CN102186069B - Remote sensing image data compression method capable of maintaining measurement performance - Google Patents

Abstract

The invention discloses a remote sensing image data compression method capable of maintaining measurement performance. A remote sensing image is processed by adopting wavelet transform, so the method is suitable for the compression of the remote sensing image. The remote sensing image is converted and coded by adopting integer wavelet transform, so transformation from an integer to another integer can be realized, the inverse transformation of the transformation from the integer to another integer can completely reconstruct the image, the uncontrollable loss of image information is avoided, and integer-wavelet-transform-based lossless image compression becomes possible. The compression efficiency is improved, and a compression ratio is increased by improved embedded zerotree wavelet (EZW) coding methods for the establishment of a zerotree marker table, the integer updating processing of a wavelet decomposition coefficient, the change of a scanning direction of the wavelet decomposition coefficient, the independent processing of the lowest frequency sub-band coefficient and the like. Characteristic points of important characteristics in the remote sensing image are extracted, and are identified in a wavelet transform process, and an integer wavelet decomposition coefficient in a high frequency sub-band is predicted and compensated by utilizing an integer wavelet decomposition coefficient recovery value in a low frequency sub-band to improve the performance of measuring the decompressed image.

Description

A kind of remote sensing image data compression method that keeps measuring performance

Technical field

The invention belongs to remote sensing image data compression and digital signal processing technique field, relate to the wavelet transformation of high resolution remote sensing image and EZW (Embedded Zerotree Wavelet Coding) compression method towards remote sensing image, relate in particular to the remote sensing image data compression method that keeps measuring performance.

Background technology

Be accompanied by the development of photoelectric technology, computer technology and remote sensing technology, the information of obtaining by remote sensing is more and more is the data that exist in digital form and process, to digital photogrammetry brought more directly, the data source of horn of plenty more.Yet in the development that promotes digital photogrammetry production technology, product quality and quantity, magnanimity remote sensing information data (being mainly remote sensing image data) are had higher requirement to storage and transmission conditions, bring acid test also for existing finite bandwidth and hardware device.The expansion of particularly digital photogrammetry networking, large-scale production; need to store a large amount of raw videos and intermediate imagery data on client station; transmitting a large amount of remote sensing image datas between different clients and between client and server; and existing Computer Storage and transmission technology can't reach this requirement, have directly had influence on the further raising of digital photogrammetry production efficiency.Therefore the remote sensing image compression technology has received increasing concern.

Data compression is exactly the signal that is sent with minimum numerical code expression information source, reduces the signal space hold given information set or data sampling set, and so-called signal space refers to certain signal set shared spatial domain, time domain and frequency space.Image compression is that the image information source is encoded, and is guaranteeing to reach under the prerequisite of expection picture quality, by deletion redundancy or unwanted information, manages to reduce necessary numeric code rate and the compression coding technology taked.Image compression also often is called Image Coding.

The digital image compression coding techniques can be traced back to the TV signal digitlization that proposes in 1948, to today, the history of more than 50 year has been arranged.A variety of image compression encoding methods have appearred during this period, particularly to the later stage eighties, foundation due to wavelet transformation theory, fractal theory, artificial neural network theories and visual simulation theory, Image Compression has obtained unprecedented development, and wherein the Image Compression based on wavelet transformation is one of focus of current research.

Image compression provides effective approach for solving the large problem of DID amount.Image compression can reduce the memory space of image, improves network transmission efficiency, reduces storage and transmission cost, plays conclusive effect for the real-time Transmission of digital picture.

In general, lossy compression method is compared Lossless Compression can obtain larger compression ratio, in multimedia technology commonly used, digital image content mostly is character image, what mostly adopt is lossy compression, namely under the prerequisite that does not produce obvious image vision loss, reduce as much as possible the amount of information of image, to realize the image compression of high compression ratio.Remote sensing images do not compress or adopt lossless compression scheme, thereby can't obtain higher compression ratio.The Remote Sensing Digital Image main cause of not carrying out lossy compression method has before:

1, remote sensing image obtains difficult, existing Aero-Space remote sensing image, all need to drop into very large manpower, financial resources and material resources and set up remotely-sensed data and obtain system, cause the cost of remote sensing image higher, can't accept the information loss in the lossy compression method process of remote sensing image;

2, the precision to remote sensing image processing and application requirements is higher, remote sensing image not only will be used for the observation of human eye, and to for search, identification and the measurement of computer, if adopt general lossy compression method mode, can produce the effect that affects remote sensing image because of the loss of information.

Owing to ensureing in the production and operation of information and product at digital mapping, the compressing data technology has the requirement of this uniqueness, if realize the compression to remote sensing image data, not only have visual effect preferably, the measurement performance of prior maintenance decompress(ion) image is when guaranteeing homework precision, can also reduce the cost that spends in deposit data and transfer aspect, further enhance productivity, particularly for the production of high-timeliness topographic support product, have great significance.

Summary of the invention

The purpose of this invention is to provide a kind of method that keeps measuring the remote sensing image data compression of performance, to adapt to the specification requirement in the fields such as data storage in full digital photogrammetric equipment, the operation of networking full digital photogrammetric and transmission, topographic support.

A kind of data compression method that keeps digital remote sensing image to measure performance, wherein: concrete steps are as follows:

Step 1), gather digital remote sensing image signal, the parameter of statistics sensed image signal: mean value, variance, entropy, average energy, definition, auto-correlation coefficient;

If digital remote sensing image is f (x, y), its pixel f (i, j)=a _k, k={1,2,3 ... L}, i, j are respectively horizontal stroke, the ordinate of pixel, and parameter L is the maximum gradation value in digital remote sensing image;

The probability that any pixel in digital remote sensing image occurs in digital remote sensing image is p (a _k), have:

$\underset{k=1}{\overset{L}{\mathrm{\Σ}}}p\left(a_k\right)=1---\left(1\right)$

Mean value Refer to the average gray value of digital remote sensing image signal, as (2) formula:

$\stackrel{\‾}{a}=\frac{1}{M\×N}\underset{y=1}{\overset{M}{\mathrm{\Σ}}}\underset{x=1}{\overset{N}{\mathrm{\Σ}}}f(x,y)---\left(2\right)$

Variances sigma ²: the variance of the average gray value of digital remote sensing image signal has reflected the discrete case of intensity profile, as (3) formula:

${\mathrm{\σ}}^2=\frac{1}{M\×N}\underset{y=1}{\overset{M}{\mathrm{\Σ}}}\underset{x=1}{\overset{N}{\mathrm{\Σ}}}{(f(x,y)-\stackrel{\‾}{a})}^2---\left(3\right)$

ENERGY E: be called again second moment, the power of representative digit remote sensing images overall intensity, as (4) formula:

$E=\frac{1}{M\×N}\underset{y=1}{\overset{M}{\mathrm{\Σ}}}\underset{x=1}{\overset{N}{\mathrm{\Σ}}}{\left(f(x,y)\right)}^2---\left(4\right)$

Average gradient g: the expressive force of the average gradient of digital remote sensing image reflection digital remote sensing image details, the definition of reflection digital remote sensing image, as (5) formula:

$g=\frac{1}{(M-1)(N-1)}\underset{1}{\overset{(M-1)(N-1)}{\mathrm{\Σ}}}\sqrt{({\left(\frac{\∂f(x,y)}{\∂x}\right)}^2+{\left(\frac{\∂f(x,y)}{\∂y}\right)}^2)/2}---\left(5\right)$

Auto-correlation coefficient R (Δ x, Δ y): the correlation of reflection digital remote sensing image signal, adopt normalized autocorrelation functions, expression formula is as (6) formula:

$R(\mathrm{\Δx},\mathrm{\Δy})=\frac{\underset{y=1}{\overset{M}{\mathrm{\Σ}}}\underset{x=1}{\overset{N}{\mathrm{\Σ}}}[f(x,y)-\stackrel{\‾}{a}][f(x+\mathrm{\Δx},y+\mathrm{\Δy})-\stackrel{\‾}{a}]}{\underset{y=1}{\overset{M}{\mathrm{\Σ}}}\underset{x=1}{\overset{N}{\mathrm{\Σ}}}{[f(x,y)-\stackrel{\‾}{a}]}^2}---\left(6\right)$

Wherein, when Δ x=1 or Δ y=1, the auto-correlation function value between R (Δ x, Δ y) value representation neighbor is referred to as auto-correlation coefficient;

Comentropy H (U): the entrained average information of each pixel in index word remote sensing images, what its physical significance characterized is under the lossless coding condition, the needed code length of each coded identification:

$H\left(U\right)=-\underset{k=1}{\overset{L}{\mathrm{\Σ}}}p\left(a_k\right){\log }_{2}p\left(a_k\right)---\left(7\right)$

Step 2), digital remote sensing image is carried out multistage integer wavelet transform, set up the structure of wavelet tree, determine the integer wavelet decomposition coefficient after the digital remote sensing image wavelet transformation; Integer wavelet transform refers to remain the discrete signal that coefficient of wavelet decomposition be integer of discrete signal through obtaining after wavelet transformation of integer set; The computing formula of integer wavelet decomposition coefficient is:

Wherein,

Expression is to d _{1, l}/ 2 do rounding operation, wherein, and conversion coefficient d _{1, l}And s _{1, l}It is all the integer set;

After digital remote sensing image process integer wavelet transform, a low frequency sub-band and three high-frequency sub-band of press from left to right, order from top to bottom consisting of different resolution; The integer wavelet decomposition coefficient that is in low frequency sub-band is called paternal number, corresponding to the whole integer wavelet decomposition coefficients in the high-frequency sub-band of each direction in space, is called descendants's coefficient of this paternal line number, has consisted of like this three tree structures take paternal number as the summit;

Step 3), carry out Embedded Zerotree Wavelet Coding:

(31), select one group to coefficient of wavelet decomposition X _iCarry out the threshold value T of importance judgement ₀, T ₁T _n-1, one by one to coefficient of wavelet decomposition X _iCarry out the importance judgement, wherein threshold value chooses according to T _i=T _i-1/ 2, and initial threshold threshold value T ₀Satisfy | X _i|＜2T _o

(32), significant coefficient and non-significant coefficient: a given coefficient of wavelet decomposition X _iIf exist for given thresholding T | X _i|＞T, wherein, T ∈ T ₀, T ₁T _n-1, this coefficient of wavelet decomposition is significant coefficient; Otherwise, be non-significant coefficient; In Embedded Zerotree Wavelet Coding, non-significant coefficient is labeled as zero; Corresponding non-significant coefficient is zero node in wavelet tree, and significant coefficient is the non-zero node;

(33), zero tree: zero tree representation is based on the correlation of intersubband coefficient of wavelet decomposition: if the coefficient of wavelet decomposition on low frequency sub-band is non-significant coefficient for thresholding T, with the high-frequency sub-band of corresponding each direction in space of this low frequency sub-band in all coefficient of wavelet decomposition be also non-significant coefficient for this thresholding T, these non-significant coefficients are represented with tree, be zero tree;

(34), zerotree root: the zero node that is positioned at the lowest frequency subband in zero tree is zerotree root, and zerotree root represents with symbols Z TR;

(35), master meter and subtabulation: in cataloged procedure, remaining table two separation, that constantly update: master meter and subtabulation; Master meter is corresponding to unessential set or the coefficient of wavelet decomposition of coding, and subtabulation is the effective information of coding; Significant coefficient in each coefficient of wavelet decomposition is preserved into subtabulation, and non-significant coefficient is preserved into master meter, and sets threshold value T ₀

(36), main scanning and subscan:

Main scanning: to all the coefficient of wavelet decomposition X in master meter _iMake a decision, if | X _i|＞thresholding T _i, this node is significant coefficient; If X _iBe zerotree root, output code ZTR does not scan all descendants's coefficients of zerotree root; If X _iBe isolated point zero point, continue all follow-up coefficient of wavelet decomposition of isolated point zero point are scanned;

Subscan: the coefficient of wavelet decomposition Y in subtabulation _iAll satisfy T _i＜Y _i＜2T _iSubscan is by all coefficient of wavelet decomposition Y in the scanning subtabulation _iValue, export different coding " 0 " or " 1 "; Satisfy T _i＜Y _i＜3T _i/ 2 coefficient of wavelet decomposition Y _iBe non-significant coefficient, 3T is satisfied in output 0 _i/ 2＜Y _i＜2T _iCoefficient of wavelet decomposition Y _iBe significant coefficient, output 1;

Step 4), utilize improved Embedded Zerotree Wavelet Coding to compress processing to digital remote sensing image:

(41), set up zero tree label table ZT, before scanning, all marker bits of zero tree label table ZT are initially set to " 1 "; Each element of coefficient of wavelet decomposition has corresponding marker bit in label table, when a definite element is zerotree root, all descendants's coefficients that will be somebody's turn to do zero tree in label table are labeled as " 0 ", simultaneously, are labeled as " 0 " at the correspondence position of each non-significant coefficient;

When each coefficient of wavelet decomposition of scanning, first check the symbol of this coefficient of wavelet decomposition in zero tree label table ZT, if symbol is " 0 ", this coefficient of wavelet decomposition is descendants's coefficient of zerotree root, this coefficient of wavelet decomposition is not judged, if symbol is 1, scan;

The marker bit of the significant coefficient in this scanning all is set as " 0 ", after this scanning is completed, all symbols of zero tree label table ZT is done the antilogical computing, and the new zero tree label table ZT that obtains this moment is used for scanning next time; In scanning process next time, the significant coefficient that is labeled as " 0 " in this scanning is skipped, do not made a decision;

(42), record the maximum of coefficient of wavelet decomposition in each subband: complete the statistics of extremes of coefficient of wavelet decomposition in the subband of new zero tree label table ZT in this scanning process; Add up the extreme value of coefficient of wavelet decomposition in each subband, be used for determining scanning threshold value T ₀If, the threshold value T of scanning _iGreater than the extreme value of subband coefficient of wavelet decomposition, this subband does not need scanning;

(43), the coefficient of wavelet decomposition in the lowest frequency subband is processed separately: the most energy that include the original figure remote sensing images due to the lowest frequency subband, coefficient of wavelet decomposition in lowest frequency subband after wavelet transformation is encoded separately, do not participate in Embedded Zerotree Wavelet Coding scanning; Remaining high-frequency sub-band after wavelet transformation is formed three tree structures;

(44), change coded identification:

In the successive approximation to quantification process of Embedded Zerotree Wavelet Coding, carry out simultaneously main scanning and subscan; For coefficient of wavelet decomposition Y _iAnd threshold T _iIf, Y _i∈ [T _i+ T _i/ 2,2T _i), output encoder symbol " 111 "; If Y _i∈ [T _i, T _i+ T _i/ 2), output symbol " 110 " is if Y _i∈ [T _i,-T _i-T _i/ 2), output symbol " 100 "; If Y _i∈ [T _i-T _i/ 2,2T _i), output symbol " 101 "; If Y _iBe zerotree root, output symbol " 00 " is if Y _iBe isolated point zero point, output symbol " 01 ";

(45), the integer of coefficient of wavelet decomposition renewal is processed:

In the successive approximation to quantification process of Embedded Zerotree Wavelet Coding, threshold value sequence T ₀, T ₁T _n-1Between relation be T _i=T _i-1/ 2, in the subscan process to significant coefficient and non-significant coefficient more the new work integer process, the significant coefficient Y that namely aligns _iIf, Y _i∈ [T _i, T _i+ T _i/ 2), modification value during renewal

If Y _i∈ [T _i+ T _i/ 2,2T _i), modification value during renewal

If coefficient of wavelet decomposition is integer, work as T _i, 1≤Y must be arranged at=1 o'clock _i＜2, if Y _iBe positive significant coefficient, Y _i=1, the modification coefficient after renewal With T _iAfter=1 threshold value had scanned, all coefficient of wavelet decomposition were all zero, have completed based on the integer wavelet transformation Embedded Zerotree Wavelet Coding.

Described maintenance digital remote sensing image measures the data compression method of performance, wherein: to the lower left corner LH in four subbands after wavelet decomposition _iWhen the coefficient of wavelet decomposition in high-frequency sub-band scans, in accordance with the order from top to bottom by left-to-right order scanning; Coefficient of wavelet decomposition in remaining low frequency sub-band and two high-frequency sub-band is according to from left to right order order scanning from top to bottom.

The present invention adopts technique scheme will reach following technique effect:

Maintenance of the present invention measures the remote sensing image data compression method of performance, has the following advantages:

1), adopt wavelet transformation that remote sensing image is processed, because wavelet transformation has the multiresolution analysis ability, be conducive to selection and the judgement of compression scheme, so technical solution of the present invention is suitable for remote sensing image compression very much;

2), adopt integer wavelet transform to carry out the remote sensing image transition coding, can realize the conversion from integer-to-integer, make its inverse transformation reconstructed image fully, avoided the uncontrollable loss of image information, make the Lossless Image Compression Algorithm based on integer wavelet transform become possibility;

3), upgrade and process coefficient of wavelet decomposition, change the scanning direction to coefficient of wavelet decomposition by setting up zero tree label table, integer, and process separately the improved EZW coding method such as lowest frequency sub-band coefficients, thereby improve compression efficiency and compression ratio;

4), by to the some feature extraction of key character in remote sensing image, and in the wavelet transformation process, characteristic point is identified, utilize integer wavelet low frequency coefficient recovery value high frequency coefficient is predicted and compensated, improve the measurement performance of decompressed image.

Description of drawings

Fig. 1 is the flow chart that maintenance of the present invention measures the remote sensing image data compression method of performance;

Fig. 2 is the tree structure of coefficient of wavelet decomposition;

Fig. 3 is improved EZW method coefficient of wavelet decomposition grid scintigram;

Fig. 4 is the initial carrier remote sensing image;

Fig. 5 is the Y-PSNR contrast figure of image after the different compression ratios of two kinds of images are rebuild;

Fig. 6 is image in Fig. 4 (a) and reconstructed image point characteristic comparison diagram;

Fig. 7 is image in Fig. 4 (b) and reconstructed image point characteristic comparison diagram;

Fig. 8 is the schematic diagram that concerns of validity feature point and compression ratio;

Fig. 9 is the schematic diagram that concerns of validity feature point tolerance statistics and compression ratio;

Figure 10 is the statistical value of two kinds of original images shown in Figure 4;

Figure 11 is standard EZW method and the efficient comparison sheet (unit: size, pixel that improves the compressed encoding of EZW method; Time, second);

Figure 12 is the Y-PSNR comparison sheet (unit: PSNR, dB) of reconstructed image;

The compression ratio comparison sheet of Figure 13 different size (unit: size, pixel).

Embodiment

The invention provides a kind of data compression method that keeps remote sensing image to measure performance, as Fig. 1, concrete steps are as follows:

Step 1), gather the remote sensing image data of different resolution size, the digitlization aviation remote sensing image of the urban area of the actual size 230 * 230mm as shown in (a) figure in Fig. 4, scanning resolution 25 μ m, and, actual size 180 * 180mm shown in (b) figure in Fig. 4, scanning resolution 25 μ m take vegetation as main digitlization aviation remote sensing image.Gather some characteristics commonly used of remote sensing image, such as mean value, variance, entropy, average energy, definition, auto-correlation coefficient and comentropy etc., as shown in figure 10;

If digital remote sensing image is f (x, y), its pixel f (i, j)=a _k, k={1,2,3 ... L}, i, j are respectively horizontal stroke, the ordinate of pixel, and L is the maximum gradation value in image; The probability that any pixel in remote sensing image occurs in remote sensing image is p (a _k), have:

$\underset{k=1}{\overset{L}{\mathrm{\Σ}}}p\left(a_k\right)=1---\left(1\right)$

Average Refer to the average gray value of picture signal, as (2) formula;

$\stackrel{\‾}{a}=\frac{1}{M\×N}\underset{y=1}{\overset{M}{\mathrm{\Σ}}}\underset{x=1}{\overset{N}{\mathrm{\Σ}}}f(x,y)---\left(2\right)$

Variances sigma ²: the variance of remote sensing images gray value has reflected the discrete case of intensity profile, as (3) formula;

${\mathrm{\σ}}^2=\frac{1}{M\×N}\underset{y=1}{\overset{M}{\mathrm{\Σ}}}\underset{x=1}{\overset{N}{\mathrm{\Σ}}}{(f(x,y)-\stackrel{\‾}{a})}^2---\left(3\right)$

ENERGY E: be called again second moment, the power of expression remote sensing images overall intensity also can be to the block image calculating energy, as (4) formula;

$E=\frac{1}{M\×N}\underset{y=1}{\overset{M}{\mathrm{\Σ}}}\underset{x=1}{\overset{N}{\mathrm{\Σ}}}{\left(f(x,y)\right)}^2---\left(4\right)$

Average gradient g: the average gradient of image can reflect the expressive force of image detail, has embodied the whether clear of image, as (5) formula;

$g=\frac{1}{(M-1)(N-1)}\underset{1}{\overset{(M-1)(N-1)}{\mathrm{\Σ}}}\sqrt{({\left(\frac{\∂f(x,y)}{\∂x}\right)}^2+{\left(\frac{\∂f(x,y)}{\∂y}\right)}^2)/2}---\left(5\right)$

Auto-correlation coefficient R (Δ x, Δ y): the correlation of response diagram image signal, relatively commonly used is normalized autocorrelation functions, expression formula is as (6) formula:

$R(\mathrm{\Δx},\mathrm{\Δy})=\frac{\underset{y=1}{\overset{M}{\mathrm{\Σ}}}\underset{x=1}{\overset{N}{\mathrm{\Σ}}}[f(x,y)-\stackrel{\‾}{a}][f(x+\mathrm{\Δx},y+\mathrm{\Δy})-\stackrel{\‾}{a}]}{\underset{y=1}{\overset{M}{\mathrm{\Σ}}}\underset{x=1}{\overset{N}{\mathrm{\Σ}}}{[f(x,y)-\stackrel{\‾}{a}]}^2}---\left(6\right)$

Wherein, when Δ x=1 or Δ y=1, the auto-correlation function value between R (Δ x, Δ y) value representation neighbor is referred to as auto-correlation coefficient;

Comentropy H (U): refer to the entrained average information of each symbol in picture signal, what its physical significance characterized is under the lossless coding condition, the needed code length of each coded identification;

$H\left(U\right)=-\underset{k=1}{\overset{L}{\mathrm{\Σ}}}p\left(a_k\right){\log }_{2}p\left(a_k\right)---\left(7\right)$

Step 2), remote sensing image is carried out multistage integer wavelet transform, integer wavelet transform refers to the discrete signal of integer set, and through after wavelet transformation, what obtain remains the discrete signal that coefficient of wavelet decomposition is integer; The computing formula of integer wavelet decomposition coefficient is:

Wherein

Expression is to d _{1, l}/ 2 do rounding operation, conversion coefficient d _{1, l}And s _{1, l}It is all the integer set;

After remote sensing image decomposes through integer wavelet, by from left to right, from top to bottom order consisted of a low frequency sub-band and three high-frequency sub-band of different resolution; The coefficient of wavelet decomposition that is in low frequency sub-band is called paternal number, corresponding to the whole coefficient of wavelet decomposition in the high-frequency sub-band of each direction in space, the descendants's coefficient that is called this paternal line number has consisted of three tree structures take paternal number as the summit, as shown in Figure 2 like this; Step 3), utilize improved EZW coding to compress processing to remote sensing image:

(31), select one group to coefficient of wavelet decomposition X _iCarry out the threshold value T of importance judgement ₀, T ₁T _n-1, one by one to coefficient of wavelet decomposition X _iCarry out the importance judgement, wherein threshold value chooses according to T _i=T _i-1/ 2, and initial threshold threshold value T ₀Satisfy | X _i|＜2T _o

(32), significant coefficient and non-significant coefficient: a given coefficient of wavelet decomposition X _iIf exist for given thresholding T | X _i|＞T, wherein, T ∈ T ₀, T ₁T _n-1, this coefficient of wavelet decomposition is significant coefficient; Otherwise, be non-significant coefficient; In Embedded Zerotree Wavelet Coding, non-significant coefficient is labeled as zero; Corresponding non-significant coefficient is zero node in wavelet tree, and significant coefficient is the non-zero node;

(33), zero tree: zero tree representation is based on the correlation of intersubband coefficient of wavelet decomposition: if the coefficient of wavelet decomposition on low frequency sub-band is non-significant coefficient for thresholding T, with the high-frequency sub-band of corresponding each direction in space of this low frequency sub-band in all coefficient of wavelet decomposition be also non-significant coefficient for this thresholding T, these non-significant coefficients are represented with tree, be zero tree;

(34), zerotree root: the zero node that is positioned at the lowest frequency subband in zero tree is zerotree root, and zerotree root represents with symbols Z TR;

(35), master meter and subtabulation: in cataloged procedure, remaining table two separation, that constantly update: master meter and subtabulation.Master meter is corresponding to unessential set or the non-significant coefficient of coding, and subtabulation is the effective information of coding.Significant coefficient in each coefficient of wavelet decomposition is preserved into subtabulation, and non-significant coefficient is preserved into master meter, and sets threshold value T ₀

(36), main scanning and subscan:

Main scanning: to all the coefficient of wavelet decomposition X in master meter _iMake a decision, if | X _i|＞thresholding T _i, this node is significant coefficient; If X _iBe zerotree root, output code ZTR does not scan all descendants's coefficients of zerotree root; If X _iBe isolated point zero point, continue all follow-up coefficient of wavelet decomposition of isolated point zero point are scanned;

Subscan: the coefficient of wavelet decomposition Y in subtabulation _iAll satisfy T _i＜Y _i＜2T _iSubscan is by all coefficient of wavelet decomposition Y in the scanning subtabulation _iValue, export different coding " 0 " or " 1 "; Satisfy T _i＜Y _i＜3T _i/ 2 coefficient of wavelet decomposition Y _iBe non-significant coefficient, 3T is satisfied in output 0 _i/ 2＜Y _i＜2T _iCoefficient of wavelet decomposition Y _iBe significant coefficient, output 1.

Step 4), utilize improved Embedded Zerotree Wavelet Coding to compress processing to digital remote sensing image:

(41), set up the label table ZT (Zero Tree) of zero tree:

Namely set up the label table of zero tree; Each element of coefficient of wavelet decomposition has corresponding marker bit in label table, when a definite element is zerotree root, all descendants's coefficients that will be somebody's turn to do zero tree in label table are made " 0 " mark, simultaneously, make " 0 " mark at the correspondence position of each non-significant coefficient;

Before scanning, all marker bits of zero tree label table ZT are initially set to " 1 ", when each coefficient of wavelet decomposition of scanning, first check the symbol of this coefficient of wavelet decomposition in zero tree label table ZT, if symbol is " 0 ", this coefficient of wavelet decomposition is descendants's coefficient of zerotree root, this coefficient of wavelet decomposition is not judged, if symbol is 1, scan;

At this moment, the marker bit of significant coefficient in this scanning all is set as " 0 ", and the significant coefficient that is labeled as " 0 " in scanning process next time in this scanning is skipped, and does not make a decision;

After this scanning is completed, all symbols of zero tree label table ZT are done the antilogical computing, the new zero tree label table ZT that obtains is used for scanning next time;

(42) record the maximum of each subband coefficient of wavelet decomposition: complete the statistics of extremes of coefficient of wavelet decomposition in the subband of new zero tree label table ZT in this scanning process; The extreme value of coefficient of wavelet decomposition in the subband of adding up can be directly used in and determine scanning threshold value T ₀If, the threshold value T of scanning _iGreater than the extreme value of coefficient of wavelet decomposition in subband, this subband does not need scanning;

(43) coefficient of wavelet decomposition in the lowest frequency subband is processed separately: include most energy of original image due to the lowest frequency subband, the coefficient of wavelet decomposition in the lowest frequency subband after wavelet transformation is encoded separately, do not participate in EZW scanning; Remaining high-frequency sub-band can form three tree structures;

(44) change coded identification:

In the successive approximation to quantification process of Embedded Zerotree Wavelet Coding, carry out simultaneously main scanning and subscan; For coefficient of wavelet decomposition Y _iAnd threshold T _iIf, Y _i∈ [T _i+ T _i/ 2,2T _i), output encoder symbol " 111 "; If Y _i∈ [T _i, T _i+ T _i/ 2), output symbol " 110 " is if Y _i∈ [T _i,-T _i-T _i/ 2), output symbol " 100 "; If Y _i∈ [T _i-T _i/ 2,2T _i), output symbol " 101 "; If Y _iBe zerotree root, output symbol " 00 " is if Y _iBe isolated point zero point, output symbol " 01 ";

(45) integer of coefficient of wavelet decomposition renewal is processed:

In the successive approximation to quantification process of Embedded Zerotree Wavelet Coding, threshold value sequence T ₀, T ₁T _n-1Between relation be T _i=T _i-1/ 2, in the subscan process to significant coefficient and non-significant coefficient more the new work integer process, the significant coefficient Y that namely aligns _iIf, Y _i∈ [T _i, T _i+ T _i/ 2), modification value Y during renewal If Y _i∈ [T _i+ T _i/ 2,2T _i), modification value during renewal

If coefficient of wavelet decomposition is integer, work as T _i, 1≤Y must be arranged at=1 o'clock _i＜2, if Y _iBe positive significant coefficient, Y _i=1, the modification coefficient of wavelet decomposition Y after renewal With T _iAfter=1 threshold value had scanned, all coefficient of wavelet decomposition were all zero, have completed based on the integer wavelet transformation Embedded Zerotree Wavelet Coding.

Change the scanning direction of coefficient:

The scan mode of traditional EZW coding to coefficient of wavelet decomposition is to be to scan according to identical direction in each subband, namely is starting point take the upper left corner, according to from left to right by from top to bottom order one one scan; Until all subbands all have been scanned;

Changed the scan mode of coefficient of wavelet decomposition in the present invention, namely to the lower left corner LH in four subbands after wavelet decomposition _iWhen the coefficient of wavelet decomposition in subband scans, seeing Fig. 3, is to adopt in accordance with the order from top to bottom by left-to-right order scanning, and the coefficient of wavelet decomposition in other subbands is still according to from left to right order order scanning from top to bottom; As Fig. 3;

This is because find LH by the analysis to coefficient of wavelet decomposition _iCoefficient of wavelet decomposition horizontal direction correlation in subband is strong, a little less than the vertical direction correlation, at LH _iSubband is in accordance with the order from top to bottom by left-to-right order scanning, taken into full account that in the high-frequency sub-band, coefficient of wavelet decomposition has the characteristics of different correlations in different directions, strengthen the correlation of adjacent scanning result, be conducive to further reject the correlation of coefficient of wavelet decomposition;

In conjunction with the correlation properties of remote sensing image, adopt remote sensing image of the present invention to improve the EZW code compression method and realize digital remote sensing image compression, and comparing aspect the mass effect of the speed of service, reconstructed image with standard EZW coding method;

(1) standard EZW method compares with the compression speed of improving one's methods:

Digitlization aviation remote sensing image radiometric resolution is 8bit; Compress by the image to different size, and the calculation code process time used; Figure 11 is that traditional standard EZW coding method and the present invention improves one's methods to the efficient comparison of different size Remote Sensing Image Compression, as can be seen from Figure 11, traditional standard EZW coding method and the present invention improve one's methods, and be very approaching to the efficient of different size Remote Sensing Image Compression; Be removal system to the randomness impact of compression process, be used alternatingly two kinds of methods the image of same size is carried out repeatedly compression experiment, the compression time to this sized image of single method is got repeatedly the mean value of compression time;

(2) the EZW method compares with the compression effectiveness of improving one's methods:

Experimental data is that in the digitlization aviation remote sensing image, size is the area part of 4096 * 4096 pixels, and radiometric resolution is 8bit; The compression method that uses is method as standard EZW coding method with after improving;

The checking of lossy compression method:

Use two kinds of methods according to different compression ratios, remote sensing images to be compressed respectively, under more identical compression ratio, picture quality after two kinds of methods decompress, Figure 12 is traditional E ZW coding method and the present invention image Y-PSNR table of comparisons of compression after rebuilding of improving one's methods, being found out by Figure 12, is 4: 1 to 64: 1 at compression ratio; The signal to noise ratio that the present invention improves one's methods all greater than traditional E ZW coding method, is 32: 1 o'clock at compression ratio, and the difference of the signal to noise ratio that the present invention improves one's methods and traditional E ZW coding method is for maximum.

To identical raw image data, under different compression ratio conditions, use improve one's methods reconstructed image after compression of the present invention, Y-PSNR (PSNR) is compared and is used traditional E ZW method to exceed 0.2～0.57, has verified the validity of improving one's methods;

The checking of Lossless Compression:

Use two kinds of methods respectively the remote sensing images of different size to be carried out Lossless Compression, the maximum Lossless Compression compression ratio that relatively uses two kinds of methods to obtain is as Figure 13; When the remote sensing images size of obtaining hour, the data volume of remote sensing images is little, the Lossless Compression ability that traditional E ZW method and the present invention improve one's methods is all poor, along with picture size increases, the compression ratio of two kinds of methods all increases to some extent; Characteristics in conjunction with remote sensing images draw the following conclusions:

(1) the remote sensing images details is many, contain much information, view data hour, correlation also a little less than; Be unfavorable for Lossless Compression;

(2) size of image is larger, and the redundancy of image information is more, and the compression ratio of image is just larger;

(3) the present invention's compression ratio of improving one's methods is better than the EZW coding method of standard on the whole, and picture size is larger, and superiority is more obvious.

Utilize some feature extraction operator (as Moravec operator, Forstner operator, Harris operator) to carry out a feature extraction to remote sensing image, the characteristic point of extracting is recorded its positional information, according to the extent of damage to picture quality and information before and after recently assessment compression of characteristic point;

The Harris operator is subjected to the inspiration of auto-correlation function in the signal processing, provides the matrix M that interrelates with auto-correlation function; The Harris method thinks, characteristic point is pixel corresponding to very big interest value in subrange; Therefore, after the interest value of each point, extract the point of all local interest value maximums from original image in having calculated image; In practical operation, certain neighborhood that can selected pixels, i.e. a certain size window was if the interest value of this pixel greater than the interest value of other pixel in window, both can judge that this pixel was characteristic point;

Amplitude and the regularity of distribution of the integer wavelet decomposition coefficient on the different resolution layer of calculated characteristics point position; And according to amplitude and the regularity of distribution of characteristic point position information or coefficient of wavelet decomposition, determine the importance of coefficient of wavelet decomposition in 9 * 9 pixel subranges centered by characteristic point; By the statistics to each frequency band integer wavelet decomposition coefficient importance after the remote sensing image integer wavelet transform, in the wavelet transformation process, characteristic point being identified, thereby make decompress(ion) image feature point have better measurement performance;

Adopt the integer wavelet inverse transformation image that is restored, the integer wavelet inverse transformation process is:

The impact of the interpretation of result image compression such as the quantity by the comparison point feature extraction, error on image feature;

To the aviation image scan digitizing, the city part that to choose respectively two sizes from original two kinds of remote sensing images of Fig. 4 be 1024 * 1024 pixels and suburb part are as experimental data according to resolution 256 gray scales of 25 μ m;

(1), adopt formula (10), (11), (12) to calculate two kinds of remote sensing images at the Y-PSNR of the reconstructed image of different compression ratios; Fig. 5 is the Y-PSNR contrast figure of image after the different compression ratios of two kinds of remote sensing images are rebuild;

Mean square error MSE:

$\mathrm{MSE}={\mathrm{\σ}}_e^2=\frac{1}{\mathrm{MN}}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{[S(i,j)-S^{\′}(x,y)]}^2---\left(10\right)$

Wherein, the picture size that M and N calculate take pixel as unit respectively, S (i, j) and S ' (x, y) are respectively the gray value that original image and reconstructed image are located at pixel (i, j);

Signal to noise ratio snr:

$\mathrm{SNR}=10\lg \frac{{\mathrm{\σ}}_s^2}{{\mathrm{\σ}}_e^2}\left(\mathrm{dB}\right)---\left(11\right)$

Y-PSNR PSNR:

$\mathrm{PSNR}=10\lg \frac{S_{p-\mathrm{<figure-callout\; id="2"\; label="p"\; filenames="HSB00000524400300032.png,HSB00000524400300033.png"\; state="\{\{state\}\}">p</figure-callout>}}^2}{{\mathrm{\&sigma;}}_e^2}\left(\mathrm{dB}\right)---\left(12\right)$

Wherein

Be the average power of original image,

Peak value for primary signal.

(2), image compression impact that a feature extraction is counted

Use improved EZW coding method as the tool of compression of remote sensing images, choose the Harris operator as a feature extraction operator; The calculation window of Harris operator is 5 * 5 pixels, the σ of Gauss's template=0.8, and the inhibition window of maximum is 15 * 15 pixels, threshold value T=5 * 10 ⁷Use the Harris operator to extract respectively original image and the some feature of the rear reconstructed image that decompresses, the quantity of measuring point feature and positional information are analyzed experimental result;

Take characteristic point quantity that raw video was extracted and position as standard, make comparisons with characteristic point and the standard extracted on reconstructed image after decompressing; The reconstructed image characteristic point that wherein relatively has a same position with the original image characteristic point is called identical point, and the reconstructed image characteristic point that can not find character pair point in original image is called erroneous point, and wherein error in point measurement is apart from s _i, referring to the characteristic point extracted on reconstructed image with respect to the error amount of the characteristic point of extracting on original image, unit is pixel; As (13) formula:

$s_i=\sqrt{{(x_o-x_r)}^2+{(y_o-y_r)}^2}---\left(13\right)$

If s _i＞5 pixels are designated as the error characteristic point of reconstructed image, and as Fig. 6 and Fig. 7, Fig. 6 is image in Fig. 4 (a) and reconstructed image point characteristic comparison diagram; Fig. 7 is image in Fig. 4 (b) and reconstructed image point characteristic comparison diagram.

(3), the impact of image compression on the available point of extraction

With respect to quantity and the position of the characteristic point of extracting on original image, if the characteristic point P that extracts on reconstructed image _rWith certain characteristic point P on original image _oError distance s _i≤ 5 pixels can be thought P _rBe the validity feature point; Fig. 8 is that the validity feature of reconstructed image is counted and the ratio of characteristic point sum and the graph of a relation of compression ratio, and Fig. 9 adds up the validity feature point, the error distance average that draws and the relation of variance and original image; Analytic explanation: when compression ratio increases, not only have influence on the characteristic information of original image, the error character of introducing the noise performance has also reduced; During less than 10: 1, average and the variance of reconstructed image available point error have obvious increase when compression ratio, and compression ratio is when larger, average and variance significantly do not change, illustrate that compression ratio is increased to a certain degree, the impact that the characteristic information of available point is subjected to diminishes, and error tends towards stability.

The present invention is on the basis of the standard of improvement EZW coding deficiency, for the large format remote sensing image data, propose a kind of improved EZW coding method, and in conjunction with the some feature extraction to remote sensing image, realized a kind of remote sensing image data compression method of maintenance dose measurement performance; Through practical application, after decompressing, image has and measures preferably accuracy of measurement; The inventive method is used for production and the operation that digital mapping ensures information and product, when guaranteeing homework precision, further enhances productivity, and has practical value preferably.

Claims (4)

1. data compression method that keeps digital remote sensing image to measure performance, it is characterized in that: concrete steps are as follows:

Step 1), collection digital remote sensing image signal, the parameter of statistics sensed image signal: mean value, variance, entropy, average energy, definition, auto-correlation coefficient;

Step 2), digital remote sensing image is carried out multistage integer wavelet transform, set up the structure of wavelet tree, determine the integer wavelet decomposition coefficient after the digital remote sensing image wavelet transformation;

After digital remote sensing image process integer wavelet transform, a low frequency sub-band and three high-frequency sub-band of press from left to right, order from top to bottom consisting of different resolution; The integer wavelet decomposition coefficient that is in low frequency sub-band is called paternal number, corresponding to the whole integer wavelet decomposition coefficients in the high-frequency sub-band of each direction in space, is called descendants's coefficient of this paternal line number, has consisted of like this three tree structures take paternal number as the summit;

Step 3), carry out Embedded Zerotree Wavelet Coding:

Step 4), utilize improved Embedded Zerotree Wavelet Coding to compress processing to digital remote sensing image;

Utilize improved Embedded Zerotree Wavelet Coding to compress the concrete steps of processing to digital remote sensing image in described step 4) as follows:

(41), set up zero tree label table ZT, before scanning, all marker bits of zero tree label table ZT are initially set to " 1 "; Each element of coefficient of wavelet decomposition has corresponding marker bit in label table, when a definite element is zerotree root, all descendants's coefficients that will be somebody's turn to do zero tree in label table are labeled as " 0 ", simultaneously, are labeled as " 0 " at the correspondence position of each non-significant coefficient;

When each coefficient of wavelet decomposition of scanning, first check the symbol of this coefficient of wavelet decomposition in zero tree label table ZT, if symbol is " 0 ", this coefficient of wavelet decomposition is descendants's coefficient of zerotree root, this coefficient of wavelet decomposition is not judged, if symbol is 1, scan;

The marker bit of the significant coefficient in this scanning all is set as " 0 ", after this scanning is completed, all symbols of zero tree label table ZT is done the antilogical computing, and the new zero tree label table ZT that obtains this moment is used for scanning next time; In scanning process next time, the significant coefficient that is labeled as " 0 " in this scanning is skipped, do not made a decision;

(42), record the maximum of coefficient of wavelet decomposition in each subband: complete the statistics of extremes of coefficient of wavelet decomposition in the subband of new zero tree label table ZT in this scanning process; Add up the extreme value of coefficient of wavelet decomposition in each subband, be used for determining scanning threshold value T ₀If, the threshold value T of scanning _iGreater than the extreme value of subband coefficient of wavelet decomposition, this subband does not need scanning;

(43), the coefficient of wavelet decomposition in the lowest frequency subband is processed separately: the most energy that include the original figure remote sensing images due to the lowest frequency subband, coefficient of wavelet decomposition in lowest frequency subband after wavelet transformation is encoded separately, do not participate in Embedded Zerotree Wavelet Coding scanning; Remaining high-frequency sub-band after wavelet transformation is formed three tree structures;

(44), change coded identification:

In the successive approximation to quantification process of Embedded Zerotree Wavelet Coding, carry out simultaneously main scanning and subscan; For coefficient of wavelet decomposition Y _iAnd threshold T _iIf, Y _i∈ [T _i+ T _i/ 2,2T _i), output encoder symbol " 111 "; If Y _i∈ [T _i, T _i+ T _i/ 2), output symbol " 110 " is if Y _i∈ [T _i,-T _i-T _i/ 2), output symbol " 100 "; If Y _i∈ [T _i-T _i/ 2,2T _i), output symbol " 101 "; If Y _iBe zerotree root, output symbol " 00 " is if Y _iBe isolated point zero point, output symbol " 01 ";

(45), the integer of coefficient of wavelet decomposition renewal is processed:

In the successive approximation to quantification process of Embedded Zerotree Wavelet Coding, threshold value sequence T ₀, T ₁T _n-1Between relation be T _i=T _i-1/ 2, in the subscan process to significant coefficient and non-significant coefficient more the new work integer process, the significant coefficient Y that namely aligns _iIf, T _i∈ [T _i, T _i+ T _i/ 2), modification value during renewal

If Y _i∈ [T _i+ T _i/ 2,2T _i), modification value during renewal

If coefficient of wavelet decomposition is integer, work as T _i, 1≤Y must be arranged at=1 o'clock _i＜2, if Yi is positive significant coefficient, Y _i=1, the modification coefficient after renewal

With T _iAfter=1 threshold value had scanned, all coefficient of wavelet decomposition were all zero, complete based on the integer wavelet transformation Embedded Zerotree Wavelet Coding.

2. maintenance digital remote sensing image as claimed in claim 1 measures the data compression method of performance, it is characterized in that: to the lower left corner LH in four subbands after wavelet decomposition _iWhen the coefficient of wavelet decomposition in high-frequency sub-band scans, in accordance with the order from top to bottom by left-to-right order scanning; Coefficient of wavelet decomposition in remaining low frequency sub-band and two high-frequency sub-band is according to from left to right order order scanning from top to bottom.

3. maintenance digital remote sensing image as claimed in claim 1 or 2 measures the data compression method of performance, it is characterized in that: in step 1), specifically being calculated as follows of described mean value, variance, entropy, average energy, definition, auto-correlation coefficient: establishing digital remote sensing image is f (x, y), its pixel f (i, j)=a _k, k={1,2,3 ... L), i, j are respectively horizontal stroke, the ordinate of pixel, a _LBe the maximum gradation value in digital remote sensing image;

The probability that any pixel in digital remote sensing image occurs in digital remote sensing image is p (a _k), have:

Mean value

Refer to the average gray value of digital remote sensing image signal, as (2) formula:

Variances sigma ²: the variance of the average gray value of digital remote sensing image signal has reflected the discrete case of intensity profile, as (3) formula:

ENERGY E: be called again second moment, the power of representative digit remote sensing images overall intensity, as (4) formula:

Average gradient g: the expressive force of the average gradient of digital remote sensing image reflection digital remote sensing image details, the definition of reflection digital remote sensing image, as (5) formula:

Auto-correlation coefficient R (Δ x, Δ y): the correlation of reflection digital remote sensing image signal, adopt normalized autocorrelation functions, expression formula is as (6) formula:

Wherein, when Δ x=1 or Δ y=1, the auto-correlation function value between R (Δ x, Δ y) value representation neighbor is referred to as auto-correlation coefficient;

Comentropy H (U): the entrained average information of each pixel in index word remote sensing images, what its physical significance characterized is under the lossless coding condition, the needed code length of each coded identification:

。

4. maintenance digital remote sensing image as claimed in claim 3 measures the data compression method of performance, it is characterized in that: step 2) described integer wavelet transform refers to remain the discrete signal that coefficient of wavelet decomposition be integer of discrete signal through obtaining after wavelet transformation of integer set;

The computing formula of integer wavelet decomposition coefficient is:

5. maintenance digital remote sensing image as claimed in claim 4 measures the data compression method of performance, and it is characterized in that: the concrete steps of carrying out Embedded Zerotree Wavelet Coding in described step 3) are as follows:

(31), select one group to coefficient of wavelet decomposition X _iCarry out the threshold value T of importance judgement ₀, T ₁T _n-1, one by one to coefficient of wavelet decomposition X _iCarry out the importance judgement, wherein threshold value chooses according to T _i=T _i-1/ 2, and initial threshold threshold value T ₀Satisfy | X _i|＜2T ₀

(32), significant coefficient and non-significant coefficient: a given coefficient of wavelet decomposition X _iIf exist for given thresholding T | X _i|＞T, wherein, T ∈ T ₀, T ₁T _n-1, judge that this coefficient of wavelet decomposition is significant coefficient, otherwise, be non-significant coefficient for judging this coefficient of wavelet decomposition; In Embedded Zerotree Wavelet Coding, non-significant coefficient is labeled as zero; Corresponding non-significant coefficient is zero node in wavelet tree, and significant coefficient is the non-zero node;

(33), zero tree: zero tree representation is based on the correlation of intersubband coefficient of wavelet decomposition: if the coefficient of wavelet decomposition on low frequency sub-band is non-significant coefficient for thresholding T, with the high-frequency sub-band of corresponding each direction in space of this low frequency sub-band in all coefficient of wavelet decomposition be also non-significant coefficient for this thresholding T, these non-significant coefficients are represented with tree, be zero tree;

(34), zerotree root: the zero node that is positioned at the lowest frequency subband in zero tree is zerotree root, and zerotree root represents with symbols Z TR;

(35), master meter and subtabulation: in cataloged procedure, remaining table two separation, that constantly update: master meter and subtabulation; Master meter is corresponding to unessential set or the coefficient of wavelet decomposition of coding, and subtabulation is the effective information of coding; Significant coefficient in each coefficient of wavelet decomposition is preserved into subtabulation, and non-significant coefficient is preserved into master meter, and sets threshold value T ₀

(36), main scanning and subscan:

Main scanning: to all the coefficient of wavelet decomposition X in master meter _iMake a decision, if | X _i|＞thresholding T _i, this node is significant coefficient; If X _iBe zerotree root, output code ZTR does not scan all descendants's coefficients of zerotree root; If X _iBe isolated point zero point, continue all follow-up coefficient of wavelet decomposition of isolated point zero point are scanned;

Subscan: the coefficient of wavelet decomposition Y in subtabulation _iAll satisfy T _i＜Y _i＜2T _iSubscan is by all coefficient of wavelet decomposition Y in the scanning subtabulation _iValue, export different coding " 0 " or " 1 "; Satisfy T _i＜Y _i＜3T _i/ 2 coefficient of wavelet decomposition Y _iBe non-significant coefficient, 3T is satisfied in output 0 _i/ 2＜Y _i＜2T _iCoefficient of wavelet decomposition Y _iBe significant coefficient, output 1.

Images