Satellite image compression quality evaluation method based on feature matching

Abstract

A satellite image compression quality evaluation method based on feature matching can extract and compare key features in a satellite image by introducing a feature matching technology so as to realize accurate evaluation of compressed image quality, and comprises the following steps: firstly, carrying out image preprocessing, extracting key feature points of an original image and a decompressed image; secondly, feature pair matching establishment is carried out; and finally, compression quality evaluation is carried out according to matching index parameters of feature pairs passing matching judgment.

Description

Technical Field

[0001] The invention relates to a satellite image compression quality assessment method based on feature matching, and belongs to the technical field of image processing and compression. Background Art

[0002] With the rapid development of satellite technology, satellite images are increasingly used in scientific research, agriculture, military and other fields. However, due to the huge amount of satellite image data, images usually need to be compressed for effective transmission and storage. During the compression process, how to achieve efficient data compression while ensuring image quality has become an urgent problem to be solved. Traditional image compression quality assessment methods are mostly based on subjective evaluation or simple objective indicators, such as MSE, PSNR, SSIM, etc., which cannot fully and accurately reflect the quality loss after image compression. Summary of the invention

[0003] The technical problem solved by the present invention is: in view of the problem that the traditional technology in the current prior art is difficult to accurately reflect the quality loss after image compression, a satellite image compression quality assessment method based on feature matching is proposed.

[0004] The present invention solves the above technical problems by the following technical solutions:

[0005] A satellite image compression quality assessment method based on feature matching, comprising:

[0006] Collect original satellite image data for image preprocessing, extract features from the preprocessed images, and generate feature vectors based on the extracted key feature point information;

[0007] Calculate the similarity between the feature vectors of key feature points, preset a distance threshold to filter the feature pairs composed of matching key feature point feature vectors, and make matching judgments on each feature pair;

[0008] For feature pairs that pass matching judgment, compression quality is evaluated using matching index parameters.

[0009] The method to obtain key point information and generate feature vectors is:

[0010] Collect the original satellite image data, crop the data image and save it as a specified format image, compress the cropped image to obtain a compressed code stream file;

[0011] After decompressing the compressed code stream file, the image information is restored to generate a decompressed image. The corresponding second-order partial derivatives are calculated according to each pixel of the decompressed image, and the second-order partial derivatives are used to construct the Hessian matrix to describe the local curvature of the decompressed image.

[0012] Detect feature points of original satellite images and decompressed images, construct feature point response pyramids based on feature points detected at different scales, search for feature point response extreme points at each layer of the feature point response pyramid, and use feature point response extreme points as candidate feature points;

[0013] According to the response values ​​of the candidate feature points, all key feature points of the original satellite image and the decompressed image are determined;

[0014] A preset area is selected around each candidate feature point, the Haar wavelet response value of the current area is calculated, the histogram of all Haar wavelet response values ​​is counted, and the feature vector of the key feature point is generated according to the histogram information.

[0015] The method for determining key feature points based on the response values ​​of candidate feature points is:

[0016] For all candidate feature points, the key feature points are screened according to the preset response threshold. If the response value of the candidate feature point is greater than the preset response threshold, the current candidate feature point is the key feature point. Otherwise, it is not a key feature point and the candidate feature point is reselected for comparison.

[0017] The similarity calculation method between the feature vectors of two key feature points is:

[0018] The similarity between two feature vectors is measured by Euclidean distance:

[0019]

[0020] In the formula, Fo i Represents the feature vector of the i-th key feature point in the original image, Fc j Represents the feature vector of the jth key feature point in the compressed image, Fo i,k Represents the k-th dimensional feature vector of the i-th key feature point in the original image, Fc j,k represents the k-th dimensional feature vector of the j-th key feature point in the compressed image, and N=64 represents the dimension of the feature vector.

[0021] The method for selecting matching feature pairs is:

[0022] For the key feature points of each original satellite image, find the key feature points of the compressed image with the highest similarity to form a feature pair;

[0023] For each feature pair, a preset distance threshold is selected. If the Euclidean distance between the two feature vectors of the feature pair is less than the distance threshold, the current feature pair is matched. Otherwise, the current feature pair is not matched, and a new feature pair is selected for comparison.

[0024] The method for matching feature pairs is as follows:

[0025] For each matching feature pair, the Euclidean distance between the key feature points of the original image and the nearest neighbor feature points in the compressed image is calculated. If the ratio of the Euclidean distances is less than the preset ratio threshold, the current matching pair is correct; otherwise, the current matching pair is not valid and a new matching pair is selected for comparison.

[0026] The method for matching feature pairs is as follows:

[0027] For each matching feature pair, a preset pixel deviation threshold is selected. If the pixel deviation between two key feature points is less than the preset pixel deviation threshold, the current matching pair is correct. Otherwise, the current matching pair is not valid and a new matching pair is selected for comparison.

[0028] The method for matching feature pairs is as follows:

[0029] For each matching feature pair, when the preset pixel deviation threshold judgment method and the preset ratio threshold judgment method are satisfied at the same time, the matching pair is completely established; each key feature point in the feature pair comes from the original satellite image and the decompressed image respectively; the key feature points of the decompressed image are not compared and judged.

[0030] The matching index parameters include the matching success rate, matching accuracy and matching distance of the matching pair:

[0031] The matching success rate is the ratio of the number of correctly matched feature pairs to the total number of feature pairs, and is calculated as follows:

[0032]

[0033] Where N sum Represents the total number of feature pairs, N right Indicates the number of correctly matched feature pairs.

[0034] The matching accuracy is the average similarity between correctly matched feature pairs, and is calculated as follows:

[0035]

[0036] In the formula, (Fo i ,Fc j ) represents the correctly matched feature pair, Dis(Fo i ,Fc j ) represents the Euclidean distance of this feature pair.

[0037] The matching distance is the average pixel offset between correctly matched feature pairs, and is calculated as follows:

[0038]

[0039] In the formula, (Fo i ,Fc j ) indicates the correct matching feature pair, Pos(Fo i ,Fc j ) represents the pixel offset of this feature pair, and the calculation formula is as follows:

[0040]

[0041] In the formula, (Fo i,x ,Fo i,y ) represents the horizontal and vertical pixel coordinates of the i-th key feature point in the original image, (Fc j,x ,Fc j,y ) represents the horizontal and vertical pixel coordinates of the jth key feature point in the compressed image.

[0042] The compression quality evaluation is quantitatively expressed according to the quality of the compressed image. The quality of the compressed image is inversely proportional to the matching success rate and the matching accuracy value; the quality of the compressed image is directly proportional to the matching distance value;

[0043] The method for calculating the quality of compressed images is:

[0044] FMCQ=P suc ×(P pre -P dis )

[0045] Where FMCQ represents the satellite image compression quality evaluation index value based on feature matching.

[0046] The advantages of the present invention compared with the prior art are:

[0047] (1) The present invention provides a satellite image compression quality assessment method based on feature matching, proposes an assessment method oriented to feature matching, and proposes a compression algorithm assessment method based on feature matching in response to the actual needs of remote sensing satellite image feature matching. The method can directly assess the impact of the compression algorithm on the key features in the image, thereby ensuring that the compressed image can still maintain a high accuracy during the feature matching process;

[0048] (2) The present invention constructs a multi-dimensional, high-precision evaluation index that not only considers the matching success rate, but also combines the two indicators of matching accuracy and matching distance to form a multi-dimensional, high-precision comprehensive evaluation system that can more accurately reflect the impact of compression on image quality, especially in the detection of subtle differences;

[0049] (3) The present invention has strong optimization guidance and provides a tool for optimizing the application effect of the compression algorithm. Feature matching plays an important role in remote sensing satellite applications such as three-dimensional reconstruction, image fusion, and surface change detection. The evaluation method of the present invention can indirectly reflect the impact of the compression algorithm on the above-mentioned application effects, thereby providing users with ideas for optimizing the compression algorithm in specific application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic diagram comparing the original image and the compressed image provided for the invention;

[0051] Figure 2 A schematic diagram for comparing key feature points extracted from the original image and compressed image provided by the invention;

[0052] Figure 3 A schematic diagram of key feature point matching pairs for matching the original image and the compressed image provided by the invention; DETAILED DESCRIPTION

[0053] A satellite image compression quality assessment method based on feature matching can extract and compare key features in satellite images by introducing feature matching technology, so as to achieve accurate assessment of compressed image quality. First, image preprocessing is performed to extract key feature points of original images and decompressed images, then feature pair matching is established, and finally compression quality assessment is performed based on matching index parameters of feature pairs judged by matching.

[0054] Satellite image compression quality assessment method based on feature matching, the method steps are as follows:

[0055] Collect original satellite image data for image preprocessing, extract features from the preprocessed images, and generate feature vectors based on the extracted key feature point information;

[0056] Calculate the similarity between the feature vectors of key feature points, preset a distance threshold to filter the feature pairs composed of matching key feature point feature vectors, and make matching judgments on each feature pair;

[0057] For feature pairs that pass matching judgment, compression quality is evaluated using matching index parameters.

[0058] The method to obtain key point information and generate feature vectors is:

[0059] Collect the original satellite image data, crop the data image and save it as a specified format image, compress the cropped image to obtain a compressed code stream file;

[0060] After decompressing the compressed code stream file, the image information is restored to generate a decompressed image. The corresponding second-order partial derivatives are calculated according to each pixel of the decompressed image, and the second-order partial derivatives are used to construct the Hessian matrix to describe the local curvature of the decompressed image.

[0061] Detect feature points of original satellite images and decompressed images, construct feature point response pyramids based on feature points detected at different scales, search for feature point response extreme points at each layer of the feature point response pyramid, and use feature point response extreme points as candidate feature points;

[0062] According to the response values ​​of the candidate feature points, all key feature points of the original satellite image and the decompressed image are determined;

[0063] A preset area is selected around each candidate feature point, the Haar wavelet response value of the current area is calculated, the histogram of all Haar wavelet response values ​​is counted, and the feature vector of the key feature point is generated according to the histogram information.

[0064] The method for determining key feature points based on the response values ​​of candidate feature points is:

[0065] For all candidate feature points, the key feature points are screened according to the preset response threshold. If the response value of the candidate feature point is greater than the preset response threshold, the current candidate feature point is the key feature point. Otherwise, it is not a key feature point and the candidate feature point is reselected for comparison.

[0066] The method for screening matching feature pairs and making matching judgments is as follows:

[0067] The similarity between two feature vectors is measured by Euclidean distance;

[0068] For the key feature points of each original satellite image, find the key feature points of the compressed image with the highest similarity to form a feature pair;

[0069] For each feature pair, a preset distance threshold is selected. If the Euclidean distance between the two feature vectors of the feature pair is less than the distance threshold, the current feature pair is matched. Otherwise, the current feature pair is not matched, and a new feature pair is selected for comparison.

[0070] For each matching feature pair, the Euclidean distance between the key feature points of the original image and the nearest neighbor feature points in the compressed image is calculated. If the ratio of the Euclidean distances is less than the preset ratio threshold, the current matching pair is correct; otherwise, the current matching pair is not valid and a new matching pair is selected for comparison.

[0071] For each matching feature pair, a preset pixel deviation threshold is selected. If the pixel deviation between two key feature points is less than the preset pixel deviation threshold, the current matching pair is correct. Otherwise, the current matching pair is not valid and a new matching pair is selected for comparison.

[0072] For each matching feature pair, when the preset pixel deviation threshold judgment method and the preset ratio threshold judgment method are satisfied at the same time, the matching pair is completely established; each key feature point in the feature pair comes from the original satellite image and the decompressed image respectively; the key feature points of the decompressed image are not compared and judged.

[0073] The matching index parameters include the matching success rate, matching accuracy and matching distance of the matching pair:

[0074] The matching success rate is the ratio of the number of correctly matched feature pairs to the total number of feature pairs. The matching accuracy is the average similarity between correctly matched feature pairs. The matching distance is the average pixel offset between correctly matched feature pairs. The compression quality evaluation is quantitatively expressed based on the quality of the compressed image. The quality of the compressed image is inversely proportional to the matching success rate and matching accuracy; the quality of the compressed image is directly proportional to the matching distance.

[0075] The following is further described in conjunction with the accompanying drawings and preferred embodiments of the specification:

[0076] In the current embodiment, a satellite image compression quality assessment method based on feature matching is provided. First, the original and JPEG 2000 compressed satellite images are prepared. Next, the images are preprocessed to restore useful information and enhance detectability. Subsequently, the SURF algorithm is used to extract key features from the image and generate feature vectors. Then, feature matching is performed by calculating the Euclidean distance between the feature vectors, and feature pairs with high similarity are screened out. Finally, the quality of the compressed image is comprehensively evaluated by combining the three indicators of matching success rate, matching accuracy and matching distance, providing guidance for optimizing the satellite image compression algorithm.

[0077] The specific steps of the invention are as follows:

[0078] Step 1: Feature extraction. Perform feature extraction on the original image and the compressed image respectively, and use the SURF algorithm to extract key points such as corner points, edge points, bright spots in dark areas and dark spots in bright areas in the image, and generate feature vectors that reflect the position, direction, scale and other characteristics of the key points, so as to facilitate accurate feature matching in the future. The specific implementation method of this step is as follows:

[0079] 1) Collect data. First, it is necessary to collect original satellite image data. The data used in this paper comes from the xView dataset, which is a set of public remote sensing satellite image datasets, all of which are RGB images collected by the WorldView-3 satellite. Secondly, the images in the xView dataset are cropped into slices of 512*512 size and saved as .png format images to reduce the amount of data processed, unify the image size and format, and improve the evaluation efficiency.

[0080] 2) Image compression. The cropped original image is compressed using the JPEG 2000 algorithm, which reduces the number of bits required for data storage while retaining important information through discrete wavelet transform, quantization, entropy coding and other steps. The image compression ratio is set to 10:1, and the compressed code stream file in .jp2 format is finally output.

[0081] 3) Image decompression. Decompress the compressed code stream to obtain compressed image data, that is, restore the image information through entropy decoding, inverse quantization, inverse transformation and other steps, and generate a decompressed image in .png format. Examples of original images and compressed images are shown below. Figure 1 shown.

[0082] 4) Construct the Hessian matrix. For each pixel in the image, calculate its second-order partial derivative to construct the Hessian matrix, which is used to describe the local curvature of the image.

[0083] 5) Detect scale space extremes. Detect feature points at different scales and construct a multi-scale feature point response pyramid; on each level of the pyramid, search for extreme points of feature point response, which are candidate feature points.

[0084] 6) Select key feature points. Select a suitable response threshold, which is set to 0.5 in the present invention. If the response value of the candidate feature point is greater than this threshold, the current candidate feature point is the final key feature point. The key feature points extracted from the original image and the compressed image are as follows: Figure 2 shown.

[0085] 7) Generate feature vectors. Select an area around each key feature point, calculate the Haar wavelet response of the area, and count the histograms of these responses. Generate a 64-dimensional feature vector based on the histogram information.

[0086] Step 2: Feature matching. Match the key feature points in the original image and the compressed image, use Euclidean distance to calculate the similarity between feature vectors, set the distance threshold to filter out matching feature pairs, and use the nearest neighbor, next nearest neighbor ratio method and pixel deviation to determine whether the matching pair is correct. The specific implementation of this step is as follows:

[0087] 1) Calculate the similarity of feature vectors. Use Euclidean distance to measure the similarity of two feature vectors. Euclidean distance measures the straight-line distance between two points in multidimensional space. The calculation formula is as follows:

[0088]

[0089] Among them, Fo i Represents the feature vector of the i-th final key feature point in the original image, Fc j Represents the feature vector of the jth key feature point in the compressed image, Fo i,k Represents the k-th dimensional feature vector of the i-th key feature point in the original image, Fc j,k The k-th dimensional feature vector of the j-th key feature point in the compressed image is represented, and N=64 represents the dimension of the feature vector. For each key feature point in the original image, the Euclidean distance between it and all key feature point descriptors in the compressed image is calculated and stored in a matrix for subsequent screening.

[0090] 2) Filter matching feature pairs. First, for each key feature point of the original image, find the key feature point of the compressed image with the highest similarity. Then select a suitable distance threshold, which is set to 0.8 in the present invention. If the Euclidean distance between two feature vectors is less than this threshold, they are considered to be a matching feature pair.

[0091] 3) Determine the correctness of feature pair matching. On the one hand, find the nearest neighbor and next nearest neighbor feature points of the key feature points of the original image in the compressed image. If the ratio of the Euclidean distance between the nearest neighbor and the next nearest neighbor is less than a preset ratio threshold, which is set to 0.85 in the present invention, the matching pair is considered correct, reducing false matches caused by background noise or repeated textures. On the other hand, select a suitable pixel deviation threshold, which is set to 0.5 in the present invention. If the pixel deviation between the two feature points is less than this threshold, the matching pair is correct. The key feature points of the original image and the compressed image are matched as follows: Figure 3 shown.

[0092] Step 3: Compression quality evaluation. Calculate the matching success rate, matching accuracy, and matching distance of the key feature points of the original image and the compressed image, and combine the above three indicators to comprehensively evaluate the quality of the compressed image. The specific implementation method of this step is as follows:

[0093] 1) Calculate the matching success rate. The matching success rate refers to the ratio of the number of correctly matched feature pairs to the total number of feature pairs. The calculation formula is as follows:

[0094]

[0095] Among them, N sumRepresents the total number of feature pairs, which is no greater than the number of key feature points in the original image. right Indicates the number of correctly matched feature pairs, which is not greater than the total number of feature pairs. The matching success rate calculated using the above formula is 0.945.

[0096] 2) Calculate the matching accuracy. The matching accuracy refers to the average similarity between correctly matched feature pairs, and the calculation formula is as follows:

[0097]

[0098] Among them, (F i ,Fc j ) represents the correctly matched feature pair, Dis(Fo i ,Fc j ) represents the Euclidean distance of this feature pair. The matching accuracy calculated using the above formula is 0.968.

[0099] 3) Calculate the matching distance. The matching distance refers to the average pixel offset between correctly matched feature pairs, and the calculation formula is as follows:

[0100]

[0101] Among them, (F i ,Fc j ) indicates the correct matching feature pair, Pos(Fo i ,Fc j ) represents the pixel offset of this feature pair, and the calculation formula is as follows:

[0102]

[0103] Among them, (F i,x ,Fo i,y ) represents the horizontal and vertical pixel coordinates of the i-th key feature point in the original image, (Fc j,x ,Fc j,y ) represents the horizontal and vertical pixel coordinates of the jth key feature point in the compressed image. The matching distance calculated using the above formula is 0.089.

[0104] 4) Calculate the quality of the compressed image. The larger the matching success rate and matching accuracy, the smaller the loss of the compressed image and the higher the quality; the smaller the matching distance, the smaller the loss of the compressed image and the higher the quality. Combining the above analysis, we can get the method for evaluating the quality of compressed images, and the calculation formula is as follows:

[0105] FMCQ=P suc ×(P pre -P dis )

[0106] Among them, FMCQ represents the satellite image compression quality evaluation index value based on feature matching, which depends on the three values ​​of matching success rate, matching accuracy and matching distance. The larger the value, the smaller the loss of the compressed image and the higher the quality. The compressed image quality calculated by the above formula is 0.831.

[0107] This embodiment introduces feature matching technology to extract and compare key features in satellite images, thereby achieving accurate evaluation of compressed image quality. This method not only improves the accuracy and objectivity of the evaluation, but also provides an effective tool for optimizing and improving satellite image compression algorithms.

[0108] Although the present invention has been disclosed as above in the form of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art may make possible changes and modifications to the technical solution of the present invention by using the methods and technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope of the technical solution of the present invention.

[0109] The contents not described in detail in the specification of the present invention belong to the common knowledge of those skilled in the art.

Claims

1. A satellite image compression quality assessment method based on feature matching, characterized in that include: Collect original satellite image data for image preprocessing, extract features from the preprocessed images, and generate feature vectors based on the extracted key feature point information; Calculate the similarity between the feature vectors of key feature points, preset a distance threshold to filter the feature pairs composed of matching key feature point feature vectors, and make matching judgments on each feature pair; For feature pairs that pass matching judgment, compression quality is evaluated using matching index parameters.

2. The satellite image compression quality assessment method based on feature matching according to claim 1, characterized in that: The method to obtain key point information and generate feature vectors is: Collect the original satellite image data, crop the data image and save it as a specified format image, compress the cropped image to obtain a compressed code stream file; After decompressing the compressed code stream file, the image information is restored to generate a decompressed image. The corresponding second-order partial derivatives are calculated according to each pixel of the decompressed image, and the second-order partial derivatives are used to construct the Hessian matrix to describe the local curvature of the decompressed image. Detect feature points of original satellite images and decompressed images, construct feature point response pyramids based on feature points detected at different scales, search for feature point response extreme points at each layer of the feature point response pyramid, and use feature point response extreme points as candidate feature points; According to the response values ​​of the candidate feature points, all key feature points of the original satellite image and the decompressed image are determined; A preset area is selected around each candidate feature point, the Haar wavelet response value of the current area is calculated, the histogram of all Haar wavelet response values ​​is counted, and the feature vector of the key feature point is generated according to the histogram information.

3. The satellite image compression quality assessment method based on feature matching according to claim 2, characterized in that: The method for determining key feature points based on the response values ​​of candidate feature points is: For all candidate feature points, the key feature points are screened according to the preset response threshold. If the response value of the candidate feature point is greater than the preset response threshold, the current candidate feature point is the key feature point. Otherwise, it is not a key feature point and the candidate feature point is reselected for comparison.

4. The satellite image compression quality assessment method based on feature matching according to claim 1, characterized in that: The similarity calculation method between the feature vectors of two key feature points is: The similarity between two feature vectors is measured by Euclidean distance: In the formula, Fo i Represents the feature vector of the i-th key feature point in the original image, Fc j Represents the feature vector of the jth key feature point in the compressed image, Fo i,k Represents the k-th dimensional feature vector of the i-th key feature point in the original image, Fc j,k represents the k-th dimensional feature vector of the j-th key feature point in the compressed image, and N=64 represents the dimension of the feature vector.

5. The satellite image compression quality assessment method based on feature matching according to claim 1, characterized in that: The method for selecting matching feature pairs is: For the key feature points of each original satellite image, find the key feature points of the compressed image with the highest similarity to form a feature pair; For each feature pair, a preset distance threshold is selected. If the Euclidean distance between the two feature vectors of the feature pair is less than the distance threshold, the current feature pair is matched. Otherwise, the current feature pair is not matched, and a new feature pair is selected for comparison.

6. The satellite image compression quality assessment method based on feature matching according to claim 1, characterized in that: The method for matching feature pairs is as follows: For each matching feature pair, the Euclidean distance between the key feature points of the original image and the nearest neighbor feature points in the compressed image is calculated. If the ratio of the Euclidean distances is less than the preset ratio threshold, the current matching pair is correct; otherwise, the current matching pair is not valid and a new matching pair is selected for comparison.

7. The satellite image compression quality assessment method based on feature matching according to claim 1, characterized in that: The method for matching feature pairs is as follows: For each matching feature pair, a preset pixel deviation threshold is selected. If the pixel deviation between two key feature points is less than the preset pixel deviation threshold, the current matching pair is correct. Otherwise, the current matching pair is not valid and a new matching pair is selected for comparison.

8. The satellite image compression quality assessment method based on feature matching according to claim 1, characterized in that: The method for matching feature pairs is as follows: For each matching feature pair, when the preset pixel deviation threshold judgment method and the preset ratio threshold judgment method are satisfied at the same time, the matching pair is completely established; each key feature point in the feature pair comes from the original satellite image and the decompressed image respectively; the key feature points of the decompressed image are not compared and judged.

9. The satellite image compression quality assessment method based on feature matching according to claim 1, characterized in that: The matching index parameters include the matching success rate, matching accuracy and matching distance of the matching pair: The matching success rate is the ratio of the number of correctly matched feature pairs to the total number of feature pairs, and is calculated as follows: Where N sum Represents the total number of feature pairs, N right Indicates the number of correctly matched feature pairs.

10. The satellite image compression quality assessment method based on feature matching according to claim 1, characterized in that: The matching accuracy is the average similarity between correctly matched feature pairs, and is calculated as follows: In the formula, (Fo i ,Fc j ) represents the correctly matched feature pair, Dis(Fo i ,Fc j ) represents the Euclidean distance of this feature pair.

11. The satellite image compression quality assessment method based on feature matching according to claim 1, characterized in that: The matching distance is the average pixel offset between correctly matched feature pairs, and is calculated as follows: In the formula, (Fo i ,Fc j ) indicates the correct matching feature pair, Pos(Fo i ,Fc j ) represents the pixel offset of this feature pair, and the calculation formula is as follows: In the formula, (Fo i,x ,Fo i,y ) represents the horizontal and vertical pixel coordinates of the i-th key feature point in the original image, (Fc j,x ,Fc j,y ) represents the horizontal and vertical pixel coordinates of the jth key feature point in the compressed image.

12. The satellite image compression quality assessment method based on feature matching according to claim 1, characterized in that: The compression quality evaluation is quantitatively expressed according to the quality of the compressed image. The quality of the compressed image is inversely proportional to the matching success rate and the matching accuracy value; the quality of the compressed image is directly proportional to the matching distance value; The method for calculating the quality of compressed images is: FMCQ=P suc ×(P pre -P dis ) Where FMCQ represents the satellite image compression quality evaluation index value based on feature matching.

Images