JP2011254257A - Image compression device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image compression device capable of improving compressibility while suppressing deterioration of image quality.SOLUTION: An input unit 1 divides input image data into a plurality of blocks so as to have overlapped portions between adjacent blocks. An orthogonal conversion unit 2 orthogonally converts each block divided by the input unit 1. A quantization coefficient setting unit 4 sets a quantization coefficient to each piece of frequency data of each block, and a quantization unit 3 quantizes the frequency data of each block by use of the quantization coefficient.

Description

  The present invention relates to an image compression apparatus that compresses image data.

  As one general method for compressing (encoding) image data, for example, a JPEG (Joint Photographic Experts Group) method disclosed in Patent Document 1 is known. The procedure for encoding image data in a general JPEG system is as follows. First, input image data is divided into a plurality of blocks (generally 8 × 8 pixels). After block division, each block is subjected to discrete cosine transform (DCT) processing to convert the image data of each block into frequency data. After DCT processing, each frequency data is quantized by dividing it by a quantization coefficient (quantization table) having a size (8 × 8 pixels) corresponding to each frequency data. In this case, the quantization coefficient is a characteristic that reduces higher-frequency data based on the visual characteristics of the human eye, that is, the quantization coefficient corresponding to the high-frequency component is generally increased. is there. After quantization, the frequency data after each quantization is rearranged so that 0 data is continuous. Finally, the rearranged quantized frequency data is subjected to encoding such as zero run length encoding or entropy encoding such as Huffman encoding. As a result, encoded data is generated.

Japanese Patent Laid-Open No. 7-74959

  As described above, in the JPEG scheme, the compression rate is increased by mainly reducing information on high frequency components by quantization. Therefore, the compression rate and the image quality (quantization error in the high frequency part of the image) are in a trade-off relationship. That is, in the JPEG system, if it is attempted to improve the image quality, it is necessary to leave a high-frequency component, so the compression rate must be reduced. Therefore, in the JPEG method, it is difficult to achieve both compression ratio and image quality at a high level.

  The present invention has been made in view of the above circumstances, and an object of the present invention is to provide an image compression apparatus capable of improving the compression rate while suppressing deterioration in image quality.

  In order to achieve the above object, an image compression apparatus according to an aspect of the present invention includes an image dividing unit that divides input image data into a plurality of blocks so that adjacent blocks have overlapping portions, and An orthogonal transform unit that orthogonally transforms each block to obtain frequency data for each block; and a quantization unit that quantizes the frequency data for each block with a different quantization coefficient for each adjacent block. Features.

  ADVANTAGE OF THE INVENTION According to this invention, the image compression apparatus which can improve a compression rate can be provided, suppressing deterioration of an image quality.

It is a block diagram which shows the structure of the image compression apparatus which concerns on one Embodiment of this invention. It is a figure for demonstrating duplication of the block at the time of block division. It is a figure which shows the example of a LOT coefficient matrix. It is a figure which shows the concept of the calculation of overlap orthogonal transformation. It is a figure which shows the example of the frequency data obtained after overlapped orthogonal transformation. It is a figure for demonstrating the quantization coefficient in one Embodiment of this invention.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing a configuration of an image compression apparatus according to an embodiment of the present invention. As shown in FIG. 1, an image compression apparatus 100 according to the present embodiment includes an input unit 1, an orthogonal transform unit 2, a quantization unit 3, a quantization coefficient setting unit 4, a rearrangement unit 5, and an encoding unit. Part 6.

  The input unit 1 having a function as an image dividing unit divides image data input from the outside into a plurality of blocks by capturing the image data while overlapping them. The orthogonal transform unit 2 performs orthogonal transform processing on the blocks divided in the input unit 1 to convert image data for each block into frequency data for each block. The quantization unit 3 quantizes the frequency data for each block obtained in the orthogonal transform unit 2. The quantization coefficient setting unit 4 sets a quantization coefficient used when the quantization unit 3 performs quantization. The rearrangement unit 5 rearranges the data quantized by the quantization unit 3. The encoding unit 6 compresses (encodes) the quantized data rearranged by the rearrangement unit 6.

  Next, the operation of the image compression apparatus 100 shown in FIG. 1 will be described. First, the input unit 1 reads image data for one frame from an image memory (not shown) and divides it into blocks for each minimum coding unit (MCU). At this time, as shown in FIG. 2, the input unit 1 divides the image data so that adjacent blocks have an overlapping portion of data that is only half the block size. For this purpose, for example, the input unit 1 reads image data stored in the image memory in units of blocks that are overlapped by an amount corresponding to half of the block size. Here, in this embodiment, it is assumed that one MCU is, for example, 16 × 16 pixels. Therefore, for example, an overlapping portion between the block A at the upper left end indicated by the solid line in FIG. 2 and the block B indicated by the broken line in FIG. 2 adjacent to the block A in the horizontal direction is indicated by C in FIG. This is a portion of × 16 pixels.

  After the division of the image data by the input unit 1, the orthogonal transform unit 2 performs orthogonal transform processing in order from the upper left block. As the orthogonal transformation process here, a method called orthogonal overlapping transformation (Lapped Orthogonal Transform: LOT) method or overlapping biorthogonal transform (LBT) method, which performs orthogonal transformation by overlapping blocks, is used. These orthogonal transform methods have a feature that block distortion can be reduced by overlapping blocks as compared with DCT (discrete cosine transform) normally used in the JPEG method. Here, as an example, the calculation of overlapped orthogonal transformation will be briefly described. The overlapped orthogonal transformation is performed by performing a matrix operation twice on the 16 × 16 pixel block divided by the input unit 1 with the 16 × 8 pixel LOT coefficient matrix shown as an example in FIG. At this time, first, as shown in FIG. 4, the 16 × 8 pixel LOT coefficient matrix is multiplied by the data of each pixel included in the 16 × 16 pixel block from the right to obtain 16 × 8 pixel frequency data (1 Dimension output). Thereafter, the one-dimensional output of 16 × 8 pixels is transposed to obtain frequency data of 8 × 16 pixels. Then, the frequency data of 8 × 8 pixels is obtained by multiplying the LOT coefficient matrix of 16 × 8 pixels by the frequency data of 8 × 16 pixels from the right. Finally, the 8 × 8 pixel frequency data is transposed to obtain final 8 × 8 pixel frequency data (two-dimensional output). In this way, the frequency data obtained after overlapping orthogonal transformation of a block of 16 × 16 pixels has a direct current (DC) component at the position of the upper left corner as shown in FIG. The data is 8 × 8 pixels.

  After the orthogonal transformation by the orthogonal transformation unit 2, the quantization unit 3 performs quantization on the frequency data of 8 × 8 pixels obtained by the orthogonal transformation unit 2. The quantization unit 3 performs quantization by dividing the frequency data of 8 × 8 pixels by the quantization coefficient of 8 × 8 pixels set by the quantization coefficient setting unit 4. At this time, the value after the decimal point is rounded down. The quantization coefficient setting unit 4 sequentially sets quantization coefficients corresponding to each frequency data for each block in the quantization unit 3.

  Here, the quantization coefficient set by the quantization coefficient setting unit 4 will be described with reference to FIG. FIG. 6 is a diagram illustrating a quantization matrix that represents quantization coefficients in a matrix. Here, the position of each quantization coefficient in FIG. 6 corresponds to the position of the frequency data in FIG. In addition, the hatching density in FIG. 6 represents the magnitude of the quantization coefficient. However, the hatching density merely represents the magnitude relationship of the coefficients for each large area in the matrix, and does not mean that the quantization coefficients in the same density area all have the same value. In the following description, the region where the hatching density is the lightest is referred to as a low frequency region, the region where the hatching density is the darkest is referred to as a high frequency region, and the region between them is referred to as an intermediate frequency region.

  The quantization coefficient having the characteristic shown in FIG. 6A is a quantization coefficient used in the JPEG method, which is one of general image data encoding methods. The quantization coefficient having the characteristics shown in FIG. 6A increases in the order of the low frequency region, the medium frequency region, and the high frequency region in both the horizontal direction (horizontal direction) and the vertical direction (vertical direction). It has characteristics. As described above, the coefficient values are different even in the same hatched area, and in the case of FIG. 6A, even in the same area, the coefficient gradually increases. It has the property of increasing the value. When quantization is performed with a quantization coefficient as shown in FIG. 6A, the higher the frequency component (lower right position), the higher the probability that the value after quantization becomes smaller or zero. Therefore, it is possible to reduce high-frequency component information by quantization and increase the compression rate of image data. On the other hand, when decoding (decompressing) the image data after encoding (compression), high-frequency components in the image are not reproduced, which may lead to image quality deterioration such as a fine pattern in the image not being reproduced.

  In contrast, in the present embodiment, a plurality of quantization coefficients, for example, a first quantization coefficient shown in FIG. 6B and a second quantization coefficient shown in FIG. 6C are used. The first quantization coefficient shown in FIG. 6 (b) has substantially the same characteristics as those shown in FIG. 6 (a), for example. Therefore, after quantization, the higher the frequency component, the higher the probability that the value after quantization will be smaller or zero. On the other hand, the second quantization coefficient shown in FIG. 6C has a characteristic in which the value of the quantization coefficient in the high frequency region is decreased and the value of the quantization coefficient in the intermediate frequency region is increased. Therefore, after quantization, in the medium frequency component, there is a high probability that the value after quantization becomes small or becomes zero.

  The quantization coefficient setting unit 4 alternately sets the quantization coefficients having such two types of characteristics for the frequency data for each block along at least one of the horizontal direction and the vertical direction. For example, considering the case where the quantization coefficient is set for the frequency data for each block along the horizontal direction, the first quantization coefficient is set for the block A in FIG. A second quantization coefficient may be set for the block B adjacent to the block B. Although not shown in FIG. 2, a first quantization coefficient is set for a block adjacent to the block B in the horizontal direction. In this way, when quantization is performed with the quantization coefficient set for each block, the block quantized with the first quantization coefficient (for example, block A) has a low frequency range and a medium frequency range. Image quality degradation (error due to quantization) is small, and a block quantized with the second quantization coefficient (for example, block B) has low image quality degradation (error due to quantization) in the low frequency range and high frequency range. Further, the overlapping portion of the two blocks is quantized with both the first quantization coefficient and the second quantization coefficient. In this case, at the time of subsequent decoding, for the overlapping part, the decoding result of the data quantized with the first quantization coefficient and the decoding result of the data quantized with the second quantization coefficient are integrated (averaged). As a result, the image quality in the high frequency region can be improved.

  After quantization by the quantization unit 3, the rearrangement unit 5 rearranges the quantized data. This rearrangement is performed so that the quantized 0 data is as continuous as possible. In the method of the present embodiment, since the characteristics of the quantization coefficient change for each block, the frequency position with a high probability of becoming 0 also changes depending on the block. Therefore, it is desirable to change the rearrangement order according to the characteristics of the quantization coefficient. As a rearrangement technique when quantization is performed using the quantization coefficients having the characteristics shown in FIGS. 6B and 6C, a zigzag scan generally used in the JPEG method or the like is used. It is possible to use a scheme. In this case, for example, for a block that has been quantized using the quantization coefficient shown in FIG. 6B, 0 is continued at the end of the frequency data after the zigzag scan. On the other hand, in the block that has been quantized using the quantization coefficient shown in FIG. 6C, 0 continues in the middle portion of the frequency data after zigzag scanning.

  After the rearrangement by the rearrangement unit 5, the encoding unit 6 performs entropy encoding such as zero-run length encoding or Huffman encoding on the frequency data rearranged by the rearrangement unit 5. Then, the encoding unit 6 outputs the encoded image data to the outside. Here, run-length encoding is an encoding method in which data having a certain same value is represented by data indicating the value and the number of repetitions thereof. Therefore, a high compression rate can be expected if many of the same values (0 data) are continued by the rearrangement by the rearrangement unit 5. In addition, entropy coding such as Huffman coding is a coding method in which a short code is assigned to data having a high probability of occurrence, and conversely, a long code is assigned to data having a low probability of occurrence, thereby reducing the code amount as a whole. Therefore, the higher the occurrence probability of the data before encoding, the higher the compression rate can be expected.

  Incidentally, both the first quantization coefficient shown in FIG. 6B and the second quantization coefficient shown in FIG. 6C indicate the number of quantization coefficients having a large value as shown in FIG. This is not a reduction compared to the conventional quantization coefficient shown in FIG. Therefore, the difference between the quantization by the first quantization coefficient and the quantization by the second quantization coefficient is that the result of quantization by the first quantization coefficient is a probability that the value in the high frequency region becomes small and becomes zero. On the other hand, the result of quantization with the second quantization coefficient is only that the value in the middle frequency range becomes smaller and the probability of becoming 0 becomes higher. In this case, the result of quantization with the first quantization coefficient is biased in the data generation probability in the high frequency range (continuous 0 data), and the result of quantization with the second quantization coefficient is the data in the middle frequency range. There is a bias in the occurrence probability (continuation of 0 data). Thus, the frequency data after quantized with these two quantization coefficients has a data bias equivalent to the frequency data quantized with the conventional quantization coefficient shown in FIG. Therefore, it is possible to obtain a compression rate equivalent to the conventional one by entropy coding.

On the other hand, if the conventional quantization coefficient shown in FIG. 6A is used alone to obtain an image quality equivalent to that of the present embodiment, it is necessary to set the quantization coefficient in the high frequency region to a small value. In this case, the occurrence probability of the data after quantization is not biased, and the compression rate does not increase.
As described above, in the present embodiment, the image data is divided into a plurality of blocks so as to have an overlapping portion between adjacent blocks, and orthogonal transformation is performed for each block, and thereafter, a different quantization coefficient is assigned to each block. It is used to perform quantization. As a result, it is possible to reflect a plurality of quantization results for the pixels included in the overlapping portion. Therefore, it is possible to improve the compression rate while suppressing the deterioration of the image quality, compared to the case where encoding is performed using a single quantization coefficient. In the present embodiment, the size of each block is matched with the size of the MCU. By dividing the image data into blocks for each MCU having overlapping portions and performing orthogonal transform, discontinuity in image quality due to changing the quantization coefficient for each block can be reduced. Furthermore, in this embodiment, the characteristics of the quantization coefficient are alternately switched for blocks along at least one of horizontal (horizontal) and vertical (vertical). Thereby, it is not necessary to have information on the switching position in order to switch the characteristic of the quantization coefficient. Therefore, quantization using a different quantization coefficient for each block can be realized with a simple configuration. Further, if the characteristics of the quantization coefficient are switched in each of the horizontal direction and the vertical direction, the types of frequencies that can be selected increase, leading to further image quality improvement.

  Here, in the above-described embodiment, an example is shown in which quantization coefficients are alternately set for frequency data for each block along the horizontal direction. On the other hand, the quantization coefficient may be alternately set for the frequency data for each block along the vertical direction, or the quantization coefficient may be alternately set for each of the horizontal direction and the vertical direction. You may do it. In addition, when setting a quantization coefficient with respect to each of a horizontal direction and a vertical direction, three types of quantization coefficients will be used.

Furthermore, the above-described example describes an example in which quantization coefficients are set for all blocks included in image data. On the other hand, for example, only the part in which the image quality in the image data is not desired to be degraded may be different from the quantization coefficient.
In addition, the first quantization coefficient and the second quantization coefficient have characteristics that increase the high-frequency and middle-frequency quantization coefficients, respectively, but are not limited thereto. For example, for the first quantization coefficient, the value of the quantization coefficient is increased as the frequency increases only in the horizontal direction, and for the second quantization coefficient, the quantization coefficient increases as the frequency increases only in the vertical direction. Any characteristic that complements each other, such as increasing the value of, may be used.

  Further, the first quantization coefficient shown in FIG. 6B is such that the quantization coefficients in the low frequency region, the medium frequency region, and the high frequency region increase in the order of the low frequency region, the intermediate frequency region, and the high frequency region. If it is a relationship, even if the quantization coefficient is different in each frequency region, it can be regarded as the first quantization coefficient. Similarly, if the second quantization coefficient shown in FIG. 6C has a relationship that increases in the order of low frequency region, high frequency region, and medium frequency region, the quantization coefficient is within each frequency region. Even if they are different, it can be regarded as the second quantization coefficient.

Although the present invention has been described above based on the embodiments, the present invention is not limited to the above-described embodiments, and various modifications and applications are naturally possible within the scope of the gist of the present invention.
Further, the above-described embodiments include various stages of the invention, and various inventions can be extracted by appropriately combining a plurality of disclosed constituent elements. For example, even if some configuration requirements are deleted from all the configuration requirements shown in the embodiment, the above-described problem can be solved, and this configuration requirement is deleted when the above-described effects can be obtained. The configuration can also be extracted as an invention.

  DESCRIPTION OF SYMBOLS 1 ... Input part, 2 ... Orthogonal transformation part, 3 ... Quantization part, 4 ... Quantization coefficient setting part, 5 ... Rearrangement part, 6 ... Encoding part, 100 ... Image compression apparatus

Claims (5)

An image dividing unit that divides input image data into a plurality of blocks so as to have overlapping portions between adjacent blocks;
An orthogonal transform unit that orthogonally transforms each block to obtain frequency data for each block;
A quantization unit that quantizes the frequency data for each block with a different quantization coefficient for each adjacent block;
An image compression apparatus comprising:   The image compression apparatus according to claim 1, wherein the orthogonal transformation unit performs overlapping orthogonal transformation or overlapping biorthogonal transformation as the orthogonal transformation.   The image compression apparatus according to claim 1, wherein the image dividing unit divides the input image data into a plurality of blocks for each minimum coding unit.   The quantization unit performs the quantization while alternately switching a plurality of quantization coefficients with respect to frequency data for each block along at least one of a horizontal direction and a vertical direction of the input image data. The image compression apparatus according to claim 1, wherein the image compression apparatus is an image compression apparatus. The quantization coefficient is
A first quantization coefficient having a characteristic in which a value sequentially increases from a low frequency region to a high frequency region;
A second quantization coefficient having a characteristic in which a medium frequency region coefficient is smaller than a low frequency region value and larger than a high frequency region value;
The image compression apparatus according to claim 1, further comprising:

Images