Star pattern compression method based on target detection

Abstract

The star map compression method based on target detection belongs to the field of image processing, and solves the problems of low target detection accuracy, and the difficulty in balancing the image compression efficiency and the local detail preservation of the existing star map coding method. The method comprises the following steps: dividing the star map into a plurality of sub-images with the same pixel size as a primary coding unit; using the maximum inter-class variance method to perform adaptive threshold target detection on the plurality of primary coding units; simultaneously counting the number of pixel points in the eight neighborhoods around the detection points which are divided into target regions and background regions; refining the coding unit according to the adaptive threshold target detection result; and using the adaptive quantization parameter method to perform compression coding on the refined coding unit. Through the adaptive threshold target detection, the detection accuracy is improved; under the same compression rate, the local detail preservation effect of the image coded and decoded by the method is better, and the peak signal-to-noise ratio is higher.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically relating to a star map compression method based on target detection. Background Technology

[0002] Image coding, also known as image compression, is a crucial aspect of image processing. It primarily aims to represent an image or its contained information using a smaller number of bits while meeting certain quality requirements (such as signal-to-noise ratio or subjective evaluation scores). Image coding forms the foundation of various image processing and applications; the quality of encoding and decoding directly impacts subsequent image processing and applications. An image coding system typically consists of an encoder and a decoder. After the image enters the encoder, it transforms the input data through mapping quantization and other transformations to output a bitstream file, which is then used to depict the image, forming a compressed image file. During decoding, the decoder uses corresponding inverse quantization and other transformations to decode the encoded bitstream file and reconstruct the image.

[0003] Star map coding methods can be broadly categorized into two types: one type is based on international image and video coding standards, including JPEG, GZIP software, JPEG-2000, and HEVC; the other type is based on non-image and video coding standards such as wavelet transform, KL transform, and compressed sensing. Methods based on international image and video coding standards make targeted improvements based on the characteristics of star map pixels, thus focusing on better utilizing the characteristics of star maps for targeted improvements.

[0004] Star maps contain target regions and a large number of flat background regions. The target region is the main basis for subsequent analysis and processing of the star map, and the flat background regions contain a lot of redundant information. Therefore, during encoding, more attention can be paid to preserving the details of the target region, and more encoding bitstream can be spent on the target region. Commonly used image target detection and segmentation methods include grayscale thresholding and edge detection. Given the large difference in pixel values ​​between the target and background regions in star maps, and the fact that star maps captured by telescopes are all grayscale images, grayscale thresholding can be used for target detection. It is suitable for images where the target and background occupy different grayscale ranges. By selecting an appropriate grayscale threshold and comparing the grayscale of each pixel with the threshold, the two can be separated. Pixels exceeding the threshold are considered elements in the target object set, while pixels below the threshold are considered elements in the background set. The key to grayscale thresholding segmentation methods lies in the selection of the threshold. Global thresholding methods include iterative global thresholding and the maximum inter-class variance method. The iterative global thresholding method divides the entire image into two classes using an initial threshold. It iterates the threshold to the average of the gray values ​​of these two classes until the difference between the iterative thresholds is less than a predetermined difference threshold. This method has significant error; the greater the difference in the number of peaks between the two classes, the greater the deviation, leading to less accurate detection. The maximum inter-class variance method divides the image into foreground and background regions based on its gray-level characteristics, obtaining a threshold that maximizes the inter-class variance between the two regions. This method is sensitive to image noise and tends to classify noisy pixels into the foreground region, causing the calculated threshold to be too high and resulting in detection errors. Summary of the Invention

[0005] To address the problems of low target detection accuracy and difficulty in balancing image compression efficiency with the preservation of local image details in existing star map encoding methods, this invention provides a star map compression method based on target detection.

[0006] The technical solution adopted by this invention to solve the technical problem is as follows:

[0007] The star map compression method based on target detection of the present invention includes the following steps:

[0008] Step 1: Divide the star map into multiple sub-images of the same pixel size as first-level coding units;

[0009] Step 2: Adaptive threshold target detection is performed on multiple first-level coding units using the maximum inter-class variance method; at the same time, the number of pixels in the eight-neighborhood of the detection point that are divided into target region and background region is counted.

[0010] Step 3: Refine the coding units based on the adaptive threshold target detection results;

[0011] Step 4: Compress and encode the refined coding units using an adaptive quantization parameter method.

[0012] Furthermore, the sub-image has a pixel size of 64×64.

[0013] Furthermore, in step two, if more than six pixels around the detection point are classified as background areas, the detection point is determined to be a noise point and classified as a background area.

[0014] Furthermore, in step two, the formula for the Otsu's method of inter-class variance is as follows:

[0015] N = n1 + n2 + ... + n i +...+n L

[0016]

[0017]

[0018]

[0019]

[0020]

[0021]

[0022]

[0023] In the formula, N represents the total number of pixels in the image, n i p represents the number of pixels with gray level i, where i = 1, 2...L, and L represents the maximum gray level of the star map. i Let j represent the normalized gray level and use it as the probability distribution of pixels with gray level i in this image, m represent the number of pixels that are determined to be background regions from the target region after pixel detection, and j represent the normalized gray level. m p represents the grayscale level of the pixels that are determined to be part of the background region from the target region through pixel detection. jm This indicates that the pixels identified as background areas by pixel detection within the target area have a gray level of j. m The pixel probability distribution is given by k, which represents the gray value threshold for dividing the image pixels into background and target regions. ω0 and ω(k) both represent the cumulative probability of pixel values ​​appearing in the background region, ω1 represents the cumulative probability of pixel values ​​appearing in the target region, μ0 represents the average gray value in the background region, μ1 represents the average gray value in the target region, and μ(k) represents the average gray level of pixels in the background region. TThis represents the average gray level of the entire image. This represents the inter-class variance between the background region and the target region.

[0024] Furthermore, in step two, when the inter-class variance reaches its maximum value, the corresponding grayscale threshold k is considered to be the optimal segmentation threshold.

[0025] Furthermore, in step two, the grayscale threshold k is judged. If k is greater than the set default threshold, it is considered that the first-level coding unit contains a target; otherwise, it is considered that the first-level coding unit does not contain a target.

[0026] Furthermore, the specific operation process of step three is as follows: Based on the adaptive threshold target detection result, if no target is detected in the first-level coding unit, then the first-level coding unit will not be further divided; if a target is detected in the first-level coding unit, then it is determined whether to further divide it based on the rate-distortion cost comparison result; if the sum of the rate-distortion costs corresponding to the optimal prediction schemes of the four second-level coding units after division is not greater than the rate-distortion cost corresponding to the optimal prediction scheme of the first-level coding unit, then it is decided to divide it, that is, the first-level coding unit is further divided into four second-level coding units of 32×32 pixels, and adaptive threshold target detection is performed on these four second-level coding units. Second-level coding units that do not detect targets will not be further divided, and second-level coding units that detect targets will be determined whether to further divide them based on the rate-distortion cost comparison result; if the sum of the rate-distortion costs corresponding to the optimal prediction schemes of the four second-level coding units after division is not greater than the rate-distortion cost corresponding to the optimal prediction scheme of the first-level coding unit, then it is determined whether to further divide it ... greater than the rate-distortion cost corresponding to the optimal prediction scheme of the first-level coding unit, then it is determined whether to further divide it based on the rate-distortion cost comparison result of If the sum of the rate-distortion costs corresponding to the schemes is not greater than the rate-distortion cost corresponding to the optimal prediction scheme of the second-level coding unit, then a partitioning decision is made. That is, the second-level coding unit is further divided into four 16×16 pixel third-level coding units, and adaptive threshold target detection is performed on these four third-level coding units. Third-level coding units that do not detect targets are not further partitioned, and third-level coding units that detect targets are further partitioned based on the rate-distortion cost comparison results. If the sum of the rate-distortion costs corresponding to the optimal prediction schemes of the four fourth-level coding units after partitioning is not greater than the rate-distortion cost corresponding to the optimal prediction scheme of the third-level coding unit, then a partitioning decision is made. That is, the third-level coding unit is further divided into four 8×8 pixel fourth-level coding units. When the coding unit is a fourth-level coding unit, target detection and rate-distortion cost comparison are not performed, and further partitioning of the coding unit is stopped.

[0027] Furthermore, in step four, encoding units containing targets are encoded by setting quantization parameters based on the complexity and activity of their image content; encoding units without targets are encoded by using larger quantization parameters.

[0028] Furthermore, in step four, the formula for the adaptive quantization parameter is as follows:

[0029]

[0030]

[0031]

[0032]

[0033] Q b =Q base +2

[0034] In the formula, a CU σ represents the average activity of the current coding unit. 2 (i) represents the variance of the i-th luma sub-block of the coding unit. The normalized average activity is represented by s, which represents the scaling factor, and a is the average activity. avg Q represents the average activity of all coding units of the same size as the current coding unit in the star diagram. PA Q represents the range of adaptive quantization parameters. base Q represents the basic quantization parameter. T Q represents the adaptive quantization parameter of the coding unit containing the target. b This represents the quantization parameter of the coding unit that does not contain the target.

[0035] The beneficial effects of this invention are:

[0036] This invention starts with target detection, processing the coding units of a star map to facilitate subsequent encoding. By dividing the star map into multiple 64×64 pixel first-level coding units, an improved maximum inter-class variance (MOV) method is used for target detection within these first-level coding units. In the adaptive threshold target detection process, the threshold selection for the input image blocks is adaptively adjusted to address the issue of low accuracy in star map target detection, thus enhancing the accuracy and universality of target detection. After target detection within the coding units, the first-level coding units are further subdivided based on the target detection results, i.e., coding units containing the target are further subdivided. The method selects the optimal partitioning method based on the rate-distortion cost to better handle local image details. Further partitioning stops for coding units that do not contain the target, significantly improving coding efficiency. Subsequently, different coding schemes are applied to different regions. This involves adaptively selecting quantization parameters for the finely partitioned coding units to achieve high-quality compression for coding units containing the target and low-quality compression for those not containing the target. Under the same compression ratio, this invention achieves better local image detail preservation and a higher peak signal-to-noise ratio after encoding and decoding, solving the problem of balancing image compression efficiency with local image detail preservation. Attached Figure Description

[0037] Figure 1This is a flowchart of the star map compression method based on target detection according to the present invention.

[0038] Figure 2 This is the result of adaptive threshold target detection. Figure 2 In the diagram, 'a' represents the original star map encoding unit, 'b' represents the detection result of the existing maximum inter-class variance method, and 'c' represents the detection result of the present invention.

[0039] Figure 3 This is a refined breakdown of the coding unit. Figure 3 In the diagram, 'a' represents the experimental result of coding unit partitioning in the HEVC standard algorithm, and 'b' represents the experimental result of coding unit partitioning in this invention.

[0040] Figure 4 This is a comparison chart of the visual quality of decoded images using different compression methods. Figure 4 In the diagram, a represents the original star map, b represents the decoded image of the orthogonal matching pursuit algorithm, c represents the decoded image of the iterative hard thresholding algorithm, and d represents the decoded image of the present invention.

[0041] Figure 5 A comparison of the visual quality of the target region in images decoded using different compression methods. Figure 5 In the diagram, a represents the original star map, b represents the decoded image using the HEVC standard algorithm, and c represents the decoded image of this invention.

[0042] Figure 6 A comparison of the visual quality of the target region in images decoded using different compression methods. Figure 6 In the diagram, a represents the original star map, b represents the decoded image using the HEVC standard algorithm, and c represents the decoded image of this invention.

[0043] Figure 7 The graph shows the rate-distortion curves of a star map with a pixel size of 256×256 after encoding using the HEVC standard algorithm and the algorithm of this invention. Detailed Implementation

[0044] The present invention will be further described in detail below with reference to the accompanying drawings.

[0045] like Figure 1 As shown, the star map compression method based on target detection of the present invention mainly includes the following steps:

[0046] Step 1: Divide the coding unit

[0047] Different sized coding units have different effects on the coding effect. Larger coding units can greatly improve the coding efficiency in flat areas, while smaller coding units can better handle local details of the image, thus making the prediction of complex areas more accurate. Therefore, in the encoding process of this invention, the input star map is first divided into multiple sub-images with a pixel size of 64×64, which are used as first-level coding units.

[0048] Step 2: Adaptive Threshold Target Detection

[0049] An improved Otsu's inter-class variance (OGV) method is used to perform adaptive threshold target detection on the aforementioned multiple first-level coding units. During the detection process, in addition to simply judging based on the threshold, it is also necessary to count the number of pixels classified as target regions and the number of pixels classified as background regions within the eight-neighborhood of a detection point. If more than six pixels around a detection point are classified as background regions, the detection point is determined to be a noise point and is classified as a background region. The formula for the improved OGV method is as follows:

[0050] N = n1 + n2 + ... + n i +...+n L

[0051]

[0052]

[0053]

[0054]

[0055]

[0056]

[0057]

[0058] In the formula, N represents the total number of pixels in the image, n i p represents the number of pixels with gray level i, where i = 1, 2...L, and L represents the maximum gray level of the star map. i Let j represent the normalized gray level and use it as the probability distribution of pixels with gray level i in this image, m represent the number of pixels that are determined to be background regions from the target region after pixel detection, and j represent the normalized gray level. m p represents the grayscale level of the pixels that are determined to be part of the background region from the target region through pixel detection. jm This indicates that the pixels identified as background areas by pixel detection within the target area have a gray level of j. mThe pixel probability distribution is given by k, which represents the gray value threshold for dividing the image pixels into background and target regions. ω0 and ω(k) both represent the cumulative probability of pixel values ​​appearing in the background region, ω1 represents the cumulative probability of pixel values ​​appearing in the target region, μ0 represents the average gray value in the background region, μ1 represents the average gray value in the target region, and μ(k) represents the average gray level of pixels in the background region. T This represents the average gray level of the entire image. This represents the inter-class variance between the background region and the target region.

[0059] When the inter-class variance reaches its maximum value, the corresponding grayscale threshold k is considered the optimal segmentation threshold. Since adaptive thresholding always calculates a threshold to divide the image into target and background regions, but some first-level coding units are entirely background, the calculated threshold needs to be evaluated to better determine whether a target region is present. If the calculated grayscale threshold k is greater than the default threshold, the first-level coding unit is considered to contain a target; otherwise, it is considered not to contain a target.

[0060] Step 3: Refinement of coding units

[0061] Based on the target detection results of each first-level coding unit in step two, if no target is detected in the first-level coding unit, the first-level coding unit will not be further divided; if a target is detected in the first-level coding unit, the decision on whether to further divide the unit will be made based on the comparison of rate-distortion costs.

[0062] The decision to divide the first-level coding unit into four second-level coding units is based on the relationship between the rate-distortion cost corresponding to the optimal prediction scheme of the first-level coding unit and the sum of the rate-distortion costs corresponding to the optimal prediction schemes of the four second-level coding units after the division. If the sum of the rate-distortion costs corresponding to the optimal prediction schemes of the four second-level coding units after the division is not greater than the rate-distortion cost corresponding to the optimal prediction scheme of the first-level coding unit, then the division is decided, that is, the first-level coding unit is further divided into four second-level coding units of 32×32 pixels in size, and adaptive threshold target detection is performed on these four second-level coding units. Based on the target detection results, the second-level coding units that do not detect targets are not further divided, while the second-level coding units that detect targets are further divided based on the comparison of rate-distortion costs.

[0063] The decision to divide a Level 2 coding unit into four Level 3 coding units is based on the relationship between the rate-distortion cost of the optimal prediction scheme for the Level 2 coding unit and the sum of the rate-distortion costs of the optimal prediction schemes for the four Level 3 coding units after the division. If the sum of the rate-distortion costs of the optimal prediction schemes for the four Level 3 coding units after the division is not greater than the rate-distortion cost of the optimal prediction scheme for the Level 2 coding unit, then the division is decided. This involves further dividing the Level 2 coding unit into four 16×16 pixel Level 3 coding units, and performing adaptive threshold target detection on these four Level 3 coding units. Based on the target detection results, Level 3 coding units that do not detect targets are not further divided, while Level 3 coding units that detect targets are further divided based on the comparison of their rate-distortion costs.

[0064] The decision to divide the Level 3 coding unit into four Level 4 coding units is based on the relationship between the rate-distortion cost corresponding to the optimal prediction scheme of the Level 3 coding unit and the sum of the rate-distortion costs corresponding to the optimal prediction schemes of the four Level 4 coding units after partitioning. If the sum of the rate-distortion costs corresponding to the optimal prediction schemes of the four Level 4 coding units after partitioning is not greater than the rate-distortion cost corresponding to the optimal prediction scheme of the Level 3 coding unit, then partitioning is decided, that is, the Level 3 coding unit is further divided into four 8×8 pixel Level 4 coding units. When the coding unit is a Level 4 coding unit, object detection and rate-distortion cost comparison are not performed, and further partitioning of the coding unit is stopped.

[0065] Step 4: Selection of Adaptive Quantization Parameters

[0066] When compressing the obtained coding units, an adaptive quantization parameter method is used. For coding units containing targets, the quantization parameter is set according to the complexity and activity of their image content; for coding units without targets, a larger quantization parameter is used. The adaptive quantization parameter is calculated using the following formula:

[0067]

[0068]

[0069]

[0070]

[0071] Q b =Q base +2

[0072] In the formula, a CU σ represents the average activity of the current coding unit. 2 (i) represents the variance of the i-th luma sub-block of the coding unit. The normalized average activity is represented by s, which represents the scaling factor, and a is the average activity. avg Q represents the average activity of all coding units of the same size as the current coding unit in the star diagram. PA Q represents the range of adaptive quantization parameters. base Q represents the basic quantization parameter. T Q represents the adaptive quantization parameter of the coding unit containing the target. b This represents the quantization parameter of the coding unit that does not contain the target.

[0073] This invention improves target detection accuracy by using adaptive threshold target detection for star maps to be compressed. Subsequently, based on the target detection results, the coding units are refined and adaptive quantization parameters are selected, solving the problem of balancing image compression efficiency and local image detail preservation. As a result, under the premise of the same compression rate, the local detail preservation effect of the encoded and decoded image is better and the peak signal-to-noise ratio is higher.

[0074] To verify the effectiveness and effect of the star map compression method based on target detection of the present invention, the following experiments were conducted.

[0075] 1. Working conditions

[0076] This experiment uses an Intel Core i5-9400F CPU @ 2.9GHz * 6 processor, a Windows 10 PC, a GeForce GTX 1050Ti graphics card, and C++ as the programming language.

[0077] 2. Experimental Content and Results Analysis

[0078] (1) Start by entering the original star map.

[0079] (2) Encoding unit division.

[0080] (3) Adaptive threshold target detection. The results of adaptive threshold target detection are as follows: Figure 2 As shown, by comparing the original star map, it can be seen that the adaptive threshold target detection method used in this invention has more accurate target detection results.

[0081] (4) Refinement of coding units. The refinement results of coding units are as follows: Figure 3 As shown, when the coding unit contains the target, the present invention selects the optimal partitioning method based on the rate-distortion cost result to achieve better processing of local image details; when the coding unit does not contain the target region, the further partitioning of the coding unit is stopped, which greatly improves the coding efficiency.

[0082] (5) Selection of adaptive quantization parameters

[0083] To better compare the performance of the star map compression method based on target detection proposed in this invention for star map compression coding, a comparative analysis will be conducted from two perspectives: subjective visual quality and peak signal-to-noise ratio (PSNR).

[0084] like Figure 4 As shown, by comparing with the original star map, it can be seen that the decoded image of the present invention has better subjective visual quality than the decoded images of the orthogonal matching pursuit algorithm and the iterative hard thresholding algorithm in the compressed sensing method, and can make the decoded image closely resemble the original star map.

[0085] like Figure 5 and Figure 6 As shown, the compression method of this invention and the HEVC standard algorithm are used to compress and encode the star map with a basic quantization parameter of 33. By comparing the local target region of the decoded image with the original star map, it can be found that this invention can better preserve and reconstruct the detailed information of the target in the star map compared with the HEVC standard algorithm.

[0086] Peak signal-to-noise ratio (PSNR) is an engineering term representing the ratio of the maximum possible power of a signal to the power of destructive noise that affects its representation accuracy. PSNR is commonly expressed in logarithmic decibels (dB), with higher values ​​indicating lower distortion. PSNR is defined using mean square error (MSE), and its calculation formula is as follows:

[0087]

[0088]

[0089] In the formula, image I is the original image, image K is the decoded image, m is the height of the image, n is the width of the image, and MAX is the maximum value. I is the maximum value of a pixel in the image, and MSE is the mean square error.

[0090] like Figure 7 As shown, the peak signal-to-noise ratio of the present invention is higher than that of the HEVC standard algorithm.

[0091] Therefore, through Figures 4 to 7 It can be seen that when performing star map compression coding, the present invention not only has a better objective peak signal-to-noise ratio (PSNR) than other existing algorithms, but also better preserves the detailed information of the target, and is also superior to other existing algorithms in terms of subjective visual quality.

[0092] The experimental results above demonstrate that, for the star map to be compressed, this invention improves detection accuracy through adaptive threshold target detection. Based on the adaptive threshold target detection result, it determines whether the current coding unit contains a target. Subsequently, different coding unit partitioning methods are selected based on whether a target is present, along with adaptive quantization parameters during encoding. This achieves refined partitioning and high-quality compression of coding units containing targets, and coarse partitioning and low-quality compression of coding units not containing targets. Consequently, the local details of the encoded and decoded image are better preserved, and the peak signal-to-noise ratio (PSNR) is higher.

[0093] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A star map compression method based on target detection, characterized in that, Includes the following steps: Step 1: Divide the star map into multiple sub-images of the same pixel size as first-level coding units; Step 2: Adaptive threshold target detection is performed on multiple first-level coding units using the maximum inter-class variance method; at the same time, the number of pixels in the eight-neighborhood of the detection point that are divided into target region and background region is counted. If more than six pixels around a detection point are classified as background areas, then the detection point is determined to be a noise point and classified as a background area. The formula for the Otsu's method is as follows: N=n1+n2+...+n i +...+n L In the formula, N represents the total number of pixels in the image, n i p represents the number of pixels with gray level i, where i = 1, 2...L, and L represents the maximum gray level of the star map. i Let j represent the normalized gray level and use it as the probability distribution of pixels with gray level i in this image, m represent the number of pixels that are determined to be background regions from the target region after pixel detection, and j represent the normalized gray level. m This represents the grayscale level of pixels that, after pixel detection, are determined to be part of the background area from the target area. This indicates that the pixels identified as background areas by pixel detection within the target area have a gray level of j. m The pixel probability distribution is given by k, which represents the gray value threshold for dividing the image pixels into background and target regions. ω0 and ω(k,m) both represent the cumulative probability of pixel values ​​appearing in the background region, ω1 represents the cumulative probability of pixel values ​​appearing in the target region, μ0 represents the average gray value of the background region, μ1 represents the average gray value of the target region, and μ(k,m) represents the average gray value of the pixels in the background region. T This represents the average gray value of the entire image. This represents the inter-class variance between the background region and the target region; Step 3: Refine the coding units based on the adaptive threshold target detection results; Step 4: Compress and encode the refined coding units using an adaptive quantization parameter method.

2. The star map compression method based on target detection according to claim 1, characterized in that, The sub-image has a pixel size of 64×64.

3. The star map compression method based on target detection according to claim 1, characterized in that, In step two, when the inter-class variance reaches its maximum value, the corresponding grayscale threshold k is considered to be the optimal segmentation threshold.

4. The star map compression method based on target detection according to claim 3, characterized in that, In step two, the grayscale threshold k is judged. If k is greater than the set default threshold, it is considered that the first-level coding unit contains a target; otherwise, it is considered that the first-level coding unit does not contain a target.

5. The star map compression method based on target detection according to claim 4, characterized in that, The specific operation process of step three is as follows: Based on the adaptive threshold target detection results, if no target is detected in the first-level coding unit, the first-level coding unit will not be further divided; if a target is detected in the first-level coding unit, it will be determined whether to further divide it based on the comparison result of rate-distortion costs; if the sum of the rate-distortion costs corresponding to the optimal prediction schemes of the four second-level coding units after division is not greater than the rate-distortion cost corresponding to the optimal prediction scheme of the first-level coding unit, then division is decided, that is, the first-level coding unit is further divided into four second-level coding units of 32×32 pixels, and adaptive threshold target detection is performed on these four second-level coding units. Second-level coding units that do not detect targets will not be further divided, and second-level coding units that detect targets will be determined whether to further divide them based on the comparison result of rate-distortion costs; if the sum of the rate-distortion costs corresponding to the optimal prediction schemes of the four second-level coding units after division is not greater than the rate-distortion cost corresponding to the optimal prediction scheme of the first-level coding unit, then division will be decided, that is, the first-level coding unit will be further divided into four second-level coding units of 32×32 pixels, and adaptive threshold target detection will be ... If the sum of the corresponding rate-distortion costs is not greater than the rate-distortion cost corresponding to the optimal prediction scheme of the second-level coding unit, then a partitioning decision is made. That is, the second-level coding unit is further divided into four 16×16 pixel third-level coding units, and adaptive threshold target detection is performed on these four third-level coding units. Third-level coding units that do not detect targets are not further partitioned, and third-level coding units that detect targets are further partitioned based on the comparison result of rate-distortion costs. If the sum of the rate-distortion costs corresponding to the optimal prediction schemes of the four fourth-level coding units after partitioning is not greater than the rate-distortion cost corresponding to the optimal prediction scheme of the third-level coding unit, then a partitioning decision is made. That is, the third-level coding unit is further divided into four 8×8 pixel fourth-level coding units. When the coding unit is a fourth-level coding unit, target detection and rate-distortion cost comparison are not performed, and further partitioning of the coding unit is stopped.

6. The star map compression method based on target detection according to claim 5, characterized in that, In step four, the encoding units containing targets are encoded by setting quantization parameters based on the complexity and activity of their image content; For coding units that do not contain a target, a larger quantization parameter is used for encoding.

7. The star map compression method based on target detection according to claim 6, characterized in that, In step four, the formula for the adaptive quantization parameter is as follows: Q b =Q base +2 In the formula, a CU σ represents the average activity of the current coding unit. 2 (i) represents the variance of the i-th luma sub-block of the coding unit. The normalized average activity is represented by s, which represents the scaling factor, and a is the average activity. avg Q represents the average activity of all coding units of the same size as the current coding unit in the star diagram. PA Q represents the range of adaptive quantization parameters. base Q represents the basic quantization parameter. T Q represents the adaptive quantization parameter of the coding unit containing the target. b This represents the quantization parameter of the coding unit that does not contain the target.

Images