JP2008147880A - Image compression apparatus and method, and its program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To make it possible to reproduce even detailed information of image data while achieving a high compression ratio in image compression. <P>SOLUTION: Even detailed information of the image data is made possible by non-linear approximation of the image data using a surface function which uses compression parameters for compression of the image data as coefficients, and at the same time, it is characterized to presume the compression parameters which make a remainder square sum with weight of image data divided in each sub-block and image data carried out nonlinear approximation of the image data by a surface function to the minimum by a nonlinear type least square method. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an image compression technique, and more particularly to an image compression technique that can be used for image transmission or image data storage in the field of information home appliances such as digital terrestrial TV, digital camera, digital video, DVD recorder, and mobile phone. .

  Conventionally, when compressing image data such as a moving image or a still image, the image is divided into sub-block areas according to a certain standard, and compression processing is performed to the minimum necessary information. In motion pictures, the motion vector search accuracy between frames greatly affects the compression efficiency. For still images, the compression efficiency depends on what criteria the image is divided into. In the past, as an area dividing method, an averaging process was performed for each square sub-block area to linearly approximate an image. In the case of a still image in which fine textures and characters are mixed, it is optimally divided by sub-blocks having different sizes for each region.

On the other hand, recently, in the field of image processing such as three-dimensional vision, a method of approximating an image using a nonlinear function is known (for example, see Non-Patent Document 1).
Xugang, Saburo Tsuji, “Three-dimensional vision”, first edition, Kyoritsu Shuppan, March 10, 2001

  However, in the conventional image compression method for linearly approximating an image, the averaging process in each sub-block does not change, so that there is a problem that detailed information is crushed. On the other hand, if the number of sub-block divisions is increased, the approximation accuracy increases, but the compression ratio decreases. In addition, there has never been an example in which the method using the nonlinear function is applied to the field of image compression technology.

  In view of the above problems, an object of the present invention is to make it possible to reproduce detailed information of image data while realizing a high compression rate in image compression.

  An image compression apparatus according to a first aspect of the present invention receives an input of image data to be compressed, stores the received image data in a storage means, and divides the image data into a predetermined number of sub-blocks. Area dividing means for storing image data for each divided sub-block in the storage means, curved surface function storage means for preliminarily storing information relating to a curved surface function using a compression parameter for compressing the image data as a coefficient, A compression parameter that minimizes the weighted residual sum of squares between the image data for each sub-block divided by the region dividing means and the image data obtained by nonlinear approximation of the image data with a curved surface function by the non-linear least square method. A compression parameter estimation unit that estimates and stores the estimated compression parameter in the storage unit for each sub-block. To.

  In the present invention, it is possible to reproduce image data up to fine information by nonlinearly approximating the image data using a curved surface function using a compression parameter for compressing the image data as a coefficient, and by a nonlinear least square method. Image data can be compressed at a high compression rate by estimating the compression parameter that minimizes the weighted residual sum of squares of the image data divided for each sub-block and the image data nonlinearly approximated by a curved surface function. can do.

  The compression parameter estimation means in the image compression apparatus estimates a compression parameter using a robust estimation method based on robust statistics.

  In the present invention, by estimating the compression parameter using a robust estimation method based on robust statistics, the influence from the outlier specific to the least squares method can be reduced, and even more detailed information can be obtained from the image data. It is possible to estimate a compression parameter that can be reproduced.

  In the image compression method according to the second aspect of the present invention, the image data receiving means receives the input of the image data to be compressed, the received image data is stored in the storage means, and the area dividing means stores the image data. A step of storing the image data for each of the divided sub-blocks in a storage unit, and information on a curved surface function using a compression parameter for compressing the image data as a coefficient by the curved surface function storage unit; Is stored in advance, and the compression parameter estimation means weights the image data for each sub-block divided by the area dividing means by the nonlinear least square method and the image data obtained by nonlinearly approximating the image data with a curved surface function. Estimate the compression parameter that minimizes the residual sum of squares, and calculate the estimated compression parameter for each sub-block. And storing the 憶 means, characterized by having a.

  An image compression program according to a third aspect of the present invention causes a computer system to execute each step in the image compression method.

  According to the present invention, image data is nonlinearly approximated using a curved surface function, and fine information of the image data can be reproduced while realizing a high compression rate.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[First Embodiment]
As shown in the block diagram of FIG. 1, the image compression apparatus 100 according to the first embodiment includes an image data receiving unit 101, a region dividing unit 102, and a curved surface function information storage that stores information on curved surface functions. Unit 103, compression parameter estimation unit 104, image data decompression unit 105, and memory 110.

  The image data receiving unit 101 receives input of image data to be compressed, and stores the received image data in the memory 110. Here, the image data is input from, for example, a camera installed on a website on a network such as the Internet, image data on the website, or a scanner or a digital camera.

  The area dividing unit 102 divides the image data into a predetermined number of sub-blocks, and stores the divided image data for each sub-block in the memory 110.

  The curved surface function information storage unit 103 stores in advance information related to a curved surface function that uses a compression parameter for compressing image data as a coefficient.

  The compression parameter estimation unit 104 minimizes the weighted residual sum of squares of the image data for each sub-block divided by the region dividing unit 102 and the image data obtained by nonlinear approximation of the image data with a curved surface function by the non-linear least square method. The compression parameter to be estimated is estimated for each sub-block, and the estimated compression parameter is stored in the memory 110 for each sub-block.

  The image data expansion unit 105 expands all image data by substituting the compression parameters estimated for each sub-block into the curved surface function.

  In the memory 110, received image data is divided and stored for each sub-block, and the estimated compression parameter is stored for each sub-block. Here, as shown in FIG. 2, for example, image data of 300 pixels × 300 pixels is divided into sub-blocks of 3 pixels × 3 pixels. A compression parameter is stored corresponding to each of the 100 sub-blocks.

  Next, an image compression procedure by the image compression apparatus will be described with reference to the flowchart of FIG. This processing procedure is based on the image compression method of the present invention.

  First, information regarding a curved surface function that uses a compression parameter for compressing image data as a coefficient is stored in advance in the curved surface function information storage unit 103. The curved surface function stored here is a quadric surface.

  Step 1: The image data receiving unit 101 receives input of image data to be compressed, and stores the received image data in the memory 110. Here, for example, image data of 300 pixels × 300 pixels is input.

Step 2: The area dividing unit 102 divides the image data stored in the memory 110 into a predetermined number of sub-blocks, and stores the divided image data for each sub-block in the memory 110. Here, as shown in FIG. 2, the image data I _{x, y} of 300 pixels × 300 pixels is divided and stored in the memory 110 for each sub-block of 3 pixels × 3 pixels. Here, the gray value of each pixel of I _{x and y} corresponds to, for example, 8-bit gradation data corresponding to any one of RGB three colors.

  Step 3: The weighted residual sum of squares of the image data for each sub-block divided by the region dividing unit 102 and the image data obtained by nonlinear approximation of the image data with a curved surface function by the compression parameter estimation unit 104 by the nonlinear least square method Estimate the compression parameter that minimizes.

Specifically, first, image data I _{x, y} for each sub-block stored in the memory 110 is converted into a quadratic curved surface function represented by the following expression (1) stored in the curved surface function information storage unit 103 in advance. Use nonlinear approximation.

Here, z (x, y) represents two-dimensional image data that is nonlinearly approximated, and the two-dimensional coordinates of x and y correspond to the pixels of the image. 6 unknown coefficients _{a 20, a 11, a 02} , a 10, a 01, a 00 is the compression parameter to be estimated.

A weighted residual sum of squares E ₁ between the image data I _{x, y for} each sub-block and the image data z (x, y) nonlinearly approximated by a curved surface function by the nonlinear least square method is expressed by the following equation (2). .

Here, R ² indicates a region in the sub-block.

In this embodiment, in order to mitigate the influence from the outlier inherent in the least squares method, the compression parameters a _20, a _11, a _02, a _10, a _01, a robust estimation method based on robust statistics are used _. a ₀₀ is estimated. A weighted residual sum of squares E ₂ shown by the following equation (3) is introduced using a robust function.

Here, ρ is a Lorentz function used as a robust function. Σ defined in the Lorentz function is a variance value of image data in a sub-block. As shown below, the partial differential for each coefficient with respect to E ₂ in Equation (3) is calculated in advance.

Next, when solving the equation (3), iterative calculation is performed by the steepest descent method represented by the following equation (4).

  Here, μ is a parameter called a learning rate, and is used to control the amount of descent. n indicates the number of iterations, for example, about 5 times. Zero was given as the initial value. At this time, the iterative calculation is performed while decreasing the variance value σ in the Lorentz function stepwise. For example, the value is increased by 0.9 from σ120 and gradually reduced. Thereby, the convergence of parameter estimation can be improved.

Step 4: The compression parameters a _20, a _11, a _02, a _10, a _01, a ₀₀ estimated in step 3 are stored in the memory 110 for each sub-block as shown in FIG. Thus, by storing the compression parameters estimated together with the information about the curved surface function for each sub-block, it is possible to save information about a large amount of image data in an information storage system such as a database or a recording medium such as a CD-ROM. It becomes.

  Next, image compression and image expansion will be described with reference to FIG. As shown in the figure, at the time of image compression, compression parameters are obtained for a certain sub-block of an original image in which the gray value of the sub-block pixel is in the range of 173-201. Further, the present inventor has confirmed that a 3 × 3 sub-block can be approximated with a certain level of accuracy with many image patterns.

In the image expansion, the image data expansion unit 105 substitutes the compression parameter estimated for each sub-block into the curved surface function to expand all the image data. Here, the compression parameters _{a 20, a 11, a 02} , a 10, a 01, a 00 into Equation (1), subblock inside the x, obtaining the value of z corresponding to the coordinate value of y . As a result, grayscale restored images 169 to 198 as shown in FIG. 4 are obtained.

  FIG. 5 shows an example in which the image compression / decompression process is applied to a captured image of a digital camera. In the figure, the left image shows the original image input from the digital camera to the image compression apparatus 100, and the right image shows the restored image compressed and expanded by the image compression apparatus 100. As shown in the figure, it can be seen that finer image information such as clothes and lines can be approximated. The central part of the can displayed in both images is not subject to evaluation.

  Next, an image compression rate according to the image compression method will be described. If the data size of the original image is 300 pixels × 300 pixels and the information amount is 8 bits per pixel, the information amount before compression is 300 × 300 × 8. On the other hand, it is represented by six compression parameters for each sub-block of 3 pixels × 3 pixels. Here, since the number of sub-blocks is 10 × 10, the amount of information after compression is 10 × 10 × 6. Therefore, the amount of information before compression / the amount of information after compression = 720000/600 = 1200, and a very high compression rate can be realized.

  Therefore, according to the present embodiment, it is possible to reproduce the image data up to fine information by nonlinearly approximating the image data using a curved surface function that uses a compression parameter for compressing the image data as a coefficient, and the nonlinear minimum A high compression ratio is obtained by estimating the compression parameter that minimizes the weighted residual sum of squares of the image data divided for each sub-block and image data obtained by nonlinear approximation of the image data with a curved surface function by the square method. Can be compressed.

[Comparative example]
Next, in order to facilitate understanding of the present invention, non-linear approximation and linear approximation of image data will be described by comparing both.

  As shown in FIG. 6, the image 300 input here uses an image having a texture as a background and a character A displayed in the center. The image 310 is a restored image when the image 300 is divided into four sub-blocks and approximated by a linear surface. That is, an average gray value is calculated for each sub-block. As a result, fine textures and characters cannot be approximated. In the case of linear approximation, the accuracy of approximation increases as the number of sub-block divisions is made finer, but 10-30% is the limit from the viewpoint of compression.

On the other hand, the image 320 is a restored image when the image 300 is divided into four sub-blocks and approximated by a non-linear surface. Here, nonlinear approximation is performed using the quadratic surface function of Equation (1), and the compression parameter that is the coefficient of the surface function is estimated by the least square method for the weighted residual sum of squares E _{1 of} Equation (2). ing. As a result, character features can be obtained somewhat more than when linear approximation is used, but finer feature amounts cannot be approximated.

Therefore, in the present embodiment, the compression parameter is estimated by the robust estimation method based on the robust statistics with respect to the weighted residual sum of squares E ₂ in Expression (3). As a result, the influence from the outlier inherent in the least squares method can be mitigated, and it is possible to estimate a compression parameter capable of reproducing image data up to finer information.

  Further, according to the present embodiment, the convergence of parameter estimation can be improved by performing the iterative calculation while decreasing the variance value in the Lorentz function used as a robust function stepwise by the steepest descent method.

  Furthermore, according to the present embodiment, it is possible to expand all image data by substituting the compression parameter estimated for each sub-block into the curved surface function.

[Second Embodiment]
Hereinafter, a second embodiment will be described. The basic configuration of the image compression apparatus according to the present embodiment is the same as that described in the first embodiment. Below, it demonstrates centering on a different point from 1st Embodiment.

  The difference from the first embodiment is that when the image compression apparatus 100a in FIG. 7 transmits image data, the image compression apparatus 100a transmits information on the compression parameter to the receiving side having the same information as the curved surface function. The compression parameter transmission unit 106 is further provided. Thereby, the image compression apparatus 100a transmits a compression parameter to the image data receiving apparatus as the receiving side, for example, in an image transmission system.

  As shown in the block diagram of FIG. 8, the image data receiving device 200 stores a compression parameter receiving unit 201 that receives the compression parameters transmitted from the image compressing device 100a, and the compression parameter information received by the compression parameter receiving unit 201. A storage unit 202 for storing curved surface function information in advance, and an image data expansion unit 204 for expanding all image data by substituting compression parameters into the curved surface function. Note that the image decompression method by the image data decompression unit 204 is the same as the method by the image data decompression unit 105 described in the first embodiment, and a description thereof will be omitted here.

  In this way, information on the same curved surface equation is held in advance on both the transmitting and receiving sides, and only a small amount of information typified by compression parameters is exchanged, enabling image compression communication in the image transmission system and a large amount of image data. The system can be made compact with high-speed communication.

  The image compression apparatus according to each of the above-described embodiments described above can also be realized by causing a computer program such as a personal computer to read the computer program for realizing it and executing the program. Programs for realizing these image compression apparatuses are read into a computer by a recording medium such as a CD-ROM or via a network.

  Such a computer generally includes a CPU (Central Processing Unit), a hard disk device for storing programs and data, a main memory, an input device such as a keyboard and a mouse, a display device such as a CRT, and a CD- It comprises a reading device that reads a recording medium such as a ROM, a communication interface for connecting to a network, and an image input device (scanner device, digital camera, etc.) for reading an image. When image data is read via a network or from a recording medium, an image input device is not necessarily provided. The hard disk device, main memory, input device, display device, reading device, communication interface, and image input device are all connected to the CPU. In this computer, a recording medium storing a program for realizing an image compression device is loaded into a reading device, the program is read from the recording medium and stored in a hard disk device, or such a program is downloaded via a network. When the CPU executes the program stored in the hard disk device and stored in the hard disk device, it functions as the above-described image compression device.

  As described above, the image compression technology according to the present invention has the potential to be the core of new image conversion technologies in domestic and overseas digital industries such as digital cameras, digital video, and digital video for home use to business use in the broadband era. is doing.

1 is a block diagram showing a schematic configuration of an image compression apparatus according to a first embodiment. It is a chart which shows an example of the data memorize | stored in the said image compression apparatus. It is the flowchart which showed the flow of the compression process of the image data by the said image compression apparatus. It is explanatory drawing for demonstrating the image compression / expansion | extension process in a certain subblock. An example in which the image compression / decompression process is applied to a captured image of a camera is shown. Explanatory drawing about the linear approximation and nonlinear approximation of image data. It is a block diagram which shows the schematic structure of the image compression apparatus which concerns on 2nd Embodiment. It is a block diagram which shows the schematic structure of the image data receiver which receives the compression information which the said image compression apparatus transmitted.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100, 100a ... Image compression apparatus 101 ... Image data reception part 102 ... Area division part 103 ... Curved surface function information storage part 104 ... Compression parameter estimation part 105 ... Image data expansion | extension part 106 ... Compression parameter transmission part 110 ... Memory 200 ... Image data Receiving device 201 ... compression parameter receiving unit 202 ... compression parameter information storage unit 203 ... curved surface function information storage unit 204 ... image data decompression unit

Claims (10)

Image data receiving means for receiving input of image data to be compressed and storing the received image data in a storage means;
Area dividing means for dividing the image data into a predetermined number of sub-blocks and storing the image data for each divided sub-block in a storage means;
A curved surface function storage means for storing in advance information relating to a curved surface function using a compression parameter for compressing the image data as a coefficient;
A compression parameter that minimizes the weighted residual sum of squares of the image data for each sub-block divided by the region dividing unit and the image data obtained by nonlinear approximation of the image data by the curved surface function by a non-linear least square method A compression parameter estimation unit that estimates for each sub-block and stores the estimated compression parameter in a storage unit for each sub-block;
An image compression apparatus comprising: The compression parameter estimation means includes
The image compression apparatus according to claim 1, wherein the compression parameter is estimated using a robust estimation method based on robust statistics. The compression parameter estimation means includes
3. The image compression apparatus according to claim 2, wherein the iterative calculation is performed while the variance value in the Lorentz function used as the robust function is reduced stepwise by the steepest descent method.   4. The image compression apparatus according to claim 1, further comprising an image expansion unit that expands all image data by substituting a compression parameter estimated for each sub-block into the curved surface function.   5. The apparatus according to claim 1, further comprising compression parameter transmission means for transmitting information on the compression parameter to a receiving side having the same information as the curved surface function when transmitting the image data. The image compression apparatus according to any one of the above. Receiving input of image data to be compressed by the image data receiving means, and storing the received image data in the storage means;
Dividing the image data into a predetermined number of sub-blocks by an area dividing unit, and storing the image data for each divided sub-block in a storage unit;
Preliminarily storing information on a curved surface function using a compression parameter for compressing the image data as a coefficient by a curved surface function storage unit;
The compression parameter estimation unit minimizes the weighted residual sum of squares of the image data for each sub-block divided by the region dividing unit and the image data obtained by nonlinear approximation of the image data by the curved surface function by the nonlinear least square method. Estimating a compression parameter to be stored, and storing the estimated compression parameter in a storage unit for each sub-block;
An image compression method characterized by comprising: In the step by the compression parameter estimation means,
The image compression method according to claim 6, wherein the compression parameter is estimated using a robust estimation method based on robust statistics. In the step by the compression parameter estimation means,
8. The image compression method according to claim 7, wherein the iterative calculation is performed while the variance value in the Lorentz function used as a robust function is reduced stepwise by the steepest descent method.   9. The image according to claim 6, further comprising a step of expanding all image data by substituting a compression parameter estimated for each sub-block into the curved surface function by an image expansion means. Compression method.   An image compression program causing a computer system to execute each step in the image compression method according to claim 6.