JP2000228771A - Image encoding apparatus and image encoding method - Google Patents

Abstract

(57) [Summary] Even in encoding at a low bit rate, encoding that can reproduce an image with less blurring of edges and ringing due to coarse quantization is performed. A spatial frequency of a DCT coefficient obtained by a DCT transform unit is evaluated by a threshold value determining unit, and a threshold value is determined for each DCT coefficient such that a DCT coefficient having a higher spatial frequency has a larger threshold value. Is determined. In the DCT coefficient conversion unit 2, the absolute value of each DCT coefficient is evaluated by the DCT coefficient conversion unit 2, and among the DCT coefficients, the value is retained for the coefficient having a large absolute value, and the value is reduced for the coefficient having a small absolute value. A continuous, non-linear transformation is performed to make it smaller. This transform is strong for higher order DCT coefficients by using a threshold that is larger for larger orders,
Performing weakly with low order DCT coefficients.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an image encoding apparatus and an image encoding method, and more particularly to an image encoding apparatus and an image encoding method for quantizing image data after orthogonally transforming the data.

[0002]

2. Description of the Related Art Compression encoding of moving images is a key to realizing high capacity and high efficiency by replacing moving image handling in the fields of storage, communication, broadcasting, etc., conventionally performed by analog, with digital. Technology.

FIG. 6 shows a configuration of an encoding unit in these moving image compressions. In FIG. 6, the input image is divided into blocks of 8 × 8 pixels, and each is input to the DCT circuit 10. The DCT circuit 10 performs an orthogonal transform called a two-dimensional discrete cosine transform (DCT). As a result, of the information of the image, information with a low frequency is converted into a low-order coefficient, and information with a high frequency is converted into a high-order coefficient.
In general, in a natural image, information is concentrated in an area having a low frequency. Therefore, it is possible to compress information by coding a low-order coefficient in which information is concentrated and discarding a high-order coefficient in which information is small. This truncation is performed by quantization.

That is, 8 × 8 DCT coefficients output from the DCT circuit 10 are quantized by the quantization circuit 11. This quantization is performed by dividing each coefficient by a preset quantization step and discarding fractions. In general, this quantization step is set fine in reduction and coarse in high frequencies. This allows
As described above, much of the high-frequency information that does not include much information is truncated, and the amount of information can be reduced. Generally, the adjustment of the information compression ratio is realized by increasing or decreasing the value of the quantization step. In order to realize a higher compression ratio, the value of the quantization step is set to be large, and coarser quantization is performed. At this time, more information will be truncated by quantization.

[0007] The quantized DCT coefficients are then input to a run-length encoding unit 12. The run-length encoding unit 12 rearranges the 8 × 8 DCT coefficients in order of frequency in an order called zigzag scan as shown in FIG. The DCT coefficients rearranged in this way have a higher order as they go to the rear, so that the probability that the value becomes 0 increases. Therefore, the coefficients are scanned in this order, and a set of the number of times (zero run) R of zero “0” continues and a non-zero coefficient value (level) L following the zero is generated. The plurality of sets of R and L thus created are then input to the Huffman encoding unit 13. The Huffman encoding unit 13 compresses information by assigning a variable-length code according to the appearance probability to the above-mentioned pair of R and L.

However, in a portion such as an edge existing in an image, information in a high spatial frequency region is important.
If the information in the high spatial frequency region is deleted by coarse quantization of such a portion, the information of the high spatial frequency is lost, so the change in the signal at the edge portion of the reproduced image is lost, and ringing and blurring occur. May be.

To solve such a problem, as a method of reducing the amount of information while preserving edges, for example, Japanese Patent Laid-Open No.
A method as shown in JP 83560 has been proposed. FIG. 8 shows a configuration example of an encoding unit based on this method.

[0008] DCT output by DCT circuit 21
The coefficient is quantized by the quantization unit 22 with a quantization coefficient that does not impair image quality. Next, in order to reduce the generated code amount corresponding to the target bit rate, the analysis processing unit 26
Based on the result of the DCT coefficient analysis processing, the zero replacement unit 23 replaces the higher-order DCT coefficients with smaller absolute values with zeros. The number of pairs of the zero run R and the level L generated by the run length encoding unit 24 can be reduced as compared to before the replacement. For this reason, the number of variable length codes generated by the Huffman coding unit 25 is reduced, and the generated coding amount is suppressed. In general, among the high-order DCT coefficients, many coefficients having a small absolute value reflect pixel values having no substantial information such as noise. The effect of deterioration is small.

[0009]

However, when encoding is performed at a particularly low rate, if the bit rate is reduced by the above-described method, the DCT caused by the originally required image information can be reduced.
By substituting even the coefficients with zero, information may be lost.

SUMMARY OF THE INVENTION The present invention has been made in view of the above points, and in particular, in moving picture coding at a low bit rate, an image capable of obtaining a good reproduced picture while reducing the code amount of DCT coefficients. It is an object to provide an encoding device and an image encoding method.

[0011]

In order to solve the above-mentioned problems, an image encoding apparatus according to the present invention comprises: means for dividing an image into blocks each having n rows and m columns of pixel data; Means for executing a discrete cosine transform, and a threshold value having a larger value is set for a higher-order coefficient, so that each of the coefficients is determined in accordance with the spatial frequency of each coefficient obtained by the discrete cosine transform. Threshold value determining means for determining a threshold value for conversion for each of the coefficients; each coefficient obtained by the discrete cosine transform; and a threshold value determined by the threshold value determining means in accordance with a spatial frequency. Based on the magnitude relationship with the threshold, a coefficient whose absolute value is larger than the corresponding threshold value is in a direction to hold the value, and a coefficient whose absolute value is less than the corresponding threshold value has a smaller absolute value. In the direction A transforming means for transforming the value of each coefficient obtained by the discrete cosine transform; a means for performing quantization on the coefficient sequence of the discrete cosine transform transformed by the transforming means; Encoding means for performing variable length encoding on the coefficient sequence.

According to this image coding apparatus, the dynamic and variable setting of the threshold value in accordance with the order of the coefficient and the continuous and non-linear conversion processing utilizing the magnitude relationship between the coefficient value and the threshold value make it possible to achieve a high level of processing. For the next DCT coefficient having a small value, the probability of being rounded by quantization to become "0" increases, and if the value is reduced to some extent, important information is lost when it is replaced with "0" after quantization. The value of a low-order coefficient can be converted to a value that is not rounded to "0" by quantization. In the Huffman code generally used for variable length coding, the smaller the value of the quantized coefficient, the shorter the code length. For this reason, the code length of the Huffman code assigned to the coefficient whose value has been reduced by the above-described conversion can be made smaller than when the coefficient before the conversion is quantized. This makes it possible to reduce the influence of quantization distortion caused by an excessively large number of quantization steps even in encoding at a low rate.

[0013]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing a configuration of an image coding apparatus according to one embodiment of the present invention. Image data is
The data is input to a discrete cosine transform (DCT) unit 1 in units of blocks divided into 8 × 8 pixels. In the DCT unit 1, two-dimensional DCT is performed on the input block, and 8 ×
It is converted into a coefficient in the spatial frequency domain consisting of eight elements.

The DCT coefficient is input to the DCT coefficient conversion section 2 and also to the threshold value determination section 6. The threshold value deciding unit 6 calculates D D according to the spatial frequency of the DCT coefficient.
This is for determining the threshold value t used in the CT coefficient conversion unit 2, and the higher the DCT coefficient order, the larger the threshold value t is determined. The threshold value t for the coefficients of the horizontal h and vertical v components can be given by the following equation using the constant a.

T = a (h + v) (1) As described above, the threshold value determining unit 6 evaluates the spatial frequency of the DCT coefficient, and the value of the threshold value t is linear with respect to the increase in the spatial frequency. To be increased.

The DCT coefficient converter 2 performs a transform on the DCT coefficient according to the threshold value t. This transformation is D
When the absolute value of the CT coefficient is larger than the threshold, the value is held, and when the absolute value is smaller than the threshold, the conversion is made smaller. DC with threshold value t
For the T coefficient x, the converted coefficient y is, for example, the coefficient r (0
≦ r ≦ 1) can be expressed as follows.

(A) When | x | <t, y = rx (b) When x> t, y = (x−t) + rt (c) When x <−t, y = ( x + t) -rt (2) According to this conversion, if x is smaller than t, the converted coefficient y becomes r times smaller. Further, if x is sufficiently larger than t, the converted coefficient y relatively approaches x, so that it can be considered that the value is substantially maintained.

As described above, the continuous and non-linear conversion processing is strong for high-order DCT coefficients and works slowly for low-order DCT coefficients in combination with the above-described threshold value. That is, for the DCT coefficient, the conversion that the absolute value is smaller if the absolute value is smaller and the value is held if the absolute value is larger can be performed strongly for the high-frequency coefficient and weakly for the low-frequency coefficient.

The transformed DCT coefficients are quantized by the quantization unit 3. The coefficient converted to a small value by the above-described DCT coefficient conversion is quantized to “0” or a value close to “0”. On the other hand, a coefficient having a large absolute value, whose value is substantially held, is quantized to a value that is substantially the same as when the coefficient before conversion is quantized. The quantized coefficients are output to run-length encoding 4.

The run-length encoding unit 4 scans the quantized 8 × 8 DCT coefficients zigzag as shown in FIG. 7 and rearranges them in the order of spatial frequency. Then, it is converted into a set of a length (zero run) R in which "0" continues and a level L of a non- "0" coefficient after "0" continues. Due to the presence of the coefficient quantized to “0” by the above-described conversion, the number of sets of R and L can be reduced as compared with the case where the coefficient before the conversion is quantized and run-length encoded. In addition, as for the coefficient whose value has been reduced by the conversion, the result of the conversion is reflected in the form that the value of the level L is small.

The set of R and L that has been run-length encoded by the run-length encoding unit 4 is encoded by a Huffman encoder 5. Examples of R and L pairs and codes assigned to them (D in MPEG2 which is an international standard for video coding)
The signs of the CT coefficients are shown in FIGS. In general, the smaller the absolute value of the quantized coefficient is, the higher the probability of occurrence is. Therefore, a smaller code tends to be assigned as L is smaller. FIG. 5 shows a plot of the value of L in FIGS. 3 and 4 and the average value of the code lengths of the codes assigned at that time. Also in this example, it can be confirmed that shorter codes tend to be assigned as L is smaller.

FIG. 2 is a flowchart showing the procedure of the encoding process according to this embodiment.

As described above, the image data is first DCT
The transform unit 1 performs an orthogonal transform process using two-dimensional DCT (step S101). And the obtained DCT
The spatial frequency of the coefficient is evaluated by the threshold value determining unit 6 (step S102), and a threshold value is determined for each DCT coefficient such that the DCT coefficient having a higher spatial frequency has a larger threshold value (step S103). Next, the absolute value of each DCT coefficient is evaluated by the DCT coefficient conversion unit 2 (Step S).
104), if | x | <t, y = rx conversion processing (step S105), and if x <−t, y = (x +
t) -rt conversion process (step S106), and if x> t, y = (xt) + rt conversion process (step S106)
107) is performed.

The DCT coefficients thus transformed are quantized by the quantizing unit 3 (step S108), and then are subjected to run-length encoding and Huffman encoding by the run-length encoding unit 4 and the Huffman encoding unit 5. Variable length coding is performed (steps S109 and S110).

As described above, in the present embodiment, D
For the CT coefficient, a conversion is performed such that the smaller the absolute value is, the smaller the absolute value is, the larger the absolute value is, the stronger the high-frequency coefficient and the weaker the low-frequency coefficient, the number of quantization steps, etc. If the conditions are the same, the amount of generated codes generated when encoding the DCT coefficients can be made smaller than in the past. In other words, the number of quantization steps can be set smaller by the reduced amount of generated codes, so that a higher quality reproduced image can be obtained.

The present invention is not limited to the above embodiment, but can be applied to, for example, a moving picture compression apparatus having a circuit for performing motion compensation before the DCT circuit 1.

Further, all the image coding procedures including the DCT coefficient conversion processing of this embodiment can be realized by software. In this case, the same effect as in the present embodiment can be obtained only by introducing a computer program including the procedure into a normal computer through a recording medium and executing the program.

[0029]

As described above, according to the present invention,
For the DCT coefficients, a transform is performed such that the smaller the absolute value is, the smaller the absolute value is, and the larger the absolute value is, the stronger the high-frequency coefficients and the weaker the low-frequency coefficients are. It is possible to reduce the amount of generated code when performing long coding, and it is possible to perform coding at a lower bit rate.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an image encoding device according to an embodiment of the present invention.

FIG. 2 is an exemplary flowchart showing the procedure of an encoding process in the image encoding device of the embodiment.

FIG. 3 is an exemplary view showing a part of a Huffman code used in the embodiment;

FIG. 4 is an exemplary view showing a remaining portion of the Huffman code used in the embodiment;

FIG. 5 is a diagram showing a relationship between a level value and an average code amount in the Huffman code used in the embodiment.

FIG. 6 is a block diagram showing a configuration of a conventional standard image encoding device.

FIG. 7 is a diagram illustrating an example of the order of scanning DCT coefficients.

FIG. 8 is a block diagram showing a configuration of a conventional encoding device that replaces a part of high-order DCT coefficients with 0.

[Explanation of symbols]

 1,10,21 DCT converter 2 DCT coefficient converter 3,11,22 Quantizer 4,12,24 Run-length encoder 5,13,25 Huffman encoder 6 ... Threshold Decision section

Claims (6)

[Claims] A means for dividing an image into blocks each including n rows and m columns of pixel data; a means for performing discrete cosine transform on the divided blocks; and a higher-order coefficient having a larger value. Threshold value determining means for determining, for each coefficient, a threshold value for converting each coefficient according to a spatial frequency of each coefficient obtained by the discrete cosine transform so that a value is set. Based on a magnitude relationship between each coefficient obtained by the discrete cosine transform and a threshold determined according to a spatial frequency by the threshold determining means, the absolute value is larger than the corresponding threshold. Conversion means for converting the value of each coefficient obtained by the discrete cosine transform in a direction in which the coefficient holds its value, and in a direction in which the absolute value is equal to or smaller than the corresponding threshold value in a direction in which the absolute value becomes smaller. When, Means for performing quantization on the coefficient sequence of the discrete cosine transform converted by the conversion means, and coding means for performing variable length coding on the quantized coefficient sequence. An image encoding device characterized by the above-mentioned. 2. The method according to claim 1, wherein the converting means sets a threshold value determined according to a spatial frequency by the threshold value determining means as t,
Assuming that each coefficient value obtained by the discrete cosine transform is x, the converted coefficient value is y, and the transform coefficient is r (0 ≦ r ≦ 1), (a) When | x | <t, y = rx (b) When x> t, y = (x−t) + rt (c) When x <−t, perform a nonlinear conversion of y = (x + t) −rt. The image encoding device according to claim 1. 3. The threshold value determining means sets a threshold value t as h, where h is the horizontal coordinate of each of the coefficient examples having n rows and m columns obtained by the discrete cosine transform, and v is the vertical coordinate. 3. The image encoding apparatus according to claim 1, wherein the value is determined by t = a (h + v) (a is a constant). 4. An image is divided into blocks composed of pixel data of n rows and m columns, and a discrete cosine transform is performed on the divided blocks, and a higher-order coefficient is set to a larger threshold value. As described above, according to the spatial frequency of each coefficient obtained by the discrete cosine transform, a threshold for transforming each coefficient is determined for each of the coefficients, and each threshold obtained by the discrete cosine transform is determined. Based on the magnitude relationship between the coefficient and the threshold value determined according to the spatial frequency, a coefficient whose absolute value is larger than the corresponding threshold value is in a direction to hold that value, and the absolute value is corresponding. The coefficient below the threshold value is converted to a value of each coefficient obtained by the discrete cosine transform in such a direction that its absolute value becomes smaller, and quantization is performed on the converted coefficient sequence. To the sequence of quantized coefficients Picture coding method and performing variable length coding with. 5. An image coding apparatus for performing orthogonal transformation on image data and quantizing a coefficient sequence obtained by the orthogonal transformation, wherein the value of each coefficient obtained by the orthogonal transformation is evaluated, and an absolute value is determined. A coefficient that is larger than the threshold value retains its value, and a coefficient whose absolute value is equal to or less than the threshold value performs continuous non-linear conversion so as to make its value smaller, and means obtained by the orthogonal transformation. The spatial frequency of each coefficient is evaluated, and the higher the frequency, the higher the conversion, and the lower the frequency, the lower the frequency, so that the threshold is varied according to the spatial frequency for each coefficient. An image encoding device comprising: a setting unit. 6. An image coding method for orthogonally transforming image data and quantizing a coefficient sequence obtained by the orthogonal transformation, wherein a low-order space is used in the coefficient sequence obtained by the orthogonal transformation. A coefficient having a frequency and whose absolute value is larger than a predetermined value retains its value, and a coefficient having a higher spatial frequency and whose absolute value is smaller than a predetermined value has a smaller absolute value. An image coding method comprising: converting the coefficient sequence into a coefficient sequence; and performing quantization on the converted coefficient sequence.