, corresponding element is multiplied, || || for norm solves symbol; In ⊙ representing matrix, corresponding element is multiplied; T represents t frame, the value 2 ~ N of t; x _tfor restored image, x _tinitial value x ₁for reference frame; y _tfor observed image, y _tinitial value y ₂it is the second two field picture; By x ₁, y ₂bring in formula 1, use LBFGS optimized algorithm to solve and obtain f ₂; By the f tried to achieve ₂and x ₁, y ₂bring in formula 2 and obtain x ₂, by that analogy, obtain restored image x in turn ₂~ x _n.

Step S3: using the 3rd two field picture as observed image, carries out blind deconvolution process for the restored image after renewal and observed image, tries to achieve the point spread function of present frame, then use the point spread function of present frame to upgrade restored image;

Step S4: the rest may be inferred, finally obtains with last frame image as the restored image after renewal during observed image; Restored image after this renewal and reference frame are done the process of blind deconvolution, try to achieve final point spread function f ₁.

Step S5: brought into by final point spread function in projections onto convex sets POCS algorithm, replaces the degradation model of image in projections onto convex sets POCS algorithm, and then improves POCS algorithm, improve the resolution of image.Wherein POCS algorithm generally comprises following concrete steps:

Step S501: bicubic linear interpolation is done to reference frame and makes it reach the high resolving power of expectation, as initial estimation.

Step S502: for each estimation pixel accurately in observation sequence, finds this pixel-map to the location of pixels in the high-definition picture of current estimation in motion vector field, and the pixel under PSF (w) effect.W is herein the degradation model of image in POCS algorithm, and this method uses required final point spread function f above at this place ₁carry out alternative w.

Step S503: analog image acquisition process, obtains the estimated value of this pixel, and calculates the residual error between actual pixel value and estimated value, if residual error exceeds the residual error limit of setting, then needs to revise estimated value, within making residual error be reduced to residual error limit.

Step S504: revise the current high resolving power of iteration and estimate, until reach acceptable scope, obtain and rebuild image.

To sum up, these are only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.Within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (2)

1. based on a underwater picture super resolution ratio reconstruction method for point spread function, it is characterized in that, comprise following step:

S1: adopt the method for technique of laser range gated to take one group of sequential frame image to target, using the first frame in sequential frame image as with reference to frame, with reference to frame as restored image, using the second two field picture as observed image;

S2: carry out blind deconvolution process for restored image and observed image, obtains the point spread function of present frame, then uses the point spread function of present frame to upgrade restored image;

S3: observed image is updated to the 3rd two field picture, carries out blind deconvolution process for the restored image after renewal and observed image, tries to achieve the point spread function of present frame, then use the point spread function of present frame to upgrade restored image;

S4: the rest may be inferred, finally obtains with last frame image as the restored image after renewal during observed image; Restored image after this renewal and described reference frame are done the process of blind deconvolution, try to achieve final point spread function;

S5: substituted into by described final point spread function in projections onto convex sets POCS algorithm, replaces the degradation model of image in projections onto convex sets POCS algorithm, obtains and rebuild image.

2. underwater picture super resolution ratio reconstruction method according to claim 1, is characterized in that, the process of described blind deconvolution comprises:

Solve the point spread function of t frame $f_t=\arg {\min }_{f_t\≥0}\left|\right|y_t-F_t{x}_{t-1}|{|}^2;$

Wherein restored image x _titerative formula be:

F in formula _trepresent the point spread function that t two field picture and restored image solve; F _tfor only and f _trelevant function, f _t* x _t=F _tx _t, * represents convolution algorithm; In ⊙ representing matrix, corresponding element is multiplied, || || for norm solves symbol; T represents t frame, the value 2 ~ N of t; x _tfor restored image, x _tinitial value x ₁for reference frame; y _tfor observed image, y _tinitial value y ₂it is the second two field picture; By x ₁, y ₂bring in formula 1, solve and obtain f ₂; By the f tried to achieve ₂and x ₁, y ₂bring in formula 2 and obtain x ₂, by that analogy, obtain restored image x in turn ₂~ x _n.