CN117409092A - A lossless compression method and device for dual-view remote sensing images based on recursive prediction - Google Patents
A lossless compression method and device for dual-view remote sensing images based on recursive prediction Download PDFInfo
- Publication number
- CN117409092A CN117409092A CN202210799512.3A CN202210799512A CN117409092A CN 117409092 A CN117409092 A CN 117409092A CN 202210799512 A CN202210799512 A CN 202210799512A CN 117409092 A CN117409092 A CN 117409092A
- Authority
- CN
- China
- Prior art keywords
- view
- image
- decoding
- remote sensing
- image block
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
- Compression Of Band Width Or Redundancy In Fax (AREA)
Abstract
The invention discloses a lossless compression method and device for a double-view remote sensing image based on recursive prediction, which simultaneously has space correlation in view angles and inter-view angle correlation from the double-view image, and JPEG2000 only eliminates space redundancy in the image. The invention divides the remote sensing images of view 1 and view 2 into a plurality of image blocks, takes one block in view 2 as a key block and adopts JPEG2000 lossless mode to encode together with view 1, decodes view 2 adopts two-dimensional wavelet transformation to remove space redundancy, generates side information by utilizing correlation among views, predicts the current image block by the decoded image block by adopting a recursion prediction structure by using coefficients obtained by affine transformation matrix and multiple linear regression model, so as to reduce code rate required by channel coding. Compared with JPEG2000, the lossless compression method of the double-view remote sensing image based on the recursion prediction can obtain lower coding bit rate.
Description
Technical Field
The invention relates to an image compression technology, in particular to a lossless compression method and device for a double-view remote sensing image based on recursive prediction, and belongs to the field of image communication.
Background
The process of collecting and processing the information of the electromagnetic wave radiated or reflected by the target ground object by using a sensor mounted on a remote sensing platform to form a remote sensing image is commonly called as a remote sensing technology. At present, the improvement of the imaging capability and the transformation of the imaging mode of the satellite lead to the corresponding rapid increase of the data scale, and a plurality of satellites in many countries are provided with earth observation sensors with different visual angles, so that the data volume of the satellite-borne image is increased. However, satellites have the problems of limited computing power and memory resources, satellite links are noisy, and the generation of large-scale mass image data brings great challenges to the transmission and storage of the data. Therefore, developing an effective dual-view remote sensing image compression algorithm is always a hotspot and difficulty problem in the remote sensing field. Currently, the mainstream compression scheme has two directions: lossless compression and lossy compression. The lossy compression may cause important information loss of the satellite-borne image, which is unfavorable for subsequent research, so that the satellite-borne multi-view image lossless compression algorithm is a hot research direction in the field of remote sensing image compression.
Disclosure of Invention
The invention aims to study lossless compression of remote sensing images, wherein double-view images have strong inter-view correlation, JPEG2000 carries out independent encoding and decoding on each image, only redundancy in the spatial direction in the image is removed, the method adopts two-dimensional wavelet transformation to remove the spatial redundancy, and coefficient generation Side Information (SI) obtained by affine transformation matrix and multiple linear regression is adopted for a recursive prediction structure to remove the inter-view redundancy.
The basic idea of the invention is that the image coding transmission of view 2 is divided into image blocks with the size of M x M, N blocks are used, one block in view 2 is taken as a key block to be coded together with view 1 by adopting a JPEG2000 lossless mode, and the rest part of view 2 adopts a distributed coding scheme. The method specifically mainly comprises the following process steps:
(1) And (5) preprocessing an image. Because the remote sensing image has high resolution and huge picture, the compression coding of the whole image is difficult to directly carry out, and therefore, the image preprocessing is carried out before the compression coding. And dividing the two remote sensing images of view 1 and view 2 into image blocks with the size of M x M, and adding N blocks. The image block for view 1 is denoted as K n ,n∈[1,N]The image block for view 2 is denoted S n ,n∈[1,N]。
(2) Encoding. Taking one block in view 2 as a key block and encoding the key block and view 1 together by adopting a JPEG2000 lossless mode. The encoding of the images of the remaining view 2 is first performed by a two-dimensional integer wavelet transform. The low frequency sub-band after the wavelet transform of the image can be regarded as a low resolution version of the original image, which is of great significance in image compression. Then, the obtained wavelet coefficient is subjected to bit plane decomposition, and bit plane coding is an effective method for progressively compressing images, so that deviation of reconstructed images can be reduced, and quality of the reconstructed images is improved. Finally, the channel coding based on Polar Codes (Polar Codes) is carried out on each bit plane in order from high to low, and the obtained check information is stored in a buffer.
(3) Decoding. And decoding the view 1 by adopting a JPEG2000 lossless mode to obtain a reconstruction value of the view 1 image. Decoding of view 2 image blocks firstly utilizes decoded images to calculate geometric mapping relations among views to generate jointly decoded high-quality side information, other image blocks adopt a recursive prediction structure to predict the current image block by the decoded image blocks, coefficients obtained by affine transformation matrix and multiple linear regression are used for generating jointly decoded high-quality side information, and therefore code rate required by channel decoding is reduced. And secondly, carrying out channel decoding based on the polarization code together with the check information transmitted by the encoding end, and if the decoding is unsuccessful, requesting more check information from the encoding end through a feedback channel until the decoding is successful or the maximum decoding times are reached. And then, carrying out wavelet inverse transformation on the reconstruction information obtained after the decoded bit planes are combined to obtain a reconstruction value of the current image block.
In the technical scheme of the invention, the lossless compression method and the lossless compression device for the double-view remote sensing image based on the recursion prediction are characterized in that the decoding of the rest image blocks in the view angle 2 is assisted by using the side information generated by the recursion prediction structure.
In the technical scheme of the invention, the lossless compression method and the lossless compression device for the double-view remote sensing image based on the recursion prediction are different from JPEG2000, only remove spatial redundancy in the image, and the method utilizes the inter-view correlation of the double-view image and the weighted prediction of the decoded image block to generate high-quality side information for the current image, so that the method can reduce the code rate required by decoding the current image, and can save more code rates under the condition that the decoded image has the same objective quality.
The method can be used for programming and executing the double-view remote sensing image lossless compression method and device based on recursive prediction.
The invention is completed based on the following thought analysis:
and (3) image preprocessing, namely dividing the two remote sensing images with view angles 1 and 2 into image blocks with the size of M, wherein the number of the image blocks is N. One of the view 2 blocks is used as a key block and the view 1 image is used as a key view for coding and transmitting, because although the images shot by the two views have a certain similarity, some differences still exist, and the direct use of the view 1 for side information is not accurate enough, so that the decoding end can calculate the correlation relationship of the two views by using the two views to assist the decoding of the subsequent view 2 image by using the one of the view 2 blocks as the key block (S1) and the view 1 image as the key view.
The side information generation first requires image registration. Image registration is a preprocessing problem in multiple fields of image processing, and when images acquired at different times and under different conditions are researched and processed, registration must be performed first. The images tested in the invention are shot by cameras with different visual angles, and certain deviation exists in imaging angles, so that the two images are aligned by registration, and an affine transformation-based registration algorithm is adopted. Affine transformation refers to two parallel straight lines which are still parallel after image transformation, and only the length and the included angle can be changed, and the transformation is suitable for the situations of inversion, translation, rotation, scaling and the like. Can be expressed by the following formula:
wherein t is 11 、t 21 、t 31 、t 12 、t 22 、t 32 Six parameters representing the transformation, the transformation template, T. Decoding of the remaining image blocks of view 2 utilizes a recursive predictive structure including generation of registration matrices, acquisition of multiple linear regression coefficients, and generation of side information for each image block as described below. Predicting a current image block from the decoded image block to generate a jointly decoded affine transformation matrix T n Specific T n The generation method is as follows, K n ,n∈[1,N]Representing the image block in view 1, S n ,n∈[1,N]Representing the image block in view 2, T a (. Cndot.) is the geometric mapping between the two:
the number of reference image blocks generated by the side information is 1, the distance between the current image block and the reference image block is l (l=1), the prediction coefficient alpha is obtained by a multiple linear regression model,for decoded view 1 tiles corresponding to the reference tiles, the average value is +.> For the average value of the current view 2 image block, the side information is expressed as
At the encoding end, a block in view 2 is taken as a key block and is encoded together with view 1 by adopting a JPEG2000 lossless mode, and the image block encoding of the rest view 2 is firstly subjected to two-dimensional integer wavelet transformation. The low frequency sub-band after the wavelet transform of the image can be regarded as a low resolution version of the original image, which is of great significance in image compression. Then, the obtained wavelet coefficient is subjected to bit plane decomposition, and bit plane coding is an effective method for progressively compressing images, so that deviation of reconstructed images can be reduced, and quality of the reconstructed images is improved. Finally, the channel coding based on Polar Codes (Polar Codes) is carried out on each bit plane in order from high to low, and the obtained check information is stored in a buffer.
At the decoding end, decoding the view 1 by adopting a JPEG2000 lossless mode to obtain a reconstruction value of the view 1 image. Decoding of view 2 image blocks firstly utilizes decoded images to calculate geometric mapping relation among views to generate high-quality side information of joint decoding, and other image blocks adopt a recursive prediction structure to predict the current image block by the decoded image blocks to generate the side information so as to reduce code rate required by channel decoding. And then, the side information and the check information transmitted by the encoding end together perform channel decoding based on the polarization code, and if the decoding is unsuccessful, more check information is requested to the encoding end through a feedback channel until the decoding is successful or the maximum decoding times are reached. And finally, carrying out wavelet inverse transformation on the reconstruction information obtained after the decoded bit planes are combined to obtain a reconstruction value of the current image block.
The resource 3 front and back view images are tested, and experimental results show that compared with a JPEG2000 lossless mode, the lossless compression method and device for the double view remote sensing image based on recursion prediction can obtain lower coding rate.
Drawings
Fig. 1 is a flow chart of the algorithm of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the following examples, which are only for further illustration and are not to be construed as limiting the scope of the present invention, and some insubstantial modifications and adjustments of the present invention should be made by those skilled in the art based on the above disclosure.
The invention relates to a lossless compression method and a lossless compression device for a double-view remote sensing image based on recursion prediction, which are compared with a JPEG2000 standard test model KAKADU8.0 method, and have the following processes:
1. opening KAKADU8.0 algorithm program, performing lossless coding and decoding on the double-view image, and recording the bit rate bpp required by JPEG2000 standard coding;
2. the object in the encoding is a resource No. 3 front-back view image;
3. encoding and decoding a dual view image by using the method of the present invention, and recording a bit per pixel (bpp) and a side information quality PSNR (Peak signal-to-noise ratio) required when side information is encoded using the view 1 directly and the side information production method provided by the method of the present invention is used;
the experimental results are shown in table 1, and it can be seen from the following table that under the condition of similar code rates, for a double-view image scene, the method provided by the invention can obtain a lower coding code rate compared with the JPEG2000 standard, which indicates that the compression performance of the method provided by the invention exceeds the JPEG2000 standard.
TABLE 1 comparison of the experimental results of the inventive method with JPEG2000
| JPEG2000 lossless | Original DVC | The algorithm of the invention | |
| bpp | 5.302152 | 5.17 | 4.72 |
TABLE 2 comparison of the quality of side information in the inventive method
| PSNR/dB | Initial side information | The algorithm of the invention |
| Front-back view angle of resource No. 3 | 20.5280 | 27.0873 |
Claims (4)
1. A lossless compression method and device for a double-view remote sensing image based on recursion prediction are characterized by mainly comprising the following process steps:
(1) Image preprocessing, namely dividing two remote sensing images with view angles 1 and 2 into image blocks with the size of M, and N blocks in total;
(2) Coding, namely taking a block in the view 2 as a key block and adopting a JPEG2000 lossless mode to code together with the view 1, firstly carrying out two-dimensional integer wavelet transform on the image block Codes of the rest view 2, then carrying out bit plane decomposition, carrying out channel coding based on Polar Codes (Polar Codes) on each bit plane in sequence from high to low, and storing the obtained check information into a buffer;
(3) Decoding, namely decoding view 1 by adopting a JPEG2000 lossless mode to obtain a reconstruction value of a view 1 image, firstly solving a geometric mapping relation between views by utilizing a decoded image to obtain high-quality side information of joint decoding, predicting the current image block by using the decoded image block to generate the side information by adopting a recursive prediction structure by the rest image blocks, secondly, carrying out channel decoding based on a polarization code together with check information transmitted by an encoding end, requesting more check information from the encoding end through a feedback channel until the decoding is successful or the maximum decoding times are reached, and carrying out wavelet inverse transformation on the reconstruction information obtained after the bit planes of the decoding are combined to obtain the reconstruction value of the current image block.
2. The lossless compression method of a dual view remote sensing image based on recursive prediction as claimed in claim 1, wherein one of view 2 blocks is used as a key block and view 1 images are used as key views for encoding and transmission, because although images shot by two views have a certain similarity, some differences still exist, and the direct use of view 1 as side information is not accurate enough, so that a decoding end can calculate their correlation relationship by using the images of two views to assist the decoding of the subsequent view 2 images by transmitting one of view 2 blocks.
3. The lossless compression method of a dual view remote sensing image based on recursive prediction as claimed in claim 1, wherein the decoded side information generation of the rest of the image blocks of view angle 2 in step (3) predicts the current image block from the decoded image block using a recursive prediction structure, K n ,n∈[1,N]Representing the image block in view 1, S n ,n∈[1,N]Representing the image block in view 2, T a (. Cndot.) is the geometric mapping relationship between the two, and represents the affine transformation matrix T of joint decoding n The resulting recurrence relation is as follows:
the side information generating module refers to the image block number as 1, the distance between the current image block and the reference image block is l (l=1), the prediction coefficient alpha is obtained through a multiple linear regression model,x,y∈[1,M],n∈[1,N]for decoded view 1 tiles corresponding to the reference tiles, the average value is +.> For the average value of the current view 2 image block, the affine transformation matrix is T n The side information is denoted->
4. A method and apparatus for performing recursive prediction based lossless compression of dual view remote sensing images as claimed in claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210799512.3A CN117409092A (en) | 2022-07-08 | 2022-07-08 | A lossless compression method and device for dual-view remote sensing images based on recursive prediction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210799512.3A CN117409092A (en) | 2022-07-08 | 2022-07-08 | A lossless compression method and device for dual-view remote sensing images based on recursive prediction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117409092A true CN117409092A (en) | 2024-01-16 |
Family
ID=89485796
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210799512.3A Pending CN117409092A (en) | 2022-07-08 | 2022-07-08 | A lossless compression method and device for dual-view remote sensing images based on recursive prediction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117409092A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119444879A (en) * | 2025-01-08 | 2025-02-14 | 浙江大学 | Model data compression and decompression method and device |
-
2022
- 2022-07-08 CN CN202210799512.3A patent/CN117409092A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119444879A (en) * | 2025-01-08 | 2025-02-14 | 浙江大学 | Model data compression and decompression method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113438481B (en) | Training method, image encoding method, image decoding method and device | |
| CN101640803A (en) | Progressive distribution type encoding and decoding method and device for multispectral image | |
| CN102291582A (en) | Distributed video encoding method based on motion compensation refinement | |
| Deng et al. | MASIC: Deep mask stereo image compression | |
| Wang et al. | Uplink-assist downlink remote-sensing image compression via historical referencing | |
| CN118354086A (en) | Unmanned aerial vehicle image processing method and device based on FPGA and computer equipment | |
| CN117409092A (en) | A lossless compression method and device for dual-view remote sensing images based on recursive prediction | |
| CN1632479A (en) | Lossless compression method for high spectrum image based on three-dimensional prediction | |
| KR101455553B1 (en) | Video coding using compressive measurements | |
| CN119211582B (en) | A static image incremental encoding and target area priority transmission method based on JPEG2000 | |
| Yu et al. | Analysis of JPEG2000 compression quality of optical satellite images | |
| Yin et al. | Learned distributed image compression with decoder side information | |
| Solís-Rosas et al. | An enhanced run length encoding using an elegant pairing function for medical image compression | |
| CN104065974A (en) | Image compression method and system | |
| CN115604486A (en) | Video image coding and decoding method and device | |
| CN116934647A (en) | Compressed light field quality enhancement method based on spatial angle deformable convolution network | |
| CN102769751A (en) | Image Coding Method Based on Directional Block Compressive Sensing | |
| CN104079930A (en) | Achieving method of remote-sensing image compression system | |
| Kumar et al. | A comparative case study on compression algorithm for remote sensing images | |
| Zhang et al. | A Low-Complexity Transformer-CNN Hybrid Model Combining Dynamic Attention for Remote Sensing Image Compression. | |
| Fu et al. | Content-Adaptive Residual Learning and Context-Aware Entropy Model for SAR Image Compression | |
| Xu et al. | High performance wavelet-based stereo image coding | |
| CN103402043A (en) | Image compression unit for large visual field TDICCD camera | |
| CN120499380B (en) | Image coding method and system based on hierarchical progressive context model | |
| Tian et al. | Measurement-domain spiral predictive coding for block-based image compressive sensing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |