JPH04196695A - Image data compression device - Google Patents

Abstract

PURPOSE:To reduce the amount of code generation by inserting zeros in the places where a certain amount of zero coefficients continuously appear in the quantized coefficients and the coefficients which come after these coefficients are assumed to have no effect. CONSTITUTION:The picture data compression device is provided with a block conversion means 102 dividing picture signal into two dimensional blocks of NXN (N is an integer) picture elements, a difference means 103 taking a difference between the block signal and the picture signal which is one frame prior one, an orthogonal transformation means 105 orthogonally transforming the difference block signals and outputting the transformation coefficients, a quantization means 106 quantizing the transformation coefficients and outputting the quantized coefficients and a variable length coding means 108 making the quantized coefficients into variable length coding. And zero coefficient discrimination means 107 checks the values of the coefficients for zero or non-zero in the direction from the lower frequency components to the higher frequency components for all NXN coefficients which are the outputs of the quantization means 106, makes all zero coefficients, if zero coefficients continuously appear for n numbers and outputs them. Thus, the degradation of picture quality is prevented and a reduction of the amount of code generation is realized.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an image data compression device that orthogonally transforms and quantizes an image signal.

[Conventional technology]

Since the amount of image information is generally enormous, when storing or transmitting it, some kind of information compression is performed. As one of the information compression means, C
There is a px64 kbit/s video encoding system (H, 261) for audiovisual services that is being standardized in CITT and 5GXV (transmission systems and equipment). In H,261, the input image signal is divided into two-dimensional blocks of 8 x 8 pixels and subjected to orthogonal transformation (discrete cosine transformation).
After being quantized, it is variable length coded and output. Huffman encoding is used for this variable-length encoding, and a continuous zero (run) followed by a non-zero if
! The code word is determined by (level).

Conventionally, the output of the quantization means is variable-length coded to obtain the generated code amount, and the generated code amount is controlled by changing the quantization step according to the result. The control method is
As stated in D-45 of the 1989 Institute of Electronics, Information and Communication Engineers Autumn National Conference, the output of the quantization means is variable-length coded to obtain the generated code amount several times by changing the quantization step, and the result is The quantization step for quantizing with a set code amount is determined from the above.

[Problem to be solved by the invention]

However, in such conventional technology, the set code amount is set to 64.
When the rate is as low as kbps, the quantization step size becomes considerably large, and coefficients close to DC coefficients are roughly quantized, so quantization noise and block distortion appear prominently in the decoded image, making it difficult to see the image. This has the problem of making it unsightly, and the problem of increasing processing time because feedback control is performed to achieve the set code amount.

Therefore, the present invention is intended to solve these problems, and its purpose is to reduce the deterioration in image quality between the original image and the decoded image, and to reduce the amount of generated code.
Furthermore, a method for preventing an increase in processing time is provided.

[Means to solve the problem]

The image data compression device of the present invention converts image signals into N×N (N
・Blocking means for dividing into two-dimensional blocks of pixels (integer); difference means for taking the difference between the block signal of N×N pixels that is the output of the blocking means and the image signal of one frame before; and the output of the difference means. orthogonal transform means for orthogonally transforming a differential block signal of N×N pixels and outputting N×N transform coefficients; and quantization means for quantizing the transform coefficients and outputting N×N quantized coefficients. In an image data compression device equipped with variable length encoding means for variable length encoding quantized coefficients, in all N×N coefficients output from the quantization means, the coefficients are changed from low frequency components to high frequency components. The present invention is characterized in that it includes a zero coefficient determining means that checks whether a value is zero or non-zero, and when n zero coefficients appear consecutively, sets all subsequent coefficients to zero and outputs them.

〔Example〕

Hereinafter, the present invention will be explained in detail based on drawings showing embodiments thereof.

FIG. 1 is a block diagram when the present invention is applied to an H.261 image encoder.

In FIG. 1, a human input image signal from an input terminal 101 is divided by a blocking means 102 into macroblocks consisting of 8×8 blocks, 4 blocks of luminance signals, and 2 blocks of color difference signals, a total of 6 blocks. Since prediction is normally performed between frames, the switches 113 and 114 are connected to the lower side as shown in FIG. 1 by the encoding control 104. The image signal output from the blocking means 102 in units of macroblocks is first manually inputted to the motion compensation prediction means 111,
Motion vector detection calls the signal of the macroblock with the minimum prediction error from the previous frame signal), and the signal of that macroblock is output. After this signal passes through a loop filter 112 for each block, a difference between it and the human image signal is taken by a difference means 103. This difference means is transformed by a discrete cosine transformation means 105, and the output is zigzag scanned. And quantization means 106
is quantized and output to the zero coefficient discriminating means 107. The output from the zero coefficient discriminating means 107 is transmitted to the variable length encoding means 1.
08, the signal is Huffman encoded and output to the output terminal 115. Furthermore, the output of the zero coefficient discriminating means 107 is dequantized by the dequantizing means 10 and inverse cosine transformed by the inverse discrete cosine transform means 110, and then the signal of the block indicated by the motion vector is passed through the loop filter 112. It is added to the passed signal and recorded in the motion compensation prediction means 111.

On the other hand, in the case of the first frame of an image or a scene change, prediction is performed within the frame, so the encoding control 104 switches the switch 113 and the switch 1
14 is connected to the upper side. The image signal output from the blocking means 102 is transformed block by block by the discrete cosine transform means 1.05, quantized by the quantization means 106, and outputted to the zero coefficient discrimination means 107. The output from the zero coefficient discriminating means is Huffman encoded by the variable length encoding means 108 and output to the output terminal 115. Furthermore, the output of the zero coefficient discrimination means is dequantized by the dequantization means 109 and subjected to inverse cosine transformation by the inverse discrete cosine transformation means 110, and then recorded in the motion compensation prediction means 111.

Here, the processing by the zero coefficient discriminating means 107 will be explained according to the flowchart of FIG.

First, 8×8 quantized coefficients, which are the output of the quantization means 106, are manually input to the zero coefficient determination means 107. Next, count, which indicates the number of consecutive zero coefficients, and rlag, which indicates whether the number of consecutive zero coefficients exceeds a threshold, are initialized to zero. (201). Here, flag is O if the number of consecutive zero coefficients does not exceed the threshold n (1≦n≦63)
If the threshold value n is exceeded, the value becomes 1. Next, it is determined whether the flag is zero or non-zero (202), and if the flag is zero, it is checked whether the coefficient is zero or non-zero (203). If the coefficient is non-zero, count is initialized to zero (204) and the process returns to flag determination (202). Also, if the coefficient is zero, co
Add 1 to un t (205), and then check whether count exceeds the threshold n (206). If the count is less than or equal to the threshold n, the process returns to flag determination (202), or if it is greater than or equal to the threshold n, 1 is assigned to the flag (202).
07), the process returns to rlag determination (202). Further, if f'lag is 1 in flag determination 202, all subsequent coefficients are set to zero and output.

Here, the coefficients are examined from low frequency components to high frequency components.

FIGS. 3(a) and 3(b) show specific examples thereof. Here, FIG. 3(a) shows the quantized coefficients manually entered by the zero coefficient discriminating means. Here, if the threshold value n is 5,
The output from the zero coefficient discriminating means is as shown in FIG. 3(b). Also, the Color Test Chart No. 1 of the Institute of Image Electronics Engineers
When the 11 images of 1 are variable-length encoded using an intra-frame mote of H, 261 and a quantization step of 5, the amount of code generated is 107 kbit when the present invention is not applied, whereas when the present invention is applied In that case, 74kbi for n-5
t and n-3, it is 55 kbit.

In the H2PO4 reference model (RM8), the quantization step size Q (2≦Q≦62) is changed for each GOB according to the output buffer bit occupancy rate.
It controlled the amount of code generated in a frame. With this method, there is no problem when the correlation between 008s is strong, that is, when quantization with the same quantization step size Q generates approximately the same amount of code, but when the correlation between 008s is weak In this case, the quantization step size Q varies greatly from GOB to GOB, and the image quality deteriorates partially within one frame. However, when the present invention is applied, there are no GOBs in which the generated code amount becomes extremely large, and the quantization step size is hardly changed, preventing extreme deterioration of image quality. Furthermore, the value of the threshold n also depends on the bit occupancy rate of the output buffer.
If it is possible to change the quantization step size Q from about -3 to about 8, the quantization step size Q need hardly be changed, so that uniformity of image quality in one frame can be maintained.

〔Effect of the invention〕

As described above, the present invention has the effect that the amount of generated code can be reduced without increasing the quantization step size too much. That is, when performing two-dimensional encoding of consecutive zeros (runs) and subsequent non-zero values (levels), it is effective to increase the number of zeros, and in the present invention, if the zero coefficients are continuous to a certain extent, By setting the subsequent coefficients to zero because they are not dominant to the image, it is possible to reduce the amount of generated code without increasing the quantization step. Furthermore, among the relatively finely quantized coefficients, low-frequency component coefficients remain, so although the decoded image has somewhat sharp edges, it is smooth overall, and the quantization You can reproduce easy-to-see images with less noise and block distortion. Furthermore, since the amount of generated code can be reduced, the number of dropped frames in the video encoder is also reduced, and a natural decoded video can be obtained.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of an image data compression device of the present invention. FIG. 2 is a flowchart of the zero coefficient discrimination means. FIG. 3 is a diagram for explaining the zero coefficient discrimination means. 101... Input terminal 102... Block division means 103... Differential means 104... Encoding control 105... Discrete cosine transformation means 106... Quantization means 107... Zero coefficient discrimination means 108 . . . Variable length encoding means 109 . . . Inverse quantization means 110 . . . Inverse discrete cosine transformation means 111 . ...Output terminals and above Applicant: Seiko Epson Corporation Agent: Patent attorney Kizobe Suzuki (and 1 other person) Figure 1 Figure 2

Claims (1)

[Scope of Claims] Blocking means for dividing an image signal into two-dimensional blocks of N×N (N: an integer) pixels, and a block signal of N×N pixels that is the output of the blocking means and one frame previous block signal. A difference block signal of N×N pixels, which is the output of the difference means, is orthogonally transformed by a difference means for taking the difference from the image signal, and then
orthogonal transform means for outputting ×N transform coefficients; quantization means for quantizing the transform coefficients and outputting N×N quantized coefficients; and variable length encoding for variable length encoding the quantized coefficients. In the image data compression apparatus, the method includes checking whether the coefficient values are zero or non-zero in order from low frequency components to high frequency components among all N×N coefficients output from the quantization means, and determining whether the coefficient values are zero or non-zero. n consecutive coefficients (n: integer, 1
≦n≦63) An image data compression device characterized by comprising: zero coefficient discriminating means which, when appearing, sets all subsequent coefficients to zero and outputs them.