JPH02132971A - Adaptive discrete cosine coding method - Google Patents

Abstract

(57) [Abstract] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] This invention relates to an adaptive discrete cosine encoding method used for image compression.

[Prior Art] In recent years, as an image compression method in fields such as videophones, itI cosine transform has been widely used as an orthogonal transform method with a high degree of energy concentration, along with Karhunen-Loeve transform.

This kind of discrete cosine transform divides the image signal into frequency components and expresses them with the same number of cosine waves as the number of input samples, which concentrates energy and encodes only the components with a large amount of energy. By doing so, image compression is achieved.

By the way, in the adaptive discrete cosine transform encoding method that encodes the image signal using the discrete cosine transform, information is expressed by assigning bits and numbers according to the result of the discrete cosine transform. To encode an element assigned to a certain number of bits, positive or negative information (1 bit)
It is expressed as encoded data consisting of a numerical value (allocated bit number - 1 bit) and normalized by normalization factor information,
To decode this, the positive/negative information (1 bit) and the normalized numerical value are denormalized using normalization factor information.

[Problem to be solved by the invention] However, according to such a method, for example, for an element assigned to 1 bit, this element is extremely "0".
Even if the value is close to In an image in which a large number of elements are assigned to bits, an image that should actually be flat after inverse cosine transformation becomes a highly uneven image, resulting in a significant deterioration in image quality.

In addition, if the normalized value of an element assigned to 2 or more bits is "0", encoded data will be transferred for the assigned number of bits, which has the disadvantage of reducing encoding efficiency. there were.

The present invention has been made in view of the above circumstances, and it is an object of the present invention to provide an adaptive discrete cosine encoding method that makes it possible to improve image quality and encoding efficiency.

[Means for Solving the Problems] This invention determines whether the normalized value of the element assigned to the number of bits is "0" or not, and if it is rOJ, only 1-bit data presence/absence information is transferred. However, if the value is other than "0", encoded data is transferred together with the data presence/absence information.

[Operation] As a result, even if the value of the element assigned to the bit is close to "0", it can be accurately decoded as rOJ from the content of the 1-bit data presence/absence information, so the image after inverse cosine transformation can be decoded as rOJ. This solves the problem of images that should be flat but have severe irregularities, and is very effective in improving image quality.

Also, for elements with bit allocation of 2 bits or more,
If the normalized value is "0", it can be expressed using only 1-bit data presence/absence information, so it is possible to improve the efficiency of encoding.

[Example code] An embodiment of the present invention will be described below with reference to the drawings.

The encoding procedure will be explained with reference to FIG. 1(a).

First, in step 1, the image signal is divided into blocks of, for example, 16×16 pixels. Next, proceeding to step 2, a two-dimensional discrete cosine transformation is performed for each divided block.

Each element obtained as a result is then sent to step 3, where it is classified into four classes based on the alternating current energy. The four classes are those obtained by calculating the sum of squares of AC components within one bronc and dividing the sum into four in descending order of magnitude. This class determines the bit assignment for each subsequent element. For example, the original pixels are 256 x 256 pixels and are divided into 1. 6
Divide into blocks of X 1 6 pixels, and each block has 1
An example of a classification table in the case of classification into 4 classes is shown in FIG. 2(a). Numbers 1 to 4 in this table indicate classes, and one number corresponds to one block. Also, based on the classification information obtained in step 3, in step 4 the variance matrix of all Bronoks in the class is calculated,
The bit allocation is determined based on the magnitude of the variance value, and further, in step 5, a normalization factor is determined so that each element in the block conforms to the binoto allocation determined in step 4. A normalization factor is obtained for an element whose normalized value has a number of allocated bits other than 0 when normalized. FIG. 2(b) shows an example of a bit allocation table for each class, and the numbers in the table indicate the number of bits to be allocated. Here, the bit assignment "8" at the top left is the DC component. Further, FIG. 2(c) shows normalization factors for each class. That is,
For example, the normalization factor for a 1-bit allocation for class 1 is r7
．． It is 40J. In this state, in step 6, step 2
Each element given by , 4, and 5, and
Now outputs encoded code. Note that normalization is a process of dividing data by a normalization factor so that the data falls within a predetermined number of bits during encoding, and then encoding the data.

FIG. 3 is a flowchart for explaining in detail the steps from normalization to encoding described in FIG. In this case, in step A1, the class information of the block currently being encoded is read from the classification information table obtained in step 3 above. Next, in step A2, the focus assignment of the corresponding class is read from the bit assignment information table obtained in step 4 above. Then, the process proceeds to step A3, and it is determined whether the bit allocation read out in step A2 is rOJ. Here, if it is determined to be "0", the process proceeds to step A13, and it is determined whether the processing for all elements has been completed.
Return to , and if YES, the process ends. On the other hand, if No is determined in step 83, the process proceeds to steno knob A4. In step A4, corresponding data is read from the block subjected to the two-dimensional discrete cosine transform in step 2 above. Then, in step A5, the normalization factor of the corresponding class is read from the normalization factor table obtained in step 5 above, and in step A6, the data/normalization factor is calculated,
Find the normalized value. Next, in step A7, if the calculation result is rOJ, the process proceeds to step A8, where the data presence/absence information is set to "0", and the process proceeds to step A12, where only 1-bit data information is transferred. .

On the other hand, if the result in step A7 is No, the process advances to steno knob A9 and data presence/absence information is set in rOJ. Next, the process proceeds to step AIO, where 1-bit positive/negative information is set from the normalization result of step A6. Furthermore, the process proceeds to step All, where encoding is performed. In this case, step A
Bit allocation information from the lower bit of the normalization result of 6 -
1 (subtracting the positive and negative information) is read, and this is encoded as is as a bit pattern.

And step A1. Proceeding to step 2, encoded data is transferred together with 1-bit data q no information. Then step A1. 3, it is determined whether the processing for all elements has been completed. If NO, the process returns to step A2; if YES, the process ends.

Here, for example, when the focus assignment is 5 bits, the normalization factor is 65.30, and the data is 400, the encoded data is rlo. It is represented by 6 bits of 0110J. Here, the last 1 bit is data a-no information, the next 1 bit is positive/negative information, and the remaining 4 bits represent normalized data (bit allocation -1). Further, encoded data when the data is 2.0 is represented by one bit of "0". One bit here represents data or no information.

Next, the decoding procedure will be explained with reference to FIG. 1(b).

In this case, the class information table, bit allocation table, and normalization factor table used for encoding are transferred in advance as additional information. First, in step 11, the classification information 12, bit allocation information 13, normalization factor 14, and encoded data are decoded, and in step 15, the classification information 12, bit allocation information 13, and data no information are used to decode the inverse quantum , then hand 1111
In step 16, using the normalized consonant value obtained in step 15 and the normalization factor decoded in step 14, the inverse normalization is performed as follows: inverse binary value x normalization factor. After performing the conversion, inverse cosine transformation is performed in step 17 to complete the decoding.

FIG. 4 is a flowchart illustrating in detail the steps from decoding to denormalization described in FIG. In this case, the class information of the block currently being decoded is read out from the class information table in SteSub B1. Next, proceed to step B2, and from the bit allocation information table,
Read the number of bits allocated to the corresponding class. Then, in step B3, the bit allocation is set to "0".
”, and if it is “0”, proceed to step B6,
Set data “0” and transfer this data to step Bll.
Stored at . On the other hand, if No in step B3,
Proceeding to step B4, data information (1 bit) is read from the compressed code. Then, in step B5, step B
Determine whether the value read in step 4 is "0" or not. Here, YE
If S, the process advances to step B6 and the data “0” is set as above.
” and store this data in step BI]. On the other hand, if No, the positive/negative information (1 bit) is read and a sign is set in the data in step B7. Furthermore, in step B8, (Binoto allocation −1
) bit and set the corresponding data. Then, the process proceeds to step B9, where the normalization factor is read from the normalization factor table, and in step BIO, the normalization factor is read out from the normalization factor table.
The data read out is multiplied by a normalization factor to perform inverse normalization. The data obtained as a result is then stored in step Bll. Next, in step B12,
It is determined whether the processing for all the elements has been completed, and if the answer is No, the process is again written to SteSob A2, and if the answer is YES, the process is ended.

It should be noted that the present invention is not limited to -1 only the embodiment described above, but can be implemented with appropriate modifications within the scope without changing the gist.

[Effect of the invention] This invention determines whether the normalized value of the element assigned to the number of bits is "0" or not, and if it is "0", outputs only 1-bit data nail information. , if the value is other than "0", encoded data is output together with the data presence/absence information.

As a result, even if the value of an element assigned to a bit is close to "0", it can be decoded as "0" accurately from the content of 1-bit data presence/absence information, and after inverse cosine transformation, The image should be flat, but
Since it is possible to solve the problem of an image having severe unevenness, it is possible to significantly improve the image quality even for an image that includes many elements that can be assigned to one bit. In addition, even for elements with bit allocation of 2 bits or more, if the normalized value is "0", it can be expressed with only 1 bit of dataless information, which can greatly improve encoding efficiency. can.

[Brief explanation of the drawing]

FIGS. 1(a) and 1(b) show an embodiment of the present invention, in which (a) shows the order of encoding and (b) shows the order of decoding, and FIG. 2(a) shows the same embodiment. FIG. 3(b) is a diagram showing an example of the bit allocation table; FIG. 4(c) is a diagram showing an example of the normalization factor; FIGS. , is a flowchart for explaining the same embodiment in detail. 1-9, 11-17...Steps. Applicant's representative Patent attorney Takehiko Suzue Figure 2oooooooooooooooooooooooooooo00QoOo ooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo c. r. - P O O O O O O O O O O
O O O OIN---. -00000000000
00NNN+-1ooOo00ooOO +2+NNk+v---000000004 F1 M
7N M-w-I+OC) O 0 0 00 arm 0
0~he~he1. --=OOOOOO Arm circle circle ~ 1- y - 0 0 0 0 0 ~ yen 0 ~ hehehe. − − − 0 0 0 0(s4
c-a-v-% O O O1'1 Fl ~N
w- w- O PoOoOoooQooOooooo ooooooOoooooooo OoQoooOoOooOooooo oooooOooOoooOooOo OooOooooooooOooo oooooQoooOoQoooOo oooOoooooOo Oooooo e-? - POOOOOOOOOOOOOO-=-000
0000000000 He PFooooOooOoooOoQ He 1 PPoOoOQ00oOoQ ~He 1, -. −． −． −． -00000000NNN1111111000000hehe1Ntoc-a CQ - - - ” O O O
O O size N~hehe-y- v- --0 0 C
) C) 0 ward co yen to hehe N to 111111. Oohehe+sNm m m Fl r- − + NNIN F
) FI FJ <M Fl (T'l cwaCwaC'
l v- 111+- ,e”4 FJ l'1 %F
To the 11- of the bar-circle-bar n - to the N-heguq meeting NNq to the 11- to the 1N~-- +s ('I N dimension bar-1-to 11N edge +s49
Gu111-Nyen11e”J 1-J1+-G-+7-
+-+-n's armpit length Nguichi +-gu ON yen one's gugu hehe Ngu--< +s heguri bar, q to q + +gugu C
W-h-J l'Q-S F'l-. T Gu C) m FI F
lhehe11ooOoooooooooooooooooo
ooooo oOooOoooOQoooo00 oooOooQooooooOoQ0 . - 1, -0000000000000? I NNP
P O O O O O O ■ CI 0 0 0FI
N csr +-- --0 0 0 0 0 0
0 0 00hehehe=””---00000 L'? To NN of 1 To NN To 1110000 yen Hehehehehehehe 1--0000

Claims (1)

[Claims] In a system where elements to which the number of bits is assigned according to the result of two-dimensional discrete cosine transform are normalized by a normalization factor and then encoded, it is determined whether the numerical value of the normalized element is "0" or not. An adaptive discrete cosine code characterized in that, if it is "0", only 1-bit data presence/absence information is transferred, and if it is other than "0", encoded data is transferred together with the data presence/absence information. method.