JPH08107502A - Quantizing table selecting method - Google Patents

Abstract

PURPOSE: To set a quantizing table with simple processing by selecting one of plural quantizing tables for two-dimensional image based on a feature amount corresponding to the dispersion of respective picture element values. CONSTITUTION: Three kinds of quantizing tables are prepared as initial setting. Next, sampling is executed for thinning the picture elements of a source image as preprocessing. After sampling, the feature amount corresponding to the dispersion of picture element values such as the difference (absolute value) between adjacent picture element values, for example, is calculated. Next, the respective calculated differential values are compared with a threshold value, and the number of differential values exceeding the threshold value is counted. Further, the quantizing table is selected based on this calculated numerical value. Concerning a numerical value in the parentheses of a column for initial setting, when the numerical value calculated like this is equal with that numerical value in the parentheses, the correspondent quantizing table is selected. Thus, the quantizing table suitable for the source image is selected, the characteristics of the image can be grasped by simple operation and the suitable quantizing table suited to the characteristics of that image can be selected.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image coding system for coding image data, and more specifically, to a quantum used for quantization in an image coding system using DCT transform (discrete cosine transform). The present invention relates to a quantization table selection method for selecting a quantization table.

[0002]

2. Description of the Related Art Conventionally, various image coding systems have been proposed, but as one of the systems capable of efficiently coding a high quality image suitable for coding a color still image, a discrete cosine transform ( The DCT) method has been proposed. The discrete cosine transform and the coding method using the discrete cosine transform are already widely known (for example, "Digital Signal Processing Handbook" edited by Institute of Electronics, Information and Communication Engineers, Ohmsha, "Interface"'911.
(See the February issue), but here, only a brief description of a portion related to the present invention described later.

FIG. 3 is a schematic explanatory diagram of encoding by the DCT method. A pixel block of 8 × 8 pixels in a two-dimensional image in which each pixel value corresponds to each pixel is input, discrete cosine transform (DCT transform) is performed, and an 8 × 8 DCT coefficient matrix is generated. The upper left element (coefficient) forming this DCT coefficient matrix represents the level of the low spatial frequency component of the input 8 × 8 pixel block, and the level of the input 8 × 8 pixel block increases toward the right side or the lower side. , Respectively, representing the levels of several components around the high space in the horizontal or vertical direction.

Next, each element of the DCT coefficient matrix is divided by the corresponding element of the prepared quantization table to generate a quantized DCT coefficient matrix in which only the integer part of the result of the division is arranged. It Although there is a quantization table recommended by the standard, the quantization table is basically defined in advance by the user according to the property of the input image.

In a normal image, this quantized DCT coefficient matrix has a value other than 0 on the upper left side corresponding to the low spatial frequency component of the image and 0 on the lower right side corresponding to the high spatial frequency component. Is likely to appear. However, depending on the image, a value other than 0 may appear even on the lower right side. After obtaining this quantized DCT coefficient matrix, each element (each quantized DCT coefficient) forming this quantized DCT coefficient matrix is scanned in a predetermined order, and each quantized DCT coefficient is sequentially encoded according to a predetermined standard. Turn into.

FIG. 4 is a diagram showing the order of scanning the quantized DCT coefficient matrix. As described above, the quantized DCT coefficient matrix has a higher probability that a value of 0 appears on the higher frequency side (lower right side), so that from the low frequency side to the high frequency side, that is, as shown by the arrow in FIG.
Zigzag scanning is performed from the upper left corner to the lower right corner.

In decoding the image data encoded in this way, the reverse order of the above description is followed.

[0008]

In the image coding as described above, the compression rate and the quality of the image after decoding greatly depend on the setting method of each coefficient in the quantization table. In setting each coefficient of this quantization table, usually, the user observes the original image to be encoded by eye to grasp the characteristic of the original image, and each coefficient is set based on the characteristic.

However, if the original image is visually observed and its characteristics are grasped, it takes a lot of time and labor, and there is a problem that the variation is large depending on the observer and the reliability is low. Also,
Instead of observing with human eyes to grasp the features of the image, DC
A technique has been proposed in which the characteristics of an image are grasped by investigating how the values after the T calculation (coefficient values of the DCT coefficient matrix) are distributed from low frequency components to high frequency components. However, this method has a problem that the processing is complicated and it takes time to determine the quantization table.

In view of the above circumstances, it is an object of the present invention to provide a quantization table selection method which can determine a quantization table by a simple process.

[0011]

According to a quantization table selection method of the present invention which achieves the above object, a discrete cosine transform is applied to a pixel value matrix of a two-dimensional image or a two-dimensional partial image in which each pixel value corresponds to each pixel. To generate a coefficient matrix, and generate a quantized coefficient matrix by quantizing each coefficient of the coefficient matrix using a quantization table consisting of each quantized coefficient for quantizing each coefficient of the coefficient matrix, While scanning each coefficient of the quantized coefficient matrix, in a quantization table selection method for selecting a quantization table used for image coding for sequentially coding each coefficient, a plurality of types of quantization tables are prepared, A feature amount corresponding to a variation in each pixel value of each pixel of the two-dimensional image is obtained, and the two-dimensional image is obtained based on the feature amount. As for an image, characterized by selecting one of a plurality of quantization tables.

In the quantization table selection method of the present invention, as the feature amount, the two-dimensional image, a thinned image obtained by thinning out pixels of the two-dimensional image, a partial image of the two-dimensional image, and a partial image of the thinned image. To obtain the absolute value of the difference between the pixel values of adjacent pixels in any one of the images, compare the absolute value with a predetermined threshold value, and select the quantization table according to the comparison result. Is preferred.

[0013]

According to the quantization table selection method of the present invention, a plurality of quantization tables are prepared, and the feature amount corresponding to the variation of the pixel values, for example, the absolute value of the difference between the pixel values of the adjacent pixels is calculated, The quantization table is selected based on the characteristic amount, the characteristic amount is easily obtained by a simple process, and an appropriate quantization table is selected based on the characteristic amount.

[0014]

Embodiments of the present invention will be described below. FIG. 1 is a flowchart showing an embodiment of the quantization table selection method of the present invention. First, as initialization, three types of quantization tables are prepared here. The three types of quantization tables are (1) a table in which all 8 × 8 elements are “FF” (2) a table recommended by the standard (3) all 8 × 8 elements are “1” 'Table. Numerical values in parentheses in the initial setting block of FIG. 1 will be described later.

The fact that all 8 × 8 elements are'FF 'means that the lower right side of the quantized DCT coefficient matrix shown in FIG. 3 tends to be 0, that is, encoding is performed by ignoring high frequency components, and conversely. The fact that all 8 × 8 elements are “1” means that high frequency components are faithfully encoded. Further, here, a threshold value is set as an initial setting.

Next, as preprocessing, subsampling for thinning out pixels of the original image is performed. It should be noted that the sub-sampling may be omitted and the following calculation may be performed using all the pixels forming the original image. FIG. 2 is a diagram showing an example of a thinned image obtained by sub-sampling original pixels. Here, for simplicity, an image with an extremely small number of pixels is displayed here.

After performing such sampling, the difference (absolute value) between adjacent pixel values is calculated. In the example shown in FIG. 2, it is calculated as shown in Table 1.

[0018]

[Table 1]

Next, each difference value obtained as described above is compared with a threshold value, and the number of difference values exceeding the threshold value is counted. Here, if the threshold value is set to 26, the difference value marked with a circle in Table 1 exceeds the threshold value. When the number of difference values exceeding this threshold value is counted, a numerical value of 7 is obtained. Next, a quantization table is selected based on the obtained numerical value. Numerical values in parentheses in the initial setting column of FIG. 1 mean that when the numerical value obtained as described above is the numerical value in the parentheses, the corresponding quantization table is selected. In the example of FIG. 2 and Table 1, the numerical value 7 is obtained, and therefore the quantization table (2) in the column of initial setting is selected.

As described above, the quantization table suitable for the original image is selected. When the original image has a plurality of components, for example, when the original image color-separated into R, G, and B exists, the above process is repeated for each component, and quantization is performed for each component. A table is selected.

[0021]

As described above, according to the present invention,
The characteristics of the image can be grasped by a simple calculation before performing the DCT transformation, and an appropriate quantization table suitable for the characteristics of the image is selected.

[Brief description of drawings]

FIG. 1 is a flowchart showing an embodiment of a quantization table selecting method of the present invention.

FIG. 2 is a diagram showing an example of an image obtained by sub-sampling original pixels.

FIG. 3 is a schematic explanatory diagram of encoding by the DCT method.

FIG. 4 is a diagram showing a scanning order of a quantized DCT coefficient matrix.

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Office reference number FI technical display location H04N 7/30

Claims (2)

[Claims] 1. A coefficient matrix is generated by applying a discrete cosine transform to a pixel value matrix of a two-dimensional image or a two-dimensional partial image in which each pixel value corresponds to each pixel, and each coefficient of the coefficient matrix is quantized. The quantization coefficient matrix is generated by quantizing each coefficient of the coefficient matrix using a quantization table composed of each quantization coefficient of, and each coefficient is scanned while scanning each coefficient of the quantization coefficient matrix. In a quantization table selection method for selecting a quantization table used for image coding to be sequentially coded, a plurality of types of quantization tables are prepared to cope with variations in each pixel value of each pixel of the two-dimensional image. The feature amount is obtained, and one of the plurality of quantization tables is selected for the two-dimensional image based on the feature amount. Quantization table selection method to be used. 2. The image as any one of the two-dimensional image, a thinned image obtained by thinning out pixels of the two-dimensional image, a partial image of the two-dimensional image, and a partial image of the thinned image as the feature amount. 2. The absolute value of the difference between the pixel values of adjacent pixels of the pixel is calculated, the absolute value is compared with a predetermined threshold value, and the quantization table corresponding to the comparison result is selected. Quantization table selection method.