CN104486631B - A kind of remote sensing image compression method based on human eye vision Yu adaptive scanning - Google Patents

Abstract

A kind of remote sensing image compression method based on human eye vision Yu adaptive scanning, belongs to remote sensing images online browse technical field.The present invention is on the basis of hardly overhead bit is increased, the visual effect for rebuilding remote sensing images can be improved, although solving existing Normal squeezing method can provide the preferable reconstruction image of quality under square errors sense, the unsatisfactory problem of reconstruction image visual effect.The technical scheme is that：First mask is weighted using importance to weight changing image；The scanning sequency of then each sub-belt energy after calculating weighting, and the descending arrangement determination intersubband according to energy；Finally, to the scanning inside weighting subband, the characteristic according to subband determines scan method.The present invention effectively improves the visual quality for rebuilding remote sensing images, meets the demand of remote sensing images online browse growing at present.Online browse of the present invention suitable for remote sensing images.

Description

A kind of remote sensing image compression method based on human eye vision Yu adaptive scanning

Technical field

It is more particularly to a kind of based on human eye vision and adaptive scanning the present invention relates to a kind of remote sensing image compression method Remote sensing image compression method, belongs to remote sensing images online browse technical field.

Background technology

With the development of sensor technology, the spatial resolution and spectral resolution of remote sensing images have obtained greatly adding By force, this brings great convenience for the application of remote sensing images.On the other hand, abundant data are as generation with huge data volume Valency.Newest space satellite all produces High Dimensional Remote Sensing Data of the number in terms of T daily, and this brings for the storage and transmission of data Greatly challenge, such case can be eased by using some traditional compress techniques, referring to document EZW [1], SPIHT [2], SPECK [3], JPEG2000 [4], or some improvement to them, referring to document [30]~[36].Generally, these compressions Method is usually the measurement compression effectiveness under mean-square error criteria, that is to say, that under the same terms, using the teaching of the invention it is possible to provide smaller is equal The compression method of square error, it is considered to be more preferable compression method.However, for a width reconstruction image, less mean square error It is not meant to that the image is adapted to all of application.In fact, the evaluation to compression method should depend on corresponding application.With The popularization of remote sensing images, online browse of the substantial amounts of application with remote sensing images is relevant.Additionally, a current study hotspot, Digital earth, it is also desirable to which the preferable remote sensing images of some visual effects are covered to it.In this case, regarded with reference to human eye The compression method that feel system (HVS) is designed can more meet application demand.HVS is a system for complexity, and it has been demonstrated It is not consistent with mean-square error criteria, referring to document [5].Therefore, it is necessary to from the vision mechanism of human eye, to remote sensing figure The compression method of picture is studied.

The visually-perceptible of human eye is always the focus of some researchs, referring to document [5].Visually-perceptible is melted with coding method Close there is several methods that.One kind is the method based on discrete cosine transform (DCT), referring to document [6]-[8]；Another kind be based on from The method for dissipating wavelet transformation (DWT), referring to document [9]~[12].Some coding methods are to distinguish error (JND) mould with reference to minimum What type was designed, this kind method is can not to be detected by human eye this feature using some minor variations in image, to compression Method is designed, referring to document [13] [14].Additionally, some compression methods relevant with vision are carried out based on JPEG2000 Design, referring to document [16] [17].Recently, some coding methods for being based on HVS are designed from the angle of information theory, example Such as, document [15] proposes the coding method that a kind of view-based access control model is perceived, it is therefore an objective to retain second order yardstick in natural image not Become feature, the visual quality of reconstruction image is ensured by this method.Document 18] some new models relevant with HVS are established, And be deduced under specific occasion, the theoretical compression limit of virtually lossless can be reached.However, all these methods both for What natural image was designed, not in view of the peculiar property of remote sensing images.

For natural image, after wavelet transformation, the dense list that can obtain signal reaches, and this contributes to the volume for obtaining Code result.However, compared with natural image, remote sensing images have the property of uniqueness.Remote sensing images generally comprise substantial amounts of atural object Information, this causes that the details of remote sensing images is extremely enriched, such as geological information, the profile at edge and texture etc., even Small object. Therefore, for remote sensing images, good compression method should be able to retain more detailed information.

In recent years, some are suggested dedicated for the compression method of remote sensing images, referring to document [19]~[22].These pressures Compression method is compressed in terms of several to image, such as directional wavelet transform (OWT) or sparse expression.Because remote sensing images are usual It is that sensor is obtained in the way of pushing away and sweeping and size is very big, therefore, the compression method based on scanning is of concern.Document [23] a kind of method based on scanning is proposed, data is obtained in the way of pushing away and sweep with JPEG2000.However, the height of JPEG2000 Coding efficiency is to be designed as cost with complicated.Consultative committee for space data system (CCSDS) compresses on star, issue One standard, the CCSDS standards are a kind of methods based on scanning, but do not allow interactive decoder, and the wavelet decomposition number of plies It is fixed as 3.2009, the document such as V í lchez [24] were extended to CCSDS standards, it is supported the small echo of any number of plies Decompose, and allow the decoding of several forms.However, all these methods are all based on scanning constant, image is not accounted for Content.

2012, document [25] proposed a kind of newest method based on scanning, referred to as self adaptation binary-tree coding (binary tree coding adaptively, BTCA), the method is directed to the compression design of remote sensing images.The method Coding efficiency can be significantly improved.Although however, the method is relevant with picture material to a certain extent, setting up binary tree Before, the scanning to changing image still uses fixed scan mode.It is well known that different images have it is different interior Hold.In other words, to different remote sensing images, the distribution of significant coefficient is different.Therefore, from the angle of scanning, using solid The coding efficiency that fixed scanning sequency can not be obtained.

The content of the invention

The purpose of the present invention is to propose to a kind of remote sensing image compression method based on human eye vision Yu adaptive scanning, several On the basis of not increasing overhead bit, it is possible to increase rebuild the visual effect of remote sensing images, solve existing Normal squeezing side Although method can provide the preferable reconstruction image of quality under square errors sense, reconstruction image visual effect is unsatisfactory to ask Topic.

The present invention is for the solution technical scheme that is used of above-mentioned technical problem：

A kind of remote sensing image compression method based on human eye vision Yu adaptive scanning of the present invention, is according to following What step was realized：The wavelet field visual sensitivity model of step one, foundation based on retina；After step 2, completion step one, knot The probability density function of human eye and remote sensing images observed range is closed, importance weighting mask is generated, and wavelet image is entered Row weighting；Each wavelet sub-band energy after step 3, calculating weighting, and it is suitable according to the scanning of the descending arrangement determination intersubband of energy Sequence；Step 4, the characteristic according to subband, determine the scanning sequency in subband；Step 5, determined according to step 3 and step 4 Scanning sequency in intersubband scanning sequency and subband, to the changing image X after weighting_wAdaptive scanning is carried out, one dimensional system is generated Number Sequence；Step 6, the one-dimensional coefficient sequence generated to step 5 using binary-tree coding method are encoded.

The beneficial effects of the invention are as follows：

1st, the present invention is carried out by visual importance on the basis of human vision model is considered to remote sensing images each position Weighting, additionally, according to the characteristic of different sub-band, the different scan method to Scan Design in intersubband scanning and subband is real Checking is bright, and in the case of to bit rates, the present invention can provide the visual quality of more preferable reconstruction image.

Brief description of the drawings

Fig. 1 is overall framework figure of the invention, represents the processing procedure to a width remote sensing images；

Fig. 2 is the mapping model schematic diagram of macula lutea and observed range, wherein 1 is retina, 2 is central fovea of retina, and 3 are One width remote sensing images, 4 project for central fovea of retina, the mapping positions on the retina of the point p on p ' expression remote sensing images, p_f Projection of the central fovea of retina on remote sensing images is represented, r represents retina radius, and e represents vision eccentricity, and u is point p to p_f Distance, v be human eye to image observed range；

Fig. 3 is the probability Distribution Model schematic diagram of observed range v；

Fig. 4 is importance weighting mask schematic diagram；Fig. 5 is scan mode schematic diagram, wherein (a) is " horizontal z scannings ", is used In the more subband of horizontal information, round dot represents the beginning of scanning, and arrow represents scanning direction, and (b) is " vertical z scannings ", uses In the more subband of vertical information, round dot represents the beginning of scanning, and arrow represents scanning direction；

Fig. 6 for the present invention checking in original image and corresponding scanning process schematic diagram, wherein (a) be Lunar (8, greatly It is small be 512 × 512), (b) is the process that adaptive scanning is carried out to the changing image after weighting, and arrow represents scanning direction, C () is the one-dimensional coefficient sequence of morton scanning generations, (d) is the one-dimensional coefficient sequence of the inventive method generation；

Fig. 7 is to test image " coastal-b1 ", using the inventive method and other scanning sides during the present invention is verified The reconstruction image comparison schematic diagram that method is obtained, wherein (a) is original image, when (b) is observed range v=5, central fovea of macula is seen The region for observing, (c) and (d) is, using SPIHT compression methods, respectively when code check is 0.0313bpp and 0.0625bpp, to obtain The reconstruction image for arriving, it, using JPEG2000 compression methods, is respectively 0.0313bpp and 0.0625bpp in code check that (e) and (f) is When, the reconstruction image for obtaining, it, using BTCA compression methods, is respectively 0.0313bpp and 0.0625bpp in code check that (g) and (h) is When, the reconstruction image for obtaining, (i) and (j) is to use the inventive method, is respectively 0.0313bpp and 0.0625bpp in code check When, the reconstruction image for obtaining, wherein vision raising effect becomes apparent in white small square frame；

Fig. 8 is for, to test image " coastal-b1 ", under different bit rates, several quality evaluations refer in present invention checking Target comparative result schematic diagram, wherein the result of (a) for FWQI, (b) is the result of VSNR, the result of (c) MS-SSIM；

Fig. 9 is the part remote sensing images schematic diagram used during the present invention is verified, wherein (a) is ocean_2kb1, (b) is Pavia1, (c) is pavia2, and (d) is houston, and (e) is pleiades_portdebouc_pan1, and (f) is pleiades_ portdebouc_pan2。

Specific embodiment

Specific embodiment of the invention is further described with reference to accompanying drawing.

Human eye is gathered by retina and processes visual information, referring to document [5], in retina, the sky of photoreceptor Between be unevenly distributed, it is maximum in the distribution density of macular region.With the increase with a distance from macula lutea, photoreceptor distribution is close Degree is strongly reduced, therefore, corresponding spatial vision frequency band is also strongly reduced.For human eye, it can not be perceived more than given The spatial frequency of cut-off frequency, that is to say, that from for the angle of HVS, it is not necessary that the extra high image of retaining space frequency Information.Therefore, under given bit rate, if it is desired to improve the visual quality of reconstruction image, must just consider retina Characteristic.

Specific embodiment one：Present embodiment is illustrated with reference to Fig. 1, the one kind described in present embodiment is based on human eye vision With the remote sensing image compression method of adaptive scanning, comprise the following steps：The wavelet field vision of step one, foundation based on retina Sensitivity model；After step 2, completion step one, with reference to human eye and the probability density function of remote sensing images observed range, generation Importance weighting mask, and wavelet image is weighted；Each wavelet sub-band energy after step 3, calculating weighting, and press Descending arrangement according to energy determines the scanning sequency of intersubband；Step 4, the characteristic according to subband, determine that the scanning in subband is suitable Sequence；Step 5, the intersubband scanning sequency determined according to step 3 and step 4 and scanning sequency in subband, to the change after weighting Change image X_wAdaptive scanning is carried out, one-dimensional coefficient sequence is generated；Step 6, using binary-tree coding method to step 5 generate One-dimensional coefficient sequence coding.

Afterwards, decoding end is decoded using Binomial model, the weighted transformation image after being rebuild, using vision plus Mask is counter is weighted for power, is that can obtain reconstruction image by inverse wavelet transform.

Specific embodiment two：Present embodiment from unlike specific embodiment one：Base is set up described in step one It is in the detailed process of the visual sensitivity model of retina：Step one by one, set up based on spatial domain visual sensitivity model；Step Rapid 1, the visual sensitivity model of wavelet field is set up.

Specific embodiment three：Present embodiment is illustrated with reference to Fig. 2, present embodiment is with specific embodiment one or two not Be：The detailed process of step described visual sensitivity model of the foundation based on spatial domain one by one is：For retina, Sensitivity highest of the vision in macular region.In document [26], it establishes macula lutea and the mapping model of observed range, such as Shown in Fig. 2.For a width remote sensing images, contrast threshold's value function in spatial domain is

Wherein f representation spaces frequency, e represents retinal eccentricity (degree), CT₀Minimum contrast threshold value is represented, α represents empty Between frequency decay constants, e₂Half-resolution eccentricity constant is represented, CT (f, e) represents visual contrast threshold value, and is f's and e Function；

To the eccentric ratio e for giving, corresponding visual cut-off frequency f is obtained using formula (1)_c, that is to say, that it is any to be higher than f_cFrequency be all that vision is sightless.Make CT (f, e)=1, you can obtain cut-off frequency f_cIt is as follows：

Obviously, cut-off frequency f_cIt is only dependent upon eccentric ratio e.

Assuming that the width of remote sensing images is N number of pixel, the corresponding picture position of central fovea of macula isIts In,Represent pixel p_fCorresponding abscissa,Represent pixel p_fCorresponding ordinate, observation from human eye to image away from It is known from v (being measured by picture traverse), is measured by pixel, point p to point p_fDistance for d (p)=| | p-p^f||₂, then by figure Image width degree is measured, point p to point p_fApart from u be u=d (p)/N, then eccentricity be

From formula (2) and (3), it can be seen that for the observed range for giving, cut-off frequency is the function of location of pixels. On the other hand, maximum visual perceived resolution is limited by display resolution r, i.e.,

According to sampling thheorem, the highest frequency without aliasing that display can be represented, i.e. nyquist frequency is

According to (2) and (5), to optional position p, final visual cut-off frequency is

Visual sensitivity model based on spatial domain is

Specific embodiment four：Unlike one of present embodiment and specific embodiment one to three：Step institute one by one The minimum contrast threshold value CT for stating₀It is 1/64, spatial frequency attenuation constant α is 0.106, half-resolution eccentricity constant e₂For 2.3。

Specific embodiment five：Illustrate present embodiment with reference to Fig. 2, present embodiment and specific embodiment one to four it Unlike one：The detailed process of the visual sensitivity model for setting up wavelet field described in step one two is：Most image Compression method is all carried out in wavelet field.For 9/7 wavelet transformation, error-detection threshold can be calculated using 2AFC methods, ginseng See document [27].By the fitting to experimental result, the error-detection threshold of wavelet coefficient is

Wherein a is constant, k is constant, f₀It is constant, g_θIt is constant, A_λ,θIt is the amplitude of 9/7 wavelet transformation basic function, λ is Wavelet decomposition layer, θ represents direction, and r is display resolution, a, k, f₀,g_θAnd A_λ,θValue see regarding for document [27] subband (λ, θ) Feel distortion sensitivity S_w(λ, θ) is

Based on (7) and (9), for wavelet field, relative to a specified central fovea of macula point, the vision of wavelet coefficient is quick Sensitivity model is

P represents the position of Arbitrary Coefficient in wavelet sub-band (λ, θ), β₁And β₂Represent respectively and be used for controlling S_wAnd S_fAmplitude Parameter；From Fig. 2, it will be seen that tan (e)=u/v=d/Nv, that is to say, that d=Nv tan (e) ≈ Nve.Here, e Unit be radian.For human visual system, highest visual acuity is in macular area and the visual angle (unit of 2 ° of covering： Radian).Therefore, d=Nv π/90.This explanation macular region is the circle that radius is d.

For macular region, the set of multiple macula lutea central points can be seen as.Assuming that having k macula lutea pointFor the wavelet coefficient at the p of position, its visual sensitivity model S is calculated according to formula (10)_i(v, p), i=1, 2 ..., k, finally, for a specified macular region, the visual sensitivity model of wavelet coefficient is

Specific embodiment six：Unlike one of present embodiment and specific embodiment one to five：The institute of step one two The control S for stating_wAnd S_fThe parameter beta of amplitude₁And β₂Respectively 0.01,3.

Specific embodiment seven：With reference to Fig. 3, Fig. 4 explanation present embodiments, present embodiment and specific embodiment one to Unlike one of six：Generation importance weighting mask described in step 2 and the specific mistake being weighted to wavelet image Cheng Wei：The purpose of importance weighting mask is to ensure to contribute larger bit first to encode and transmit visual quality.According to formula (10), visual sensitivity is closely connected with observed range.However, in actual applications, encoder is not aware that solution in advance The observed range v of code end subscriber, document [26] takes the probability distribution of observed range into account when design factor is weighted, and leads to Cross this mode and determine Weighted distance.

The probability density function of observed range is

Wherein, v represents human eye to the observed range of image, and μ is the average of function, and σ is the standard deviation of function；Observed range Probability density curve it is as shown in Figure 3.

The importance weighting mask of wavelet coefficient is at the p of position

Assuming that the observation of central fovea of macula point is picture centre, according to formula (11)~(13), generation importance weighting is covered Mould, as shown in Figure 4.

Specific embodiment eight：Illustrate present embodiment with reference to Fig. 4, present embodiment and specific embodiment one to seven it Unlike one：The standard deviation sigma of function is 0.4 in the probability density function of observed range, and the mean μ of function is 1.2586.

Adaptive scanning (HAS) based on HVS；Mask is weighted according to importance to be weighted each wavelet coefficient；

Assuming that changing image is X, its size is M × N, to the scanning process of image, be defined as from closed interval [1, 2 ..., M × N] arrive scanning sequency { (i, j):1≤i≤M, 1≤j≤N } bijective function f, the coordinate pair in its scanning sequency The position of coefficient in image is represented, after being scanned referring to document [28], the changing image of two dimension is converted into one-dimensional sequence, can It is expressed as [X_f(1),X_f(2),...,X_f(MN)].Different scan methods, such as morton are scanned, zigzag scannings, raster scannings Deng the bijective function f actually with different definition.Once it is determined that the definition of function f, to the encoded of changing image Journey, has reformed into one-dimensional sequenceCompression process.

For a width changing image, the coefficient for first scanning first is encoded.Generally, code stream is organized according to the order of scanning 's.This explanation, for the coefficient for first scanning, corresponding code stream is general before whole code stream.Obviously, this partial code streams Can first be decoded, be directly displayed after being decoded when necessary.Therefore, under identical bit rate, different scanning sequencies can give birth to Into the quality of different reconstruction images.If those coefficients bigger to image reconstruction contribution can be scanned first, then reconstruction image Quality be bound to be improved.

Usually, two dimensional image is to be converted into one-dimensional sequence, such as raster scanning by some classical scan methods, Zigzag is scanned, and morton scannings and Hilbert (Hilbert) are scanned.Different scan methods is suitable for different applications. Wherein, it is the Predicting Technique based on context that Lossless Compression is designed, typically using raster scanning.Document [28] is pointed out, adopted During with based on the technology predicted, the spatial coherence of image makes raster scanning be better than other scan modes in itself.It is right In the coding method based on dct transform, zigzag scan methods can effectively organize conversion coefficient.However, zigzag is scanned Substantially it is a kind of " line " scanning, that is to say, that it is only capable of representing unessential " line ".For wavelet image, Wavelet field, unessential " block " is more common.If these unessential " blocks " can be scanned reasonably, coding efficiency is incited somebody to action To very big improvement.

Morton scanning can preferably using this " block " relation, some classical scan methods, such as EZW, SPIHT, is all based on morton scannings.But, although morton scannings can in the way of " block " scan image, but not There is the characteristic in view of subband.Remote sensing images usually contain the detailed information compared with horn of plenty, after this causes wavelet transformation, high frequency Information in band is still relatively abundanter.And, the information content that each subband is included may differ by a lot, and this depends on the interior of image Hold.Therefore, fixed scanning may cause very big influence to coding efficiency.Additionally, Hilbert scannings can also regard one as " block " scanning is planted, because with good local retention performance, Hilbert scannings were once considered as the best of compression of images Scan method.However, similar with morton scannings, Hilbert scannings still do not take the characteristic of subband into account.

According to analysis above, the scanning to wavelet image, the scan method based on " block " is more suitable for.Problem It is：For giving wavelet sub-band, which kind of " block " scan method is more suitable forAdditionally, relative to natural image, the height of remote sensing images Frequency sub-band information typically more enriches, and it is larger that this causes that different intersubband scanning sequencies can be produced to the quality of reconstruction image Influence.Therefore, how to determine the scan method of intersubband, be the problem that another needs to consider.

Specific embodiment nine：Unlike one of present embodiment and specific embodiment one to eight：Described in step 3 Calculating weighting after each wavelet sub-band energy, and according to energy descending arrangement determine intersubband scanning sequency detailed process For：If changing image is X, the wavelet decomposition number of plies is J, step 3 one, to changing image X, calculates corresponding using formula (13) Importance weighting mask W；Step 3 two, with importance weight mask changing image is weighted, by the Transformation Graphs after weighting As being expressed as X_w, i.e. X_w=XW；Step 3 three, calculating X_wIn each subband energy, and be denoted as E_λ,θ, λ represents this The corresponding small echo number of plies of subband (λ=1,2 ..., J), θ represents the direction of the subband, θ=1,2,3,4, wherein, " 1 " represents minimum Frequency subband, " 2 " represent horizontal direction subband, and " 3 " represent diagonally opposed subband, and " 4 " represent vertical direction subband.To weighting subband X_w(λ, θ), corresponding subband energy is

R and C represents the line number and columns of the subband, X respectively_w(λ, θ) (i, j) represents that in the small echo number of plies be λ, and direction is θ Weighting subband in, positioned at the coefficient value at point (i, j) place；

Step 3 four, to all subbands, by ENERGY E_λ,θThe order of descending, determines the scanning sequency of intersubband；

From mathematical angle, it is expressed as

Permu represents the rearrangement of sub-band sequence.

Specific embodiment ten：Unlike one of present embodiment and specific embodiment one to nine：Described in step 4 The characteristic according to subband, the detailed process for determining the scanning sequency in subband is：

To each weighting subband X_w(λ,θ)：

(1) if the subband direction is " 1 " or " 2 ", using horizontal_z scan modes；

(2) if the subband direction is " 4 ", using vertical_z scan modes；Vertical_z scan modes are shown in Fig. 5 (b).The scan mode is that the subband more to vertical information is designed, and the principle for designing scanning is " first vertical rear water It is flat ", while scanning element is carried out in the way of " block ", take into account the advantage of " block " scanning.

(3) if the subband direction is " 3 ", scan mode is by this layer of direction for the subband of " 2 " and " 4 " is together decided on；

If E_λ,2≥E_λ,4, then the subband is using horizontal_z scan modes；

If E_λ,2<E_λ,4, then the subband is using vertical_z scan modes.

Main thought of the invention is as follows：First mask is weighted using importance to weight changing image；Then weighting is calculated The scanning sequency of each sub-belt energy, and the descending arrangement determination intersubband according to energy afterwards；Finally, to sweeping inside weighting subband Retouch, the characteristic according to subband determines scan method.To horizontal direction subband, its reflection is image information in the horizontal direction, Therefore " horizontal_z scannings " is used as scan method.Equally, to those vertical direction subbands, its reflection is that image hangs down Nogata to information, if scanning can be carried out along vertical direction, under to constant bit rate, more vertical informations can be protected Stay.In the present invention, the method that a kind of " vertical_z scannings " is proposed to vertical subband." horizontal_z scannings " and " vertical_z scannings " respectively such as (a) in Fig. 5, shown in (b).For diagonal subband, scan method depends on image in itself.If In current wavelet layer, the horizontal information of image is more than vertical information, then diagonal subband uses " horizontal_z scannings ", instead It, diagonal subband uses " vertical_z scannings ".

It is as follows to checking of the invention：

In original image such as Fig. 6 shown in (a).For the image for giving, its visual sensitivity model is first calculated, and combine sight Find range from probability density function, generation importance weighting mask.Then mask is weighted to each wavelet coefficient according to importance It is weighted.If the wavelet decomposition number of plies is 3, the method according to the invention, the energy of subband after each weighting is calculated, the results are shown in Table 1.According to table 1, it may be determined that the scanning sequency of subband, i.e. LL after weighting₃、LH₃、HH₃、HL₃、LH₂、HH₂、HL₂、LH₁、HL₁、HH₁。

Then, to each weighting subband, its inner scanning order is determined by subband characteristic.To subband LL₃、HL₃、HL₂And HL₁, using " horizontal_z scannings " mode；To subband LH₃、LH₂And LH₁, by the way of " vertical_z scannings ". According to table 1, it can be seen that to every layer of wavelet sub-band, vertical direction information is more than horizontal direction information, therefore to all to silver coin Band HH₃、HH₂And HH₁, using " vertical_z scannings ".In whole scanning process such as Fig. 6 shown in (b).Finally, one can be generated The coefficient sequence of dimension.(c) and (d) represents the one-dimensional coefficient matrix using morton scannings and present invention generation respectively in Fig. 6.From Fig. 6 can be seen that the inventive method and can come one-dimensional sequence with the scan convertion image of self adaptation, the coefficient for making those important Above.Simultaneously as the interior scanning of subband is for the purpose of retaining image texture information as far as possible, therefore can be in the base of visual weight On plinth, the visual effect of reconstruction image is further improved.

The energy (× 10 of each subband in the changing image of table 1⁹)

Overhead bit：

Because importance weighting mask is unrelated with picture material.Therefore, the mask need not pass to decoding end.Only by macula lutea Center concave point, and picture traverse is sent to decoding end as side information.In the present invention, record yellow with four integers (wherein, two integers record the abscissa positions of the point to spot center concave point, and two other integer records the ordinate position of the point Put), two integers are used for recording picture traverse.Therefore, the side information needed for whole weighting mask, it is only necessary to six integers.

Additionally, for adaptive scanning, the scanning sequency of side packet enclosed tool interband and the scan method of diagonal subband.It is right J layers of wavelet transformation, has 3J+1 subband, therefore needs 3J+1 integer altogether and represent the scanning sequency of intersubband.Additionally, also needing J Individual integer represents the scan method of all diagonal subbands.Therefore, the side information required for adaptive scanning, needs (3J+1)+J altogether =4J+1 integer.Based on analysis above, total side information of coding side needs 6+ (4J+1)=4J+7 integer altogether.

By taking (a) in Fig. 6 as an example, the size of image " lunar " is 512 × 512, and the wavelet decomposition number of plies is 3 layers, then the present invention Total overhead bit be (4 × 3+7)/(512 × 512) ≈ 0.00725%.That is, image is bigger, the ratio shared by expense It is smaller.And, if further using entropy code to overhead-bits, overhead bit can be smaller.It is right according to analysis above In compression method proposed by the present invention, total overhead bit is very small, it might even be possible to ignore.

Binary-tree coding：

For compression of images, at present using the in the majority of embedded encoded method, because image is supported in this coding method Progressive transmission, and can code stream optional position decode.

Document [25] proposes a kind of newest embedded encoded method based on binary tree, i.e. self adaptation binary-tree coding Method (binary tree coding adaptively, BTCA).The method does not need complicated step, is such as based on context Model, rate-distortion optimization etc. greatly reduces algorithm complex.Its basic process is as follows：First, according to scanning after one Dimension sequence sets up binary tree.Then, the binary tree is encoded, coding is that order from left to right is entered according to from bottom to up OK.The adjacent coefficient that make use of significant coefficient in an encoding process is generally also this important thought, is carried to a certain extent Coding efficiency high.The implementation process of binary-tree coding is given in detail below.

Function code=BTCA (T_k)

Input：Binary tree is represented with Γ, the position of the upper node of tree, T are represented with i_kRepresent threshold value, initial thresholdAnd T_k=T₀/2^k；

Initialization:If the height of tree of whole binary tree is D, d=D is made；

While(d>1)

{

(1)

(2) Let ct={ } .If Γ (i) >=T_k-1

If the adjacent node of Γ (i) is Γ (i+1), and Γ (i+1) is inessential, then ct=Bin_Tree_Enc (Γ, i+1, T_k)；

Otherwise

If the adjacent node of Γ (i) is Γ (i-1), and Γ (i-1) is inessential, then ct=Bin_Tree_Enc (Γ, i-1, T_k)；

(3) code={ code, ct },

(4) d=d-1；

Output:It is T in threshold value_kWhen, the code stream of correspondence bit plane output.

}

In above-mentioned BTCA algorithms, to giving nodes encoding, Bin_Tree_Enc algorithms are employed.The algorithmic procedure is such as Under：

Function code=Bin_Tree_Enc (Γ, i, T_k)

Input：Binary tree is represented with Γ, the position of the upper node of tree, T are represented with i_kRepresent threshold value, initial thresholdAnd T_k=T₀/2^k；

(1) if Γ (i) is encoded in bigger threshold value, i.e. Γ (i) >=T_k-1, then

If Γ (i) is not in the bottom of tree, two child nodes of coding Γ (i)；Otherwise encode the sign bit of Γ (i).

(2) if Γ (i) has an important father node, and Γ (i) adjacent node be it is unessential, then：

If Γ (i) is not in the bottom of tree, two child nodes of coding Γ (i)；Otherwise encode the sign bit of Γ (i).

(3) if Γ (i) >=T_k

If Γ (i) adds " 1 " not in the bottom of tree, two child nodes of coding Γ (i) before code stream；Otherwise compile The sign bit of code Γ (i), and add " 1 " before code stream.

(4) otherwise

Output ' 0 '.

Output：The code stream of the subtree with Γ (i) as tree root.

Because binary-tree coding method is relatively easy, and efficiency is higher, therefore the present invention is carried out based on content to changing image Adaptive scanning after, using binary-tree coding method to generate sequence encode.

Quality evaluation index：

Compression method proposed by the present invention is to meet the demand of growing remote sensing images online browse.Therefore, The compression method of proposition should be weighed using the index related to human eye vision.In the present invention, using FWQI [26], VSNR And MS-SSIM [38] is as evaluation index [37],.

A、FWQI

In document [29], Zhou Wang etc. propose a kind of image quality evaluation index, and the method regards image fault as It is three kinds of functions of factor, i.e. degree of correlation distortion, luminance distortion, contrast distortion.Then, the distortion is extended into wavelet field, Define wavelet field corresponding distortion FWQI be

Here, M represents wavelet coefficient number, c (x_n) represent position in x_nWavelet coefficient, Q (x_n) represent in quality evaluation Position x in figure_nMass value.Due to S (v, x_n) it is to change with v, therefore, for a width test image, its FWQI is the letter of v Number.

B、VSNR

In document [37], Chandler etc. proposes a kind of effective matrix, i.e. peak vision signal to noise ratio matrix (VSNR), Nearly threshold value and threshold value property excessively according to human eye, carry out the vision distortion of quantification image.Relative to other vision distortion matrixes, VSNR more can effectively reflect the truly feels of human eye.VSNR (units：DB) can be defined as follows

Here, C (f) represents the contrast of original image f, d_pcRepresent the contrast distortion that human eye is perceived, d_gpRepresent whole The distortion level of body, α is set as 0.04.

C、MS-SSIM

MS-SSIM is Multi-scale model similarity measure, and this is estimated and considers various observation conditions, is surveyed than structural similarity Degree (SSIM) is more flexible.Therefore, the present invention uses MS-SSIM as a kind of measuring quality method.

Here, l (x, y), c (x, y) and s (x, y) represent brightness, contrast and structural comparing respectively.

Experiment and result：

In order to prove the performance of compression method proposed by the present invention, some experiments are done, and under different bit rates, with Other methods for being based on scanning compare.

D, pretreatment：

The remote sensing images of a few width difference locating depths are have chosen in order to prove the validity of the inventive method, in experiment, it is most of Remote sensing images all have resolution ratio higher.

Some test remote sensing images are obtained from CCSDS test libraries, referring to document [39], including " lunar ", " coastal-b1 ", " ocean-2kb1 ", " pleiades_portdebouc_pan ".To these images, we intercept the upper left corner Size tested for 512 × 512 image, it is therefore an objective to be compared under the same conditions.Additionally, also have chosen other two Width remote sensing images are tested.A wherein width " pavia " is the Pavia ground in North of Italy by QuickBird sensors What area obtained, resolution ratio is 0.6m.Another piece image " houston " is in U.S. by sensor WorldView-2 in 2013 What the Houston area of state obtained, resolution ratio is 0.5m.The size of image " pavia " and " houston " is 512 × 512. Whole test image is concentrated, and the locating depth of " lunar " and " coastal-b1 " is 8, and the locating depth of " ocean_2kb1 " is 10, The locating depth of " pavia " and " houston " is 11bit, and the locating depth of " pleiades_portdebouc_pan " is 12.Image In " lunar " such as Fig. 6 shown in (a), image " coastal-b1 " is shown in that (a) is shown in Fig. 7.Remaining test image is shown in Fig. 9.

Subjective quality compares：

In order to compare regarding for the reconstruction images that compression method proposed by the present invention and other methods for being based on scanning are obtained Feel quality, first tested using image " coastal-b1 ".Assuming that central fovea of macula point observation be image center, and see It is 5 to find range from v, and the wavelet decomposition number of plies is 5 layers.SPIHT, JPEG2000, BTCA is respectively adopted, and the inventive method is pressed Contracting.The visual quality comparative result of reconstruction image is as shown in Figure 7 under different bit rates.

From Fig. 7, it can be seen that under to bit rates, using compression method proposed by the present invention, its reconstruction for obtaining The total quality of image is better than other methods based on scanning.One reason for this is that, the method for proposition employs vision Weighting mask is weighted to Wavelet image, and this can guarantee that contributes big bit to be first scanned and compile to rebuilding visual quality of images Code.Further it is proposed that adaptive scanning process can retain more texture informations.Therefore, the entirety of reconstruction image Visual quality is necessarily improved.

In order to further prove the validity of the inventive method, for the test image (a) in Fig. 7, do more real Test into line justification.When bitrate range is 0.0313bpp to 1bpp, FWQI, VSNR and MS-SSIM that all methods are obtained Result, respectively such as (a) in Fig. 8, shown in (b) and (c).As can be seen that from the angle of objective evaluation, in whole given bit In the range of rate, the visual quality of the reconstruction image that method proposed by the present invention is obtained still is based on the method for scanning better than other.

The present invention is based on the Performance comparision of scan method with other：

Usually, different images has different contents, including complexity, texture etc..From the angle of evaluation algorithms, no It should be evaluated only with piece image, and multiple image should be taken and tested, and take the average of result.The present invention has done more The validity of the extracting method to verify is tested more.In experiment, choose several test images and tested, including in Fig. 6 (a), (a) in Fig. 7, used as test image, each image carries out the decomposition of the biorthogonal wavelet of Pyatyi 9/7 to (a)~(f) in Fig. 9. According to the probability distribution of the observed range be given in Fig. 3, it can be seen that the observed range v of maximum possible is 3.Therefore, at these In experiment, the value of v is set to 3.Under different bit rates, the result point of institute methodical FWQI, VSNR, and MS-SSIM 2, table 3, table 4, table 5, table 6 and table 7 are not shown in Table.

In 2~table of table 7, with " J2K " expression " JPEG2000 ", it can be seen that for all given bit rates, the present invention The average of the FWQI, VSNR, and the MS-SSIM that obtain all is highest.This explanation, the method for scanning is based on compared to other, this Invention can provide more preferable reconstructed image quality.

The FWQI of the inventive method of table 2 and other compression methods for being based on scanning compares

The FWQI of the inventive method of table 3 and other compression methods for being based on scanning compares

The VSNR (dB) of the inventive method of table 4 and other compression methods for being based on scanning compares

The VSNR (dB) of the inventive method of table 5 and other compression methods for being based on scanning compares

The MS-SSIM of the inventive method of table 6 and other compression methods for being based on scanning compares

The MS-SSIM of the inventive method of table 7 and other compression methods for being based on scanning compares

Result and conclusion：

The present invention generates visual weight mask first according to the visual characteristic of human eye.Secondly, to the Transformation Graphs after weighting Picture, according to the importance of visual weight subband, designs the scanning sequency of intersubband.For the scanning in subband, purpose of design is Retain more detailed information as far as possible, this contributes on the basis of excessive data is not increased, improve the vision of reconstruction image Quality.Finally, the one-dimensional coefficient sequence of generation is encoded using binary-tree coding device.The compression method expense of proposition is minimum, very To can be ignored.The results show, the method for being based on scanning with other is compared, and the present invention can provide preferably reconstruction The visual quality of image.Whole scanning process of the invention can be regarded as being realized by two stages.The purpose of first stage It is the visual characteristic according to human eye, generates importance weighting mask, this helps to ensure that those contribute bigger to visual quality Bit is first scanned.Second stage, for being scanned in intersubband scanning and subband, sets respectively to the changing image after visual weight The different scanning sequency of meter.Finally, the one-dimensional coefficient sequence for generating is encoded using binary-tree coding device.

The present invention effectively improves the visual quality for rebuilding remote sensing images, meets remote sensing images growing at present online The demand for browsing.Online browse of the present invention suitable for remote sensing images.

Bibliography of the present invention is as follows：

[1]J.M.Shapiro,“Embedded image coding using zerotrees of wavelet coefficients,”IEEE Trans.Signal Process.,vol.41,no.12,pp.3445–3462,Dec.1993.

[2]A.Said and W.A.Pearlman,“A new,fast,and efficient image codec based on set partitioning in hierarchical trees,”IEEE Trans.Circuits Syst.Video Technol.,vol.6,no.3,pp.243–250,Jun.1996.

[3]W.A.Pearlman,A.Islam,N.Nagaraj,and A.Said,“Efficient low complexity image coding with a set-partitioning embedded block coder,”IEEE Trans.Circuits Syst.Video Technol.,vol.14,no.11,pp.1219–1235,Nov.2004.

[4]JPEG2000Image Coding System,ISO/IEC Std.15 444-1,2000.

[5]A.Beghdadi,M.C.Larabi,A.Bouzerdoum,and K.M.Lftekharuddin,“A survey of perceptual image processing methods,”Signal Processing:Image Communication,vol.28,no.8,pp.811-831,Sep.2013.

[6]B.Macq,and H.Q.Shi,“Perceptually weighted vector quantization in the DCT domain,”Electronics Letters,vol.29,no.15,pp.1382–1384,Jul.1993.

[7]I.Hontsch and L.J.Karam,“Locally adaptive perceptual image coding,”IEEE Trans.Image Process.,vol.9,no.9,pp.1472–1483,Sep.2000.

[8]I.Hontsch and L.J.Karam,“Adaptive image coding with perceptual distortion control,”IEEE Trans.Image Process.,vol.11,no.9,pp.213–222, Mar.2002.

[9]M.G.Albanesi and F.Guerrini,“An HVS-based adaptive coder for perceptually lossy image compression,”Pattern Recognition,vol.36,no.4,pp.997– 1007,Apr.2003.

[10]M.J.Nadenau,J.Reichel,and M.Kunt,“Wavelet-based color image compression:exploring the contrast sensitivity function,”IEEE Trans.Image Process.,vol.12,no.1,pp.58–70,Jan.2003.

[11]Z.Liu,L.J.Karam,and A.B.Watson,“JPEG2000 encoding with perceptual distortion control,”IEEE Trans.Image Process.,vol.15,no.7,pp.1763–1778, Jul.2006.

[12]G.Sreelekha,P.S.Sathidevi,“An HVS based adaptive quantization scheme for the compression of color images,”Digital Signal Processing, vol.20.no.4.pp.1129–1149,Jul.2010.

[13]D.Wu,D.M.Tan,M.Baird,and J.DeCampo,etc,“Perceptually lossless medical image coding,”IEEE Trans.Medical Image,vol.25,no.3,pp.335–344, Mar.2006.

[14]X.H.Zhang,W.S.Lin,P.Xue,“Just-noticeable difference estimation with pixels in images,”J.Vis.Commun.Image R,vol.19,no.1,pp.30-41,Jan.2008.

[15]Y.Niu,X.L.Wu,G.M.Shi,and X.T.Wang,“Edge-based perceptual image coding,”IEEE Trans.Image Process.,vol.21,no.4,pp.1899–1910,Apr.2012.

[16]D.M.Tan,C.S.Tan,and H.R.Wu,“Perceptual Color Image Coding With JPEG2000,”IEEE Trans.Image Process.,vol.19,no.2,pp.374–383,Feb.2010.

[17]H.Oh,A.Bilgin,and M.W.Marcellin,“Visually Lossless Encoding for JPEG2000,”IEEE Trans.Image Process.,vol.22,no.1,pp.189–201,Jan.2013.

[18]A.L.N,T.D.Costa,M.N.Do,“A Retina-Based Perceptually Lossless Limit and a Gaussian Foveation Scheme With Loss Control,”IEEE J.Sel.Topics.Signal Process,vol.8,no.3,pp.438–453,Jun.2014.

[19]B.Li,R.Yang,and H.X.Jiang.“Remote-sensing image compression using two-dimensional oriented wavelet transform,”IEEE Trans.Geosci.Remote Sens., vol.49,no.1,pp.236–250,Jan.2011.

[20]A.Karami,M.Yazdi,and G.Mercier.“Compression of hyperspectral images using discerete wavelet transform and tucker decomposition,”IEEE J.Sel.Topics Appl.Earth Observ.,vol.5,no.2,pp.444–450,Apr.2012.

[21]X.Zhan,R.Zhang,D.Yin,and A.Z.Hu.“Remote sensing image compression based on double-sparsity dictionary learning and universal trellis coded quantization,”in Proc.IEEE Int.Conf.Image Process.,2013,pp.1665-1669.

[22]C.Jiang,H.Y.zhang,H.F.Shen,and L.P.Zhang.“Two-Step Sparse Coding for the Pan-Sharpening of Remote Sensing Images,”IEEE J.Sel.Topics Appl.Earth Observ.,vol.7,no.5,pp.1792–1805,May.2014.

[23]P.Kulkarni,A.Bilgin,M.W.Marcellin,and J.C.Dagher.“Compression of earth science data with JPEG2000,”in Hyperspectral Data Compression.pp.347– 378,2006.

[24]F.García-Vílchez and J.Serra-Sagristà,“Extending the CCSDS recommendation for image data compression for remote sensing scenarios,”IEEE Trans.Geosci.Remote Sens.,vol.47,no.10,pp.3431–3445,Oct.2009.

[25]K.K.Huang and D.Q.Dai,“A new on-board image codec based on binary tree with adaptive scanning order in scan-based mode,”IEEE Trans.Geosci.Remote Sens.,vol.50,no.10,pp.3737-3750,Oct.2012.

[26]Z.Wang and A.C.Bovik,“embedded foveation image coding,”IEEE Trans.Image Process.,vol.10,no.10,pp.1397-1410,Oct.2001.

[27]A.B.Watson,G.Y.Yang,J.A.Solomon,and J.Villasenor.“Visibility of wavelet quantization noise,”IEEE Trans.Image Process.,vol.6,no.8,pp.1164- 1175,Aug.1997.

[28]N.Memon,D.L.Neuhoff,and S.Shende,“An analysis of some common scanning techniques for lossless image coding,”IEEE Trans.Image Process., vol.9,no.11,pp.1837-1848,Nov.2000.

[29]Z.Wang and A.C.Bovik,“A universal image quality index,”IEEE Signal Process.Lett.,vol.9,no.3,pp.81–84,Mar.2002.

[30]S.Patel,and S.Srinivasan,“Modified embedded zerotree wavelet algorithm for fast implementation of wavelet image codec,”Electronics Letters,vol.36,no.20,pp.1713-1714,Sep.2000.

[31]V.N.Ramaswamy,K.R.Namuduri,and N.Ranganathan,“Context-based lossless image coding using EZW framework,”IEEE Trans.Circuits Syst.Video Technol.,vol.11,no.4,pp.554–559,Apr.2001.

[32]S.R.Chang and L.Carin,“A modified SPIHT algorithm for image coding with a joint MSE and classification distortion measure,”IEEE Trans.Image Process.,vol.15,no.3,pp.713-725,Mar.2006.

[33]Z.J.Fang,N.X.Xiong,L.T.Yang,X.M.Sun,and Y.Yang,“Interpolation- Based Direction-Adaptive Lifting DWT and Modified SPIHT for Image Compression in Multimedia Communications,”IEEE Systems Journal,vol.5,no.4,pp.584-593, Dec.2011.

[34]Y.Jin and H.J.Lee,“A block-based pass-parallel SPIHT algorithm,” IEEE Trans.Circuits Syst.Video Technol.,vol.22,no.7,pp.1064-1075,Jul.2012

[35]Z.Y.Wu,A.Bilgin,and M.W.Marcellin,“Joint source/channel coding for image transmission with JPEG2000 over memoryless channels,”IEEE Trans.Image Process.,vol.14,no.8,pp.1020-1032,Aug.2005.

[36]J.Y.Yang,Y.Wang,W.L.Xu,and Q.H.Dai,“Image coding using dual-tree discrete wavelet transform,”IEEE Trans.Image Process.,vol.17,no.9,pp.1555- 1569,Sep.2008.

[37]D.M.Chandler and S.S.Hemami.“VSNR:A wavelet-based visual signal- to-noise ratio for natural images,”IEEE Trans.Image Process.,vol.16,no.9, pp.2284-2298,Sep.2007.

[38]Z.Wang,E.P.Simoncelli and A.C.Bovik,″Multiscale structural similarity for image quality assessment,″in Proc.IEEE Asilomar Conf.Signals, Syst.,Comput.,Pacific Grove,CA,Nov.2003,pp.1398–1402.

[39]CCSDS reference test image set,Apr.2007.[Online].Available: http://cwe.ccsds.org/sls/docs/sls-dc/.

Claims (2)

1. a kind of remote sensing image compression method based on human eye vision Yu adaptive scanning, it is characterised in that methods described include with Lower step：

The wavelet field visual sensitivity model of step one, foundation based on retina；

After step 2, completion step one, with reference to human eye and the probability density function of remote sensing images observed range, generation importance adds Power mask, and wavelet image is weighted；

Step 3, calculate weighting after each wavelet sub-band energy, and according to energy descending arrangement determine intersubband scanning sequency；

Step 4, the characteristic according to subband, determine the scanning sequency in subband；

Step 5, the intersubband scanning sequency determined according to step 3 and step 4 and scanning sequency in subband, after weighting Changing image X_wAdaptive scanning is carried out, one-dimensional coefficient sequence is generated；

Step 6, the one-dimensional coefficient sequence generated to step 5 using binary-tree coding method are encoded；

The detailed process of visual sensitivity model of the foundation based on retina described in step one be：

Step one by one, set up based on spatial domain visual sensitivity model；

Step one two, the visual sensitivity model for setting up wavelet field；

The detailed process of step described visual sensitivity model of the foundation based on spatial domain one by one is：

For a width remote sensing images, contrast threshold's value function in spatial domain is

$CT(f,e)={\mathrm{CT}}_0\exp \left(\mathrm{\&alpha;}f\frac{e+e_2}{e_2}\right)---\left(1\right)$

Wherein f representation spaces frequency, e represents retinal eccentricity, CT₀Minimum contrast threshold value is represented, α representation space frequencies decline Subtract constant, e₂Half-resolution eccentricity constant is represented, CT (f, e) represents visual contrast threshold value, and is the function of f and e；

To the eccentric ratio e for giving, corresponding visual cut-off frequency f is obtained using formula (1)_c, CT (f, e)=1 is made, ended Frequency f_cIt is as follows：

$f_c=\frac{e_{2}ln\left(\frac{1}{{\mathrm{CT}}_0}\right)}{\mathrm{\&alpha;}(e+e_2)}---\left(2\right)$

Assuming that the width of remote sensing images is N number of pixel, the corresponding picture position of central fovea of macula isWherein, Represent pixel p_fCorresponding abscissa,Represent pixel p_fCorresponding ordinate, the observed range v from human eye to image is It is known, measured by pixel, point p to point p_fDistance for d (p)=| | p-p^f||₂, then measured by picture traverse, point p to point p_f Apart from u be u=d (p)/N, then eccentricity be

$e(v,p)={\tan }^{-1}\left(\frac{d\left(p\right)}{Nv}\right)---\left(3\right)$

Maximum visual perceived resolution is limited by display resolution r, i.e.,

$r=\frac{\mathrm{\&pi;}Nv}{180}---\left(4\right)$

According to sampling thheorem, the highest frequency without aliasing that display can be represented, i.e. nyquist frequency is

$f_d=\frac{r}{2}=\frac{\mathrm{\&pi;}Nv}{360}---\left(5\right)$

According to (2) and (5), to optional position p, final visual cut-off frequency is

$f_m\left(p\right)=\min (f_c,f_d)=\min (\frac{e_{2}ln\left(\frac{1}{{\mathrm{CT}}_0}\right)}{\mathrm{\&alpha;}(e+e_2)},\frac{\mathrm{\&pi;}Nv}{360})---\left(6\right)$

Visual sensitivity model based on spatial domain is

$S_f(v,f,p)=\left\{\begin{array}{cc}\frac{CT(f,0)}{CT(f,e)}=\exp (-0.0461f\&CenterDot;e)& f\&le;f_m\left(p\right)\\ 0& f>f_m\left(p\right)\end{array}\right.---\left(7\right);$

Step minimum contrast threshold value CT described one by one₀It is 1/64, spatial frequency attenuation constant α is 0.106, and half-resolution is inclined Heart rate constant e₂It is 2.3；

The detailed process of the visual sensitivity model for setting up wavelet field described in step one two is：

The error-detection threshold of wavelet coefficient is

$T_{\mathrm{\&lambda;},\mathrm{\&theta;}}=\frac{a{10}^{k\left(\log {\left(g_{\mathrm{\&theta;}}{f}_0{2}^{\mathrm{\&lambda;}}\right)}^2\right)}}{A_{\mathrm{\&lambda;},\mathrm{\&theta;}}}---\left(8\right)$

Wherein a is constant, k is constant, f₀It is constant, g_θIt is constant, A_λ,θIt is the amplitude of 9/7 wavelet transformation basic function, λ is small echo Decomposition layer, θ represents direction；

The vision distortion susceptibility S of subband (λ, θ)_w(λ, θ) is

$S_w(\mathrm{\&lambda;},\mathrm{\&theta;})=\frac{1}{T_{\mathrm{\&lambda;},\mathrm{\&theta;}}}---\left(9\right)$

Based on (7) and (9), for wavelet field, relative to a specified central fovea of macula point, the visual sensitivity of wavelet coefficient Model is

$S(v,p)={\&lsqb;S_w(\mathrm{\&lambda;},\mathrm{\&theta;})\&rsqb;}^{{\mathrm{\&beta;}}_1}\&CenterDot;{\&lsqb;S_f(v,f,p)\&rsqb;}^{{\mathrm{\&beta;}}_2}---\left(10\right)$

P represents the position of Arbitrary Coefficient in wavelet sub-band (λ, θ), β₁And β₂Represent respectively and be used for controlling S_wAnd S_fThe parameter of amplitude；

Assuming that having k macula lutea pointFor the wavelet coefficient at the p of position, its vision is calculated according to formula (10) quick Sensitivity model S_i(v, p), i=1,2 ..., k, finally, for a specified macular region, the visual sensitivity of wavelet coefficient Model is

$S(v,p)=\underset{i=\mathrm{1...},k}{max}\left(S_i\right(v,p\left)\right)---\left(11\right);$

Control S described in step one two_wAnd S_fThe parameter beta of amplitude₁And β₂Respectively 0.01,3；

Generation importance weighting mask described in step 2 simultaneously to the detailed process that wavelet image is weighted is：

Using the probability density function of following observed range：

$f\left(v\right)=\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;}v}\exp (-(\frac{{(lnv-\mathrm{\&mu;})}^2}{2{\mathrm{\&sigma;}}^2}\left)\right),v\&Element;(0,+\mathrm{\&infin;})---\left(12\right)$

Wherein, v represents human eye to the observed range of image, and μ is the average of function, and σ is the standard deviation of function；

The importance weighting mask of wavelet coefficient is at the p of position

$W\left(p\right)={\&Integral;}_0^{+\mathrm{\&infin;}}f\left(v\right)S(v,p)d\left(v\right)---\left(13\right)$

Assuming that the observation of central fovea of macula point is picture centre, according to formula (11)~(13), importance weighting mask is generated, used The importance weighting mask of generation is weighted to wavelet image, the changing image after generation weighting；

The standard deviation sigma of function is 0.4 in the probability density function of the observed range, and the mean μ of function is 1.2586；

Described in step 3 calculating weighting after each wavelet sub-band energy, and according to energy descending arrangement determine intersubband scanning The detailed process of order is：If changing image is X, the wavelet decomposition number of plies is J,

Step 3 one, to changing image X, corresponding importance weighting mask W is calculated using formula (13)；

Step 3 two, with importance weight mask changing image is weighted, the changing image after weighting is expressed as X_w, i.e.,

X_w=XW；

Step 3 three, calculating X_wIn each subband energy, and be denoted as E_λ,θ, λ represents the corresponding small echo number of plies of the subband, λ=1,2 ..., J, θ represent the direction of the subband, θ=1,2,3,4, wherein, " 1 " represents lowest frequency subband, and " 2 " represent level side To subband, " 3 " represent diagonally opposed subband, and " 4 " represent vertical direction subband, to weighting subband X_w(λ, θ), corresponding subband energy ForR and C represents the line number and columns of the subband, X respectively_w(λ, θ) (i, j) is represented in small echo The number of plies is λ, during direction is for the weighting subband of θ, positioned at the coefficient value at point (i, j) place；

Step 3 four, to all subbands, by ENERGY E_λ,θThe order of descending, determines the scanning sequency of intersubband.

2. a kind of remote sensing image compression method based on human eye vision Yu adaptive scanning according to claim 1, it is special It is the characteristic according to subband described in step 4 to levy, and the detailed process for determining the scanning sequency in subband is：

To each weighting subband X_w(λ,θ)：

(1) if the subband direction is " 1 " or " 2 ", using horizontal_z scan modes；

(2) if the subband direction is " 4 ", using vertical_z scan modes；

(3) if the subband direction is " 3 ", scan mode is by wavelet decomposition layer direction for the subband of " 2 " and " 4 " is together decided on；

If E_λ,2≥E_λ,4, then the subband is using horizontal_z scan modes；

If E_λ,2<E_λ,4, then the subband is using vertical_z scan modes.