CN101276470A - Remote sensing image compression method capable of protecting destination image information - Google Patents

Abstract

The invention relates to a remote sensing image compression method protecting object image information, belonging to the remote sensing image processing technology field. The method comprises: estimating pixel points and a length-width ratio of an object on an digital image based on the size of the object; extracting all of object images from a background image by an edge extraction method; searching an suspected object in the digital image according to the estimated pixel points of the object; extracting the suspected object image, and performing lossless compression singly on the suspected object image; compressing the suspected object image in a lossless manner; and performing loss compression with large extent on the background image. The method can perform priority protection for the object image information.

Description

A kind of remote sensing image compression method of protecting target image information

Technical field

The invention belongs to technical field of remote sensing image processing, relate generally to a kind of remote sensing image compression method of protecting target image information, mainly the image information of the target of investication in the remote sensing investigation image is protected in the process of compression.

Background technology

Data compression technique can be traced back to nineteen fifty-one Shannon the earliest and propose DPCM (Differential PulseCode Modulation, differential pulse coding) technology, what be applied to Remote Sensing Image Compression the earliest is SPOT-1HRV full-colour image compression in 1980, and it adopts the DPCM data compression technique of 8:6.At present, the compression of remote sensing images mainly concentrates on two aspects: predictive coding and transition coding.

The ratio of compression of predictive coding is very low, generally all in conjunction with entropy coding method; The Huffman coding is a practicality entropy coding the most widely, under the situation that some information source probability distribution equates, the Huffman coding can make compression effectiveness reach optimum by variable-length encoding, but in a lot of actual application, the probability distribution of information source is constantly to change and the unknown, has caused the limit of actual compression effect and ideal theoretical compression ratio often to greatly differ from each other like this.Transition coding has discrete cosine transform (Discrete Cosine Transform is called for short DCT) coded system, discrete Fourier transformation coding, K-L (Karhunen-Loeve) transition coding.The Karhunen-Loeve transformation coding is optimum in theory, ratio of compression is the highest, but operand is big and do not have fast algorithm, and what extensively adopt in the practical application is dct transform, DCT is based on the coding method of piece, but parallel processing is low to request memory, but image will be divided into blockage, blocking artifact just appears when ratio of compression is slightly high, lose image detail, thereby when picture quality was had relatively high expectations, compressibility was limited.According to relevant bibliographical information, relevant studies show that of data compression method of carrying out at the follow-up star 3S2 of SPOT5: the DCT algorithm is difficult to satisfy ratio of compression greater than 4.

The method for compressing image that changes based on small echo begins to rise at present, more effective Wavelet image compression method has following several: a kind of is the embedded zero-tree wavelet coding method (EZW:Embedded Zerotree Wavelet) that was proposed according to the similarity between the wavelet coefficient not at the same level by American scholar Jerome M.Shapiro in 1993, another kind is many thoughts that A.Said in 1996 and W.A.Pearlman draw the zerotree image algorithm, the multistage tree set partitioning algorithm that proposes (being called for short SPIHT), also having a kind of is up-to-date Joint Photographic Experts Group: JPEG2000 (Joint Photographic ExpertsGroup2000), these several algorithms all have simple in structure, support multi code Rate of Chinese character, advantages such as the reconstructed image quality is better, but have the algorithm too much and higher shortcoming of complexity consuming time simultaneously.

Above-mentioned several compaction coding method generally speaking respectively has relative merits, but has a common problem (list of references " based on the research of the remote sensing image compression method of JPEG2000 " (Zhang Baowei for example, Yu Shanshan, Zhang Ye, Harbin Institute of Technology's journal, 2007,39 (3): 420-423); " a kind of embedded image compress technique that changes based on small echo " (Northwestern Polytechnical University, China, patent of invention, 03134421.6,2003,7,23).They all are the compressions of finishing at entire image, and some has the investigation remote sensing images of special requirement, for example marine salvage, and target search etc., the pixel that the image information of the ferret out in the image information often accounts for is fewer, but is of great rarity.

Summary of the invention

For avoiding the defective of prior art; the present invention proposes a kind of method for compressing image of to greatest extent information of target image being laid special stress on protecting; target image information is extracted from background image information; carry out compressed encoding with harmless method separately; and adopt the bigger compression coding mode of ratio of compression to carry out compressed encoding for background image information; can make that like this under bigger ratio of compression condition, the information of target image still can be intact.

The present invention solves above-mentioned technical matters by following technical scheme, and it comprises the steps:

1, estimates target shared pixel and Aspect Ratio on digital picture according to the size of target of investication;

2, utilize the method for edge extracting to extract all subject image from background image;

3, according to the estimation pixel of target, in digital picture, search for suspected target;

4, the suspected target image is extracted, carry out lossless compress separately;

5, the harmless method of suspected target imagery exploitation is compressed;

6, background image is carried out the bigger lossy compression method of ratio of compression.

The present invention has following technique effect:

1, this project has proposed a kind of target of investication to be extracted from background image; adopt harmless method to carry out compressed encoding separately; lay special stress on protecting for the target of investication image information; make in entire image under bigger ratio of compression situation; the target of investication image has very high Y-PSNR, and target image information can be preserved well.

2, because background image separates with the target of investication image and carries out compressed encoding, therefore can adopt the bigger compression method of ratio of compression to carry out, improve the ratio of compression of image for the compression of background image.

3, the compression coding mode of background image is after background image is carried out the small echo variation, to extract the small echo low frequency coefficient, utilizes predictive coding to carry out, and has improved the speed of compression of images.

Description of drawings

The method for compressing image block diagram of Fig. 1 based target image information protection.

Embodiment

Method for compressing image of the present invention as shown in Figure 1, at first, the method by edge extracting extracts all subject image from background image, the size estimation according to target obtains the suspected target image from background image again.At last, adopt different compression methods to compress respectively for suspected target image and background image.Concrete steps are as follows:

1, target image pixel estimation:

According to size, shooting distance and the camera resolution of spot object, estimate the shared pixel number scope of spot;

Suppose that the camera resolution of satellite under certain flying height is 4m, the investigation naval vessel is of a size of: long by 270, wide 55 meters.Then can estimate the shared pixel of target of investication and be roughly 60 * 14.Therefore, the target setting image is: wide shared pixel scope is a 7-30 pixel, considers that the naval vessel can cause the variation of stern sea ripples in traveling process, and therefore the upper range with length suitably increases, and is set at: 30-200 pixel.

2, the remote measurement image is carried out edge extracting:

Remote sensing images are carried out edge extracting, a plurality of imageable target are wherein estimated shared pixel according to marginal information;

The method in common of rim detection is the uncontinuity of sensed luminance value so far.Like this discontinuous detects with single order or second derivative.This paper uses first order derivative to calculate, if with f (x, y) presentation video, then first order derivative be two-dimensional function f (x, gradient y) is defined as vector:

$\▿f=\left[\begin{array}{c}{G}_x\\ G_y\end{array}\right]=\left[\begin{array}{c}\frac{\∂f}{\∂x}\\ \frac{\∂f}{\∂y}\end{array}\right]$

This vectorial amplitude is:

$\mathrm{mag}(\▿f)={[G_x^2+G_y^2]}^{1/2}={[{(\∂f/\∂x)}^2+{(\∂f/\∂y)}^2]}^{1/2}$

Image Edge-Detection obtains by Canny edge detector method, and is specific as follows:

(1) image uses the Gaussian filter that has the specified value deviations to come smoothly, thereby can reduce noise;

(2) calculate partial gradient at the every bit place $g(x,y)={[G_x^2+G_y^2]}^{1/2}$ With edge direction α (x, y)=arctan (G _y/ G _x).G _xAnd G _yObtain by the Sobel method, i.e. G _x=(z ₇+ 2z ₈+ z ₉)-(z ₁+ 2z ₂+ z ₃), G _y=(z ₃+ 2z ₆+ z ₉)-(z ₁+ 2z ₄+ z ₇), marginal point is defined as the local maximum point of its intensity on the gradient direction.

(3) marginal point of determining in (2) bar can cause ridge occurring in the gradient amplitude image.Therefore, algorithm is followed the trail of the top of all ridges, and with all not the pixel at the top of ridge be made as zero so that provide a fine rule in output, just non-maximal value suppresses to handle.The ridge pixel uses two threshold value T1 and T2 to make threshold process, wherein T1＜T2.Value is called strong edge pixel greater than the ridge pixel of T2, and the ridge pixel between T1 and the T2 is called weak edge pixel.

(4) by discontinuous weak pixel is integrated into strong pixel, carry out edge link.

According to the step of Canny edge detector, set T1=0.04, T2=0.15 carries out edge-detected image to ship images;

3, suspected target obtains:

The information of doubtful spot is write down in the comparison of imageable target and spot pixel;

According to the shared pixel number of the target of edge extracting is 19*130, meets the estimated value in the step 1, therefore, is judged as suspected target.Record object image top left corner pixel point position is (89,15), and the suspected target image is extracted separately;

4, the processing of background image:

The position of doubtful spot image is replaced with the background average gray, to reduce the data of background image compression;

5, doubtful spot image is left intact, keeps original image information;

6, background image is carried out wavelet decomposition after, only low-frequency data is carried out predictive coding compression;

Test shows, this method can guarantee under bigger ratio of compression condition that the information of target image still can preserve well.

Claims (1)

1, a kind of remote sensing image compression method of protecting target image information is characterized in that may further comprise the steps:

(1), estimates target shared pixel and Aspect Ratio on digital picture according to the size of target of investication;

(2), utilize the method for edge extracting to extract all subject image from background image;

(3), according to the estimation pixel of target, in digital picture, search for suspected target;

(4), the suspected target image is extracted, carry out lossless compress separately;

(5), the harmless method of suspected target imagery exploitation is compressed;

(6), background image is carried out the bigger lossy compression method of ratio of compression.

Images