WO2020215433A1

WO2020215433A1 - Image compression method and apparatus

Info

Publication number: WO2020215433A1
Application number: PCT/CN2019/088690
Authority: WO
Inventors: 陈云娜
Original assignee: 深圳市华星光电技术有限公司
Priority date: 2019-04-26
Filing date: 2019-05-28
Publication date: 2020-10-29
Also published as: CN110113609A; CN110113609B; WO2020215433A9

Abstract

The present application uses a block method to respectively reconstruct image blocks and, during the process of reconstructing each image block, on the basis of the magnitude of quantisation parameters in the image block, obtains the two closest mapping matrices and uses the two mapping matrices for reconstructing image blocks to obtain two reconstructed image blocks, the two reconstructed image blocks being fused by means of a linear interpolation method to obtain a final reconstructed image block, and all of the final reconstructed image blocks being combined to obtain a final reconstructed image. The present application solves the block effect caused by image compression and improves the compression distortion caused by the incompatibility of multiple quantisation coefficients in mapping matrices used in the prior art.

Description

Image compression method and device

Technical field

This application relates to an image processing technology, in particular to an image compression method and device.

Background technique

Block discrete cosine transform (BDCT) is widely used in image compression coding such as JPEG, H.264/AVC and H.265/HEVC. This image is compressed and encoded and often quantized after transformation. However, due to the individual transformation and coarse quantization of each image block, this coding technique usually produces blocking effects. In order to ensure image quality, each image block in a frame of image will adopt a different quantization parameter (QP). Sparse representation can be used to reduce blockiness, but the traditional training model of sparse representation is only applicable to one quantization parameter. When the quantization parameter is other values, the processed image cannot get the best effect.

technical problem

The purpose of the present application is to provide an image compression method and device to solve the block effect caused by image compression and improve the compression distortion caused by the incompatibility of multiple quantization coefficients in the prior art using a mapping matrix.

Technical solutions

In order to achieve the foregoing objective, one aspect of the present application provides an image compression method, the method including:

Using one or more processors and a memory storing programs executable by the one or more processors to divide an image into multiple image blocks, wherein the multiple image blocks use different quantization parameters during the compression process ；

According to the quantization parameter corresponding to each image block, reconstruct each image block to obtain the final reconstructed image block; and

Combine the final reconstructed image blocks corresponding to each image block to obtain a compressed image of the image,

The step of reconstructing each image block to obtain the final reconstructed image block includes:

Judging the position of the quantization parameter of the image block in the quantization parameter fitting curve;

If the landing point is between the first predetermined quantization parameter and the second predetermined quantization parameter, the first mapping matrix corresponding to the first predetermined quantization parameter is used to reconstruct the image block to obtain a first reconstructed image block , And reconstructing the image block by using a second mapping matrix corresponding to the second predetermined quantization parameter to obtain a second reconstructed image block; and

Using a linear interpolation method to fuse the first reconstructed image block and the second reconstructed image block to obtain the final reconstructed image block,

After the step of determining the location of the landing point, the method further includes:

If the falling point is smaller than the smallest predetermined quantization parameter, reconstruct the image block by using the mapping matrix corresponding to the smallest predetermined quantization parameter to obtain the final reconstructed image block; and

If the drop point is greater than the maximum predetermined quantization parameter, the mapping matrix corresponding to the maximum predetermined quantization parameter is used to reconstruct the image block to obtain the final reconstructed image block.

In the embodiment of the present application, the quantified parameter fitting curve is obtained through the following steps:

In the training phase, the high-definition training images are compressed with different training quantization parameters to obtain compressed training images, and the compressed training images corresponding to different training quantization parameters are respectively used as output, and the high-definition training images are input into the mapping model To train the mapping model to obtain corresponding mapping matrices under different training quantization parameters;

In the test phase, the high-definition test image is compressed with different quantization parameters for testing to obtain a first compressed test image, and the mapping model is used to map the high-definition test image into a corresponding mapping matrix under different quantization parameters for training. A second compressed test image, finding the best training quantization parameter used by the second compressed test image and its corresponding mapping matrix that have the smallest difference from the first compressed test image under the test quantization parameter; as well as

In the preprocessing stage, a fitting curve between the quantized parameter for testing and the quantized parameter for optimal training is obtained, and a mapping matrix corresponding to the quantized parameter for optimal training is stored.

In the embodiment of the present application, after the step of dividing the image into the multiple image blocks, the method further includes:

Perform block discrete cosine transform on the plurality of image blocks.

Another aspect of the present application provides an image compression method, the method including:

A linear interpolation method is used to fuse the first reconstructed image block and the second reconstructed image block to obtain the final reconstructed image block.

In the embodiment of the present application, after the step of determining the location of the landing point, the method further includes:

If the falling point is smaller than the smallest predetermined quantization parameter, the mapping matrix corresponding to the smallest predetermined quantization parameter is used to reconstruct the image block to obtain the final reconstructed image block.

Perform block discrete cosine transform on the plurality of image blocks.

Another aspect of the present application provides an image compression device, which includes:

One or more processors;

Memory; and

One or more program modules are stored in the memory and can be executed by the one or more processors to implement an image compression method, the method includes:

Dividing an image into multiple image blocks, wherein the multiple image blocks adopt different quantization parameters in the compression process;

If the drop point is greater than the largest predetermined quantization parameter, the mapping matrix corresponding to the largest predetermined quantization parameter is used to reconstruct the image block to obtain the final reconstructed image block.

Perform block discrete cosine transform on the plurality of image blocks.

Beneficial effect

This application uses a block method to reconstruct image blocks separately. In the reconstruction process of each image block, the two closest mapping matrices are obtained according to the size of the quantization parameter in the image block, and the two mapping matrices are used to reconstruct the image block Two reconstructed image blocks are obtained, and the two reconstructed image blocks are merged by linear interpolation to obtain the final reconstructed image block, and all the final reconstructed image blocks are combined to obtain the final reconstructed image. The present application solves the block effect caused by image compression, and improves the compression distortion caused by the incompatibility of multiple quantization coefficients when a mapping matrix is used in the prior art.

Description of the drawings

Fig. 1 shows a flowchart of an image compression method according to the present application.

Fig. 2 shows a schematic diagram of an image divided into a plurality of image blocks according to the present application.

Fig. 3 shows a flow chart of deriving a quantitative parameter fitting curve according to the present application.

Figure 4 shows a schematic diagram of the quantified parameter fitting curve according to the present application.

Fig. 5 shows a flowchart of reconstructing each image block to obtain a final reconstructed image block according to the present application.

Fig. 6 shows a schematic diagram of an image compression device according to the present application.

Embodiments of the invention

In order to make the purpose, technical solutions and effects of this application clearer and clearer, the following further describes this application in detail with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application, and the word "embodiment" used in the specification of the application means serving as an example, example or illustration, and is not intended to limit the application.

Please refer to Figure 1. The image compression method of this application includes the following steps:

Step S10—divide an image into multiple image blocks, wherein the multiple image blocks use different quantization parameters in the compression process;

Step S20—reconstruct each image block according to the quantization parameter (QP) corresponding to each image block to obtain a final reconstructed image block; and

Step S30—combine the final reconstructed image blocks corresponding to each image block to obtain a compressed image of the image.

The image compression method of the present application divides an image into multiple image blocks, and each image block is processed by discrete cosine transform (DCT), and each image block uses different quantization parameters. As shown in Fig. 2, the 16*24 size image is divided into 6 8*8 image blocks, and each image block is compressed using a different quantization parameter (step S10). Each image block is reconstructed according to its corresponding quantization parameter to obtain a final reconstructed image block (step S20), that is, each image block corresponds to a final reconstructed image block after reconstruction. Finally, all the final reconstructed image blocks are combined to obtain the compressed image of the image (step S30), that is, according to the positions of the corresponding image blocks, all the final reconstructed image blocks are stitched and combined to obtain the compressed image. Describe compressed images.

In this application, the quantization parameter fitting curve needs to be referred to in the reconstruction process of each image block. The following describes how to obtain the quantization parameter fitting curve with reference to FIG. 3.

In the training phase, the compressed image and the corresponding high-definition image under different quantization parameters QPtr_n are used for training respectively (step S42) to obtain the corresponding mapping matrix Mtr_n under different quantization parameters QPtr_n (step S44). Specifically, the high-definition training images are compressed with different training quantization parameters QPtr_n to obtain compressed training images, and the compressed training images corresponding to different training quantization parameters QPtr_n are used as outputs, and the high-definition training images are input to the mapping model. To train the mapping model to obtain the corresponding mapping matrix Mtr_n under different training quantization parameters QPtr_n.

In the test phase, the compressed images under different quantization parameters QPte_n are tested using the corresponding mapping matrices Mtr_n obtained by different QPtr_n training (step S46) to obtain the best QPtr_n values that the compressed images under the corresponding quantization parameters QPte_n should take. The mapping matrix Mtr_n (step S48). Specifically, the high-definition test image is compressed with different test quantization parameters QPte_n to obtain the first compressed test image, and the mapping model is used to convert the high-definition test image with the corresponding mapping matrix Mtr_n under different training quantization parameters QPtr_n. Mapping into a second compressed test image, and find the best training used by the second compressed test image and its corresponding mapping matrix Mtr_n that have the smallest difference from the first compressed test image under the test quantization parameter QPte_n Use the quantization parameter QPtr_n.

In the preprocessing stage, the fitting relationship curve between the mapping matrix Mtr_n under the best QPtr_n and the test image QPte_n is obtained (step S50), and t mapping matrices Mtr_1 to Mtr_t are stored (step S52). Specifically, the fitting curve of the test quantization parameter QPte_n and the best training quantization parameter QPtr_n is obtained, and the mapping matrix Mtr_n corresponding to the best training quantization parameter QPtr_n is stored.

The fitting curve of the test quantization parameter QPte_n and the best training quantization parameter QPtr_n obtained according to the above steps is f(QPte_n) shown in FIG. 4. In this example, five mapping matrices corresponding to QPtr_1, QPtr_2, QPtr_3, QPtr_4, and QPtr_5 are stored.

Referring to FIG. 5, the above step of reconstructing each image block to obtain a final reconstructed image block (ie, step 20) includes the following steps:

Step S202 — Determine the position of the quantization parameter of the image block on the quantization parameter fitting curve.

As shown in Figure 4, assuming that the quantization parameter selected for the image block is k, the position of its landing point in the quantization parameter fitting curve f(QPte_n) may be smaller than the smallest predetermined quantization parameter, or between two predetermined quantization parameters Time, or greater than the maximum predetermined quantization parameter.

Step S204 — if the drop point is smaller than the smallest predetermined quantization parameter, use the mapping matrix corresponding to the smallest predetermined quantization parameter to reconstruct the image block to obtain the final reconstructed image block.

In this step, if k<QPte_1, that is, the quantization parameter k is less than the minimum predetermined quantization parameter QPte_1, the mapping matrix Mtr_1 is used to reconstruct the image block, and the final reconstructed image block is directly obtained.

Step S206 — if the landing point is between the first predetermined quantization parameter and the second predetermined quantization parameter, use the first mapping matrix corresponding to the first predetermined quantization parameter to reconstruct the image block to obtain the first reconstructed image block, And using the second mapping matrix corresponding to the second predetermined quantization parameter to reconstruct the image block to obtain the second reconstructed image block; and using the linear interpolation method to fuse the first reconstructed image block and the second reconstructed image block to obtain Finally reconstruct the image block.

In this step, if QPte_1<k<QPte_2, that is, the quantization parameter k is located between the first predetermined quantization parameter QPte_1 and the second predetermined quantization parameter QPte_2, then the first mapping matrix Mtr_1 is used to reconstruct the image block at this time to obtain The first reconstructed image block YQP1 uses the second mapping matrix Mtr_2 to reconstruct the image block to obtain the second reconstructed image block YQP2. Then use linear interpolation to get the final reconstructed image block, the calculation method is as follows:

Step S208 — if the drop point is greater than the maximum predetermined quantization parameter, use the mapping matrix corresponding to the maximum predetermined quantization parameter to reconstruct the image block to obtain the final reconstructed image block.

In this step, if k>QPte_5, that is, the quantization parameter k is greater than the maximum predetermined quantization parameter QPte_5, the mapping matrix Mtr_5 is used to reconstruct the image block, and the final reconstructed image block is directly obtained.

This application uses a block method to reconstruct image blocks separately. In the reconstruction process of each image block, the two closest mapping matrices are obtained according to the size of the quantization parameter in the image block, and the two mapping matrices are used to reconstruct the image block Two reconstructed image blocks are obtained, and the two reconstructed image blocks are merged by linear interpolation to obtain the final reconstructed image block, and all the final reconstructed image blocks are combined to obtain the final reconstructed image. The present application solves the block effect caused by image compression, and improves the compression distortion caused by the incompatibility of multiple quantization coefficients when the prior art adopts a mapping matrix.

As shown in FIG. 6, the present application provides an image processing apparatus 500, which includes one or more processors 501 and a memory 502, the memory 502 is connected to the one or more processors 501, and one or more program modules store It is stored in the memory 502 and can be executed by the one or more processors 501 to implement all or part of the steps in the various methods of the foregoing embodiments. It should be noted that those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above-mentioned embodiments can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium. The medium may include: read only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, etc.

In summary, although this application has been disclosed as above in preferred embodiments, the above preferred embodiments are not intended to limit the application. Those of ordinary skill in the art can make various modifications without departing from the scope of this application. Therefore, the protection scope of this application is subject to the scope defined by the claims.

Claims

An image compression method, characterized in that the method includes:

Using one or more processors and a memory storing programs executable by the one or more processors to divide an image into multiple image blocks, wherein the multiple image blocks use different quantization parameters during the compression process ；

According to the quantization parameter corresponding to each image block, reconstruct each image block to obtain the final reconstructed image block; and

Combine the final reconstructed image blocks corresponding to each image block to obtain a compressed image of the image,

The step of reconstructing each image block to obtain the final reconstructed image block includes:

Judging the position of the quantization parameter of the image block in the quantization parameter fitting curve;

If the landing point is between the first predetermined quantization parameter and the second predetermined quantization parameter, the first mapping matrix corresponding to the first predetermined quantization parameter is used to reconstruct the image block to obtain a first reconstructed image block , And reconstructing the image block by using a second mapping matrix corresponding to the second predetermined quantization parameter to obtain a second reconstructed image block; and

Using a linear interpolation method to fuse the first reconstructed image block and the second reconstructed image block to obtain the final reconstructed image block,

After the step of determining the location of the landing point, the method further includes:

If the falling point is smaller than the smallest predetermined quantization parameter, reconstruct the image block by using the mapping matrix corresponding to the smallest predetermined quantization parameter to obtain the final reconstructed image block; and

If the drop point is greater than the largest predetermined quantization parameter, the mapping matrix corresponding to the largest predetermined quantization parameter is used to reconstruct the image block to obtain the final reconstructed image block.
The image compression method according to claim 1, wherein the quantization parameter fitting curve is obtained through the following steps:

In the training phase, the high-definition training images are compressed with different training quantization parameters to obtain compressed training images, and the compressed training images corresponding to different training quantization parameters are respectively used as output, and the high-definition training images are input into the mapping model To train the mapping model to obtain corresponding mapping matrices under different training quantization parameters;

In the test phase, the high-definition test image is compressed with different quantization parameters for testing to obtain a first compressed test image, and the mapping model is used to map the high-definition test image into a corresponding mapping matrix under different quantization parameters for training. A second compressed test image, finding the best training quantization parameter used by the second compressed test image and its corresponding mapping matrix that have the smallest difference from the first compressed test image under the test quantization parameter; as well as

In the preprocessing stage, a fitting curve between the quantized parameter for testing and the quantized parameter for optimal training is obtained, and a mapping matrix corresponding to the quantized parameter for optimal training is stored.
The image compression method according to claim 1, wherein after the step of dividing the image into the plurality of image blocks, the method further comprises:

Perform block discrete cosine transform on the plurality of image blocks.
An image compression method, characterized in that the method includes:

Using one or more processors and a memory storing programs executable by the one or more processors to divide an image into multiple image blocks, wherein the multiple image blocks use different quantization parameters during the compression process ；

According to the quantization parameter corresponding to each image block, reconstruct each image block to obtain the final reconstructed image block; and

Combine the final reconstructed image blocks corresponding to each image block to obtain a compressed image of the image,

The step of reconstructing each image block to obtain the final reconstructed image block includes:

Judging the position of the quantization parameter of the image block in the quantization parameter fitting curve;

If the landing point is between the first predetermined quantization parameter and the second predetermined quantization parameter, the first mapping matrix corresponding to the first predetermined quantization parameter is used to reconstruct the image block to obtain the first reconstructed image block , And reconstructing the image block by using a second mapping matrix corresponding to the second predetermined quantization parameter to obtain a second reconstructed image block; and

A linear interpolation method is used to fuse the first reconstructed image block and the second reconstructed image block to obtain the final reconstructed image block.
4. The image compression method according to claim 4, wherein after the step of determining the location of the drop point, the method further comprises:

If the falling point is smaller than the smallest predetermined quantization parameter, the mapping matrix corresponding to the smallest predetermined quantization parameter is used to reconstruct the image block to obtain the final reconstructed image block.
4. The image compression method according to claim 4, wherein after the step of determining the location of the drop point, the method further comprises:

If the drop point is greater than the largest predetermined quantization parameter, the mapping matrix corresponding to the largest predetermined quantization parameter is used to reconstruct the image block to obtain the final reconstructed image block.
The image compression method according to claim 4, wherein the quantization parameter fitting curve is obtained through the following steps:

In the training phase, the high-definition training images are compressed with different training quantization parameters to obtain compressed training images, and the compressed training images corresponding to different training quantization parameters are respectively used as output, and the high-definition training images are input into the mapping model To train the mapping model to obtain corresponding mapping matrices under different training quantization parameters;

In the test phase, the high-definition test image is compressed with different quantization parameters for testing to obtain a first compressed test image, and the mapping model is used to map the high-definition test image into a corresponding mapping matrix under different quantization parameters for training. A second compressed test image, finding the best training quantization parameter used by the second compressed test image and its corresponding mapping matrix that have the smallest difference from the first compressed test image under the test quantization parameter; as well as

In the preprocessing stage, a fitting curve between the quantized parameter for testing and the quantized parameter for optimal training is obtained, and a mapping matrix corresponding to the quantized parameter for optimal training is stored.
4. The image compression method according to claim 4, wherein after the step of dividing the image into the plurality of image blocks, the method further comprises:

Perform block discrete cosine transform on the plurality of image blocks.
An image compression device, characterized in that the device includes:

One or more processors;

Memory; and

One or more program modules are stored in the memory and can be executed by the one or more processors to implement an image compression method, the method includes:

Dividing an image into multiple image blocks, wherein the multiple image blocks adopt different quantization parameters in the compression process;

According to the quantization parameter corresponding to each image block, reconstruct each image block to obtain the final reconstructed image block; and

Combine the final reconstructed image blocks corresponding to each image block to obtain a compressed image of the image,

The step of reconstructing each image block to obtain the final reconstructed image block includes:

Judging the position of the quantization parameter of the image block in the quantization parameter fitting curve;

If the landing point is between the first predetermined quantization parameter and the second predetermined quantization parameter, the first mapping matrix corresponding to the first predetermined quantization parameter is used to reconstruct the image block to obtain a first reconstructed image block , And reconstructing the image block by using a second mapping matrix corresponding to the second predetermined quantization parameter to obtain a second reconstructed image block; and

A linear interpolation method is used to fuse the first reconstructed image block and the second reconstructed image block to obtain the final reconstructed image block.
The image compression device according to claim 9, wherein after the step of determining the location of the drop point, the method further comprises:

If the falling point is smaller than the smallest predetermined quantization parameter, the mapping matrix corresponding to the smallest predetermined quantization parameter is used to reconstruct the image block to obtain the final reconstructed image block.
9. The image compression device of claim 9, wherein after the step of determining the location of the drop point, the method further comprises:

If the drop point is greater than the largest predetermined quantization parameter, the mapping matrix corresponding to the largest predetermined quantization parameter is used to reconstruct the image block to obtain the final reconstructed image block.
9. The image compression device according to claim 9, wherein the quantization parameter fitting curve is obtained through the following steps:

In the training phase, the high-definition training images are compressed with different training quantization parameters to obtain compressed training images, and the compressed training images corresponding to different training quantization parameters are respectively used as output, and the high-definition training images are input into the mapping model To train the mapping model to obtain corresponding mapping matrices under different training quantization parameters;

In the test phase, the high-definition test image is compressed with different quantization parameters for testing to obtain a first compressed test image, and the mapping model is used to map the high-definition test image into a corresponding mapping matrix under different quantization parameters for training. A second compressed test image, finding the best training quantization parameter used by the second compressed test image and its corresponding mapping matrix that have the smallest difference from the first compressed test image under the test quantization parameter; as well as

In the preprocessing stage, a fitting curve between the quantization parameter for testing and the optimal training quantization parameter is obtained, and a mapping matrix corresponding to the quantization parameter for optimal training is stored.
9. The image compression device according to claim 9, wherein after the step of dividing the image into the plurality of image blocks, the method further comprises:

Perform block discrete cosine transform on the plurality of image blocks.