JPH0329582A - Coding system for picture - Google Patents

Abstract

PURPOSE:To prevent the reduction in the processing speed by deciding a round-off bit number of a quantization orthogonal transformation coefficient in a way that all data quantity of the quantization orthogonal transformation coefficient after round- off is smaller than a predetermined threshold level and sending the round-off bit number and class classification information of each block as header information. CONSTITUTION:The round-off control section 8 compares all bit number of the quantization orthogonal transformation coefficient counted by a quantization data bit number counter 7 with a predetermined threshold level. The threshold level gives a lower limit of the compression rate of a picture and specifies the allowed maximum value as a data quantity after compression. When all bit number of the quantization orthogonal transformation coefficient is larger thant the threshold level, a round-off control section 8 controls a low-order bit round-off section 5, the low-order bit of the quantization orthogonal transformation coefficient is rounded off and the result is sent as a code data. In this case, the quantity of the round-off of the low-order bit is decided by the round-off control section 8 so that all bit number of the quantization orthogonal transformation coefficient after round-off is smaller than the threshold level. Thus, the decrease in the processing speed is prevented.

Description

[Detailed Description of the Invention] Industrial Application Field The present invention relates to an image encoding method. Conventional technology Orthogonal transform encoding is known as a redundant image compression method, but this method divides an image into rectangular blocks. In recent years, orthogonal transformations include Hadamard transform, Fourier transform, discrete cosine transform, etc. Even though discrete cosine transform, in which a lot of energy is concentrated in the low-order frequency region, is often used, image data compression in the frequency space concentrates a lot of energy in the low-order region of the frequency components obtained by orthogonal transform. Taking advantage of the In order to achieve this, a method was devised that classifies rectangular blocks that undergo orthogonal transformation into multiple classes based on the statistical characteristics of their frequency distribution, and assigns an appropriate number of quantization bits and quantization characteristics to each class. For example, “Adaptive Coding of Monochrome and Color Images” I Triple E Transactions on Communications V
ol, Com-25, No. 11, November
1977 (W, Chen, and H, Smith,
'Adaptive Coding of Mono
chrome andColor Images, '
IEEE Transactions on Comm
unications, Vol.1. Com-25, No.
11, November 1977) {Above, even if blocks are classified into four classes based on the AC power of each block and adaptive quantization is performed for each class, the number of blocks belonging to each class is Problem to be Solved by the Invention Even if all the values are made to be equal, the problem that the invention attempts to solve is that with the method of classifying blocks that perform orthogonal transformation into multiple classes and quantizing them appropriately for each class, the final compression ratio may not be predetermined in advance. The total number of bits after quantization in each block varies depending on the class to which the block belongs. If an image contains many blocks of classes with a large number of quantized bits, the compression rate will deteriorate. This can be a major problem when the lower limit of the compression rate is fixed for video, etc. Also, to avoid this, when dividing blocks into classes, the ratio of the number of blocks belonging to each class is controlled so that it is a constant value. To do this, it is necessary to trace the entire image once and derive a classification standard that makes the ratio of the number of blocks belonging to each class a constant value. The present invention solves the above problems by dividing the image into blocks into rectangular areas and performing orthogonal transformation for each block. Each block is classified into multiple classes according to the statistical characteristics of the distribution pattern of orthogonal transform coefficients, and the orthogonal transform coefficients of each block are quantized and quantized according to the number of quantization bits and quantization characteristics set according to each class. Focus the orthogonal transform coefficients. Store these quantized orthogonal transform coefficients in the buffer memory for all blocks of the image, and at the same time count the total data amount. If the total data amount is smaller than a predetermined threshold, the contents of the buffer memory If the total amount of data is larger than a predetermined threshold (Friend), truncate one or more bits from the least significant bit of all quantized orthogonal transform coefficients stored in the buffer memory and send them. The number of truncated bits of the quantized orthogonal transform coefficients is determined using a means such that the total data amount of the quantized orthogonal transform coefficients after truncation is smaller than a predetermined threshold. Prior to sending the quantum orthogonal transform coefficients, the number of truncated bits and the class classification information of each block are sent as header information to perform high-efficiency encoding of the image. FIG. 1 is a block diagram showing a first embodiment of an apparatus for implementing the image encoding method according to the invention; FIG. A block (rectangular area) is extracted as a unit of 2
Even if the orthogonal transform is performed in the DCT calculation section 2, the orthogonal transform coefficients from the DCT calculation section 2 are quantized in the quantization section 3.
Observe the AC divided power of the orthogonal transform coefficients from LX., and classify each block based on the AC component power. The information on which class the block belongs to is given to the quantization data pit number counters of the quantization units 3 and 7.The quantization unit 3 allocates a predetermined number of quantization bits corresponding to the class to which the processing block belongs. The orthogonal transform coefficients are quantized using the quantization characteristic. These quantized orthogonal transform coefficients are sequentially stored in the quantization data buffer memory 4, and at the same time, the quantization data pit number counter 7.
The number of bits of the quantized orthogonal transform coefficients outputted from the quantization unit 3 is sequentially added and counted for each class, and the sum of the number of bits of each class, that is, the total number of bits is also counted at the same time. When all blocks of the image have been processed, the quantized data buffer memory 4 stores all quantized orthogonal transform coefficients corresponding to the processed image, and the quantized data pit number counter 7 stores the number of bits for each class and the total number of bits. The truncation control section 8 compares the total number of bits of the quantized orthogonal transform coefficient counted by the quantized data pit number counter 7 with a predetermined threshold value. The threshold gives the lower limit of the image compression rate, and even if it specifies the maximum allowable amount of data after compression, if the total number of bits of the quantized orthogonal transform coefficient is smaller than the threshold Truncation control unit 8 (friend) Sends the contents of the quantization data buffer memory as coded data, but if the total number of bits of the quantized orthogonal transform data is greater than the threshold value, the truncation control unit 8 outputs the lower bits of 5. The truncation unit 8 controls the lower bits of the quantized orthogonal transform coefficients and sends them out as truncated code data. Here, the amount of lower bits to be truncated is If the number is determined to be smaller than the threshold value, the class classification unit 6 and the quantization unit 3 of the apparatus shown in FIG.
, the truncation control unit 8 and the lower bit truncation unit 5 will be explained in more detail.
The result of the addition represents the AC power of the block, and is squared by the 11 adders for all the components in the block and stored in the 12 registers. Figure 3 shows an example where the number of classes is 4 and the image is orthogonally transformed using one block as an 8x8 rectangle. . In the figure, classes 1, 2, and 2 are classified in descending order of AC component power.
3. Based on the above classification, the first
The quantization unit 3 in the figure quantizes each block. For quantization, the quantization bit allocation shown in FIG. 4, for example, is performed depending on the class. More quantization bits are given to classes with larger AC component power. (a) in the figure shows class 1,
(b) is class 2, (c) is class 3, (d
) shows the bit allocation for class 4, and even though the upper left is the DC component and the lower right is the higher frequency component, the quantized orthogonal transform coefficients obtained in this way are the quantum At the same time that the converted data is stored in the buffer memory 4,
Even if the total number of bits is counted by the quantized data pit number counter 7 in FIG. 1, if the total number of bits exceeds a predetermined threshold, the truncation control section 8 and the lower bit truncation section in FIG. The first method is to truncate the lower bits of the quantized orthogonal transform coefficients even though the amount of code data is reduced. In this case, in the example shown in Figure 4 (above), note that the minimum allocation of quantization bits is 0.If the overall reduction is 1 bit, class I is 64 bits and class 2 is 49 bits. 33 bits for class 3 and 30 bits for class 4 are respectively reduced.A second method is to change the number of bits to be reduced for each class.For example, for the example in Figure 4. 3 bits from class 1, 2 bits from class 2, class 3
, 4, if we reduce 1 bit from the lower bit of the quantized orthogonal transform coefficient, then 177 bits of data for class 1, 74 bits for class 2, 33 bits for class 3, and 30 bits for class 4 will be reduced. Furthermore, a third method is to change the number of bits to be reduced depending on the frequency component of the orthogonal transform coefficient.For example, in DC precepts, the number of bits to be reduced is small<,L, and the higher the frequency, the more bits are reduced. Although it is possible to reduce the number of bits of each class counted by the quantization data pit number counter 20 and the total Referring to the number of bits, the method of reduction as described above and the number of bits are determined. After determining the parameters, the method of reducing the lower bits described above can always prevent the compression ratio from falling below the preset lower limit value. Parameter information and class classification information for each block are sent as header information, but during decoding, the reduced number of bits of the quantized orthogonal transform coefficient is recognized from the header information, and the reduced bits are set to 0 (or l) and quantized. Reproduce orthogonal transform coefficients.

Next, FIG. 5 is a block diagram showing a second embodiment of an apparatus for realizing the image encoding method according to the present invention.
17 Quantization 18 Quantization data buffer memory 19 Class classification enemy 20 Quantization data pit number counter (Friend) A friend of the embodiment: A case where the amount of each data stored in the quantization data buffer memory exceeds the threshold.In contrast to the second embodiment, where the lower bits of the quantized orthogonal transform coefficients were truncated to reduce the amount of coded data. Then, the amount of code data is reduced by zone sampling.The read data position control unit 21 in FIG. Perform zone sampling by doing this.

Zone sampling is effective only in the DC component and low frequency precept areas and deletes high frequency components. Figure 6 shows an example of the zone sampling area for an 8 x 8 block. The shaded area in the figure is zone sampling. When the zone sampling shown in Fig. 6 is applied to the quantization bit allocation of each class shown in Fig. 4, class I has 102 bits, class 2 has 23 bits, class 3 has 5 bits, and class 4 Although 2-bit data is reduced in this case, it is also possible to set the area for zone sampling independently for each class, and in this case, the amount of data can be reduced more efficiently than in the zone sampling method described above. With this method, it is possible to always prevent the compression ratio from falling below the preset lower limit.
In addition, before sending code data, frequency domain information for zone sampling and class classification information for each block are sent as header information.As explained above, the present invention has an effect on the compression rate in image data compression. This makes it possible to specify the lower limit of

[Brief explanation of drawings]

FIG. 1 shows the blocks of an apparatus for implementing a high-efficiency image encoding system in the first embodiment of this description, and FIG.
The figure shows the class classification of each block of an image. Figure 4 shows the distribution of the number of quantization bits for each class. Figure 5 shows the realization of a high-efficiency encoding method for images in the second embodiment of the present invention. Figure 6 is a diagram showing an example of zone sampling.

Claims (5)

[Claims] (1) Divide the image into blocks into rectangular areas, perform orthogonal transformation for each block, classify each block into multiple classes according to the statistical characteristics of the distribution pattern of orthogonal transformation coefficients, and classify the orthogonal transformation coefficients of each block into multiple classes. is quantized according to the number of quantization bits and quantization characteristics set according to the class to obtain quantized orthogonal transform coefficients, and the quantized orthogonal transform coefficients are stored in a buffer memory for all blocks of the image, and at the same time all of them are The amount of data is counted, and if the total amount of data is smaller than a predetermined threshold, the contents of the buffer memory are sent as is, and if the total amount of data is larger than the predetermined threshold, One or more bits are truncated from the least significant bit of all the quantized orthogonal transform coefficients stored in the buffer memory and sent out, and the number of bits to be truncated at this time is equal to the quantized orthogonal transform coefficient after truncation. The total data amount of the coefficients is determined to be smaller than the threshold value, and prior to sending out the quantized orthogonal transform coefficients, the number of truncated bits and the class classification information of each block are sent out as header information. An image encoding method characterized by: (2) If the total amount of data of the quantized orthogonal transform coefficients stored in the buffer memory is larger than the threshold, weighting the number of cutoff bits of the quantized orthogonal transform coefficients for each class;
2. The image encoding method according to claim 1, wherein control is divided into a class in which more bits are truncated and a class in which fewer bits are truncated. (3) When the total amount of data of the quantized orthogonal transform coefficients stored in the buffer memory is larger than the threshold, the number of cutoff bits of the quantized orthogonal transform coefficients is weighted by frequency component, and more bits are 2. The image encoding method according to claim 1, wherein control is performed separately into a frequency component to be truncated and a frequency component to be truncated to a smaller number of bits. (4) Divide the image into blocks into rectangular areas, perform orthogonal transformation for each block, classify each block into multiple classes according to the statistical characteristics of the distribution pattern of orthogonal transformation coefficients, and classify the orthogonal transformation coefficients of each block into multiple classes. is quantized according to the number of quantization bits and quantization characteristics set according to the class to obtain quantized orthogonal transform coefficients, and the quantized orthogonal transform coefficients are stored in a buffer memory for all blocks of the image, and at the same time all of them are The amount of data is counted, and if the total amount of data is smaller than a predetermined threshold, the contents of the buffer memory are sent as is, and if the total amount of data is larger than the predetermined threshold, Among the quantized orthogonal transform coefficients stored in the buffer memory, control is performed so that only the DC component and low frequency component regions are transmitted and the high frequency region is not transmitted, and the frequency region to be transmitted at this time is The total amount of data to be transmitted is determined to be smaller than the threshold value, and prior to transmitting the quantized orthogonal transform coefficients, the frequency domain information to be transmitted and the class classification information of each block are transmitted as header information. An image encoding method characterized by: (5) When the total amount of data of the quantized orthogonal transform coefficients stored in the buffer memory is larger than a threshold value, a frequency region to be transmitted is independently determined for each class. Image encoding method described.