JPH1066078A - Picture encoding device and picture decoding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve a signal-to-noise ratio(SN ratio) obtained after post processing and the quality of a decoded picture by compressively encoding and then decoding an original signal, finding out interruption frequency minimizing a difference between a signal obtained by band limiting the decoded signal and the original signal and outputting the interruption frequency. SOLUTION: A picture decoding device for compressivly decoding a picture by a non-reversible encoding method is provided with an encoder 1 for compressively encoding an original picture P, a local decoder 4 for decoding the compressed data, a differeratiator 5 for finding out an error picture, an orthogonal transformation means 6 for the original picture, an orthogonal transformation means 7 for the error picture, and an interruption frequency determining means 8 for finding out interruption frequency by using respective orthogonal transformation results. The means 8 finds out and outputs interruption frequency minimizing a difference between a signal obtained by band limiting a signal outputted from the local decoder 4 and the original signal so as to improve the SN ratio of the decoded picture to the original picture. The picture decoding device limits a band based on the interruption frequency.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image encoding / decoding apparatus for improving an image in which a signal-to-noise ratio (hereinafter, referred to as an S / N ratio) is reduced by high-efficiency encoding of the image by band limitation. Things.

[0002]

2. Description of the Related Art Generally, image data has a much larger data amount than character and numerical data. Therefore, it is necessary to reduce the time required for transmission using a communication line, reduce communication costs, and efficiently use a storage device such as a magnetic disk. For this reason, various encoding methods for compressing image data have been invented and put to practical use.

[0003] When an original image is restored after being compressed, an encoding method capable of restoring the original image without any distortion is called lossless encoding. On the other hand, a high compression ratio is used instead of allowing some distortion. Is called lossy coding. At present, for continuous tone images, lossy coding is generally more widely used. The present invention relates to lossy coding.

[0004] An image encoded and decoded by the irreversible encoding method is generally different from the original image, and the difference between the two is an error image corresponding to distortion. As a method of reducing this distortion, post-processing of the decoded image has conventionally been performed. For post-processing, low-pass filters have often been used. The reason for this is to improve the image quality by removing the high frequency components, which are relatively large as the frequency components of the distortion.

[0005]

However, by applying a low-pass filter to the decoded image, an effective high-frequency component contained in the original image signal is also removed at the same time. Therefore, it is desirable to find the cut-off frequency of the low-pass filter in consideration of this. However, since filters with fixed characteristics have been used in the past, post-processing that is not always ideal is performed. In some cases, the processing has an adverse effect such as a reduction in the SN ratio.

An object of the present invention is to realize a post-processing in which the S / N ratio is always improved by the post-processing, or at worst, remains unchanged without such adverse effects.

[0007]

In order to achieve the above-mentioned object, in the present invention, on the image encoding side, for example, a local decoder for decoding compressed data, a differentiator for obtaining an error image, and an orthogonal transform of an original image , An orthogonal transform of the error image, and means for obtaining a cutoff frequency using the orthogonal transform results. On the image decoding side, there is provided a band limiter that limits the band of the decoded image at the cutoff frequency obtained on the encoding side and obtains a final decoded image.

That is, a band limiter having a variable cutoff frequency is provided on the decoding side. The encoding side is provided with means for obtaining a cutoff frequency so as to optimize the SN ratio of the final decoded image output from the band limiter on the decoding side with respect to the original image. This cut-off frequency is sent from the encoding side to the decoding side, and the decoding side limits the band by the cut-off frequency obtained on the encoding side.

The operation of the present invention configured as described above is as follows.
It is as follows. In improving the SN ratio of decoded images,
The cutoff frequency given to the band limiter is determined on the encoding side, and this is transmitted to the decoding side via a communication line or the like. For this reason, on the encoding side, a function for obtaining the cutoff frequency is required. This function is realized as follows. That is, by giving the compressed data output from the encoder to the local decoder, the same decoded image as that output from the decoder on the decoding side is generated. Next, an error image is obtained by a differentiator from the original image and the decoded image obtained by the local decoder. This error image corresponds to the distortion caused by the encoding. Next, the original image and the error image obtained in this way are orthogonally transformed, and given to the cutoff frequency determination unit. The cut-off frequency determining unit determines a cut-off frequency based on these pieces of information by a method described later and sends the cut-off frequency to the decoding side.

[0010]

Embodiments of the present invention will be described below with reference to the drawings. In FIG. 1, an original image (P) 10 passes through an encoder 1 and a decoder 2 of an arbitrary encoding method (for example, JPEG, block encoding, etc.) to become a decoded image (Q) 12, and further a band limiter 3 Through to the final decoded image (R) 13. Assuming that the image is, for example, a monochrome square having N pixels on each side, P, Q, and R can all be represented by an N × N square matrix.

Assuming that the encoding system of the encoder 1 and the decoder 2 is irreversible, the decoded image (Q) 12 becomes the original image (P)
Generally different from 10. An image obtained by subtracting the original image (P) 10 from the decoded image (Q) 12 for each pixel is referred to as an error image (QP) 14.
In P) 14, the original image component is dominant in the low frequency region, and the error image component is dominant in the high frequency region. Therefore, by removing high-frequency components by the band limiter 3 having low-pass characteristics, the SN ratio of the final decoded image (R) 13 to the original image (P) 10 can be improved.

As a configuration method of the band limiter 3, there is a method using a digital filter, but in the present embodiment,
Two-dimensional orthogonal transformation is applied to the entire image as one block. That is, the band limiter 3 includes the orthogonal transform unit 3
1, a replacement means 32 for replacing some coefficients with 0, and an orthogonal inverse transformation means 33.

The orthogonal transform means 31 outputs the decoded image (Q) 12
Is converted into a coefficient matrix by an N × N two-dimensional orthogonal transform. The replacement means 32 for some coefficients at 0 is m × m
The coefficients other than the number (where m <N) are replaced with 0. The orthogonal inverse transform means 33 obtains a final decoded image (R) 13 by performing an inverse transform and returning it to the pixel area. If these are expressed by one expression,

[0014]

(Equation 1)

## EQU1 ## Here, M is the following matrix.

[0016]

(Equation 2)

The integer m is a parameter for determining the cutoff frequency. By changing the value of m, the cutoff frequency of the band limiter 3, that is, the boundary frequency between the passband and the stopband can be changed.

On the other hand, T is an orthogonal transformation matrix used for the orthogonal transformation means 31 and the orthogonal inverse transformation means 33. Also, ^t T
Is the transposed matrix with its length and width reversed. There are various types of orthogonal transforms, and among them, a discrete Fourier transform (Discrete Fouri
er Transform (DFT) and discrete cosine transform (Discrete
Cosine Transform (DCT).

In the present embodiment, DCT that is easier to handle than discrete Fourier transform is used. Generally, in DCT used for image coding, the size of a block is as small as about 8 × 8 pixels, but in the present embodiment, DCT is performed with the entire screen as one block. In this case, T is the following DCT matrix.

[0020]

(Equation 3)

Since T is a unitary matrix, the inverse matrix of T is ^tT . The value of the cutoff frequency (m) 17 is selected such that the total sum e (m) of the square error of the final decoded image (R) 13 with respect to the original image (P) 10 over the entire image becomes small. That is,

[0022]

(Equation 4)

The value m that minimizes is obtained by the local decoder 4, differentiator 5, orthogonal transform means 6, 7 and cut-off frequency determining means 8 on the encoding side, and is sent to the decoding side. Where Tr
Is the eigensum of the matrix, that is, the sum of the diagonal elements.

On the encoding side, a final decoded image (R) 13 in which the band is limited for all values of m is produced, and the value of the square error e (m) between them and the original image (P) is minimized. May be obtained, but the amount of calculation becomes enormous. Therefore, if almost equivalent processing is performed on the frequency axis as described below, the amount of calculation is small. Let X be the one obtained by two-dimensional orthogonal transformation of the original image (P) 10, and let Y be the two-dimensional orthogonal transformation of the P and Q error image (Q-P) 14.

[0025]

(Equation 5)

By substituting the above equations (1), (2), (3) and (5) into the equation (4), the following equation is obtained.

[0027]

(Equation 6)

Once X and Y have been calculated in equation (5), then, using equation (6), the value of e (m) for different values of m can be obtained with a small amount of calculation. Equation (5)
Is calculated by the orthogonal transform means 6 and 7, and the calculation of the equation (6) is performed by the cut-off frequency determining means 8.

FIG. 2 is an example of an expression based on a flowchart relating to the processing function of the cut-off frequency determining means 8 just described.
(A) to (j) in the following description are (a) to (j) shown in FIG.
Corresponds to (j).

(A) First, an orthogonal transformation (X) 15 of an original image (P) 10 and an orthogonal transformation (Y) 16 of an error image (PQ) 14 are input. Note that X and Y are calculated by the above equation (5).
Is calculated by

(B) Next, a variable emax for entering the maximum value of e (m) and a variable mmax for entering the cutoff frequency m giving the maximum value are initialized to zero. (c) After setting m to 1 first, the following processing is performed.

(D) e (m) is calculated according to the above equation (6). (e) determine whether e (m) is greater than emax,
If not, proceed to processing (g).

(F) If e (m) is greater than emax,
The value of e (m) is set to emax, and the value of m is set to mmax. (g) Add 1 to m.

(H) Judge whether m is equal to or less than N,
If N or less, the process returns to the process (d), and the process is repeated in the same manner. If it is larger than N, the process proceeds to the next process (i). (i) Let m be the value of mmax.

(J) Output the value of m, and use that value as the cutoff frequency. FIG. 3 is a diagram illustrating the principle of the above embodiment. In FIG. 3, the horizontal axis conceptually shows the spatial frequency, and the vertical axis conceptually shows the power spectrum of the original image and the error image corresponding to the distortion caused by the encoding.

Usually, the original image component is dominant in the low frequency region, and the error image component is dominant in the high frequency region. Therefore, it is intuitively understood that the cut-off of the frequency band above the frequency at which the two cross the decoded image can improve the SN ratio.

[0037]

FIG. 4 is a diagram showing experimental results of the embodiment.
The horizontal axis is the cutoff frequency m, and the vertical axis is the SN ratio. The image used for the experiment was a standard image named GIRL (N = 25).
6 pixels, 8 bits / pixel). As a coding method, a JPEG baseline sequential process and a block coding shown in the following reference were used. The encoding method referred to here is an encoding method used for encoding in the encoder 1 and decoding in the decoder 2 and the local decoder 4 in FIG. The SN ratio improvement effect was about 0.4 dB for JPEG and about 1 dB for block coding.

[References] Kishimoto, Mitsuya, Hoshida, Kamae:
"Block coding of still images," IEICE (B), Vo
l. J62-B, No. 1, 1979.

[0039]

According to the present invention, as described above, the distortion caused by encoding can be efficiently removed, and the image quality of the final decoded image is improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment.

FIG. 2 is a flowchart of cut-off frequency determining means.

FIG. 3 is a diagram illustrating the principle of the embodiment;

FIG. 4 is a diagram showing experimental results of an example.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Encoder 2 Decoder 3 Band limiter 4 Local decoder 5 Difference device 6, 7 Orthogonal transformation means 8 Cutoff frequency determination means 32 Substitution means for replacing some coefficients with 0 33 Orthogonal inverse transformation means 10 Original image (P) 11 Compressed Data 12 Decoded Image (Q) 13 Final Decoded Image (R) 14 Error Image (Q-P) 15 Orthogonal Transformation of Original Image X 16 Orthogonal Transformation of Error Image Y 17 Cutoff Frequency (m)

Claims (3)

[Claims] 1. An image coding apparatus for compressing and coding an image by a lossy coding method, comprising: coding means for compressing and coding an original signal; local decoding means for decoding compressed data output from the coding means; Means for obtaining a cutoff frequency at which the difference between the signal obtained by band-limiting the signal obtained from the output of the local decoding means and the original signal is minimized, and outputting the cutoff frequency. . 2. The image encoding apparatus according to claim 1, wherein a difference between the band-limited signal and an original signal is a square error. 3. An image decoding apparatus for decoding image data compressed and encoded by a lossy encoding method, means for inputting the compressed and encoded data and a cut-off frequency determined on the image encoding side. Decoding means for decoding input compressed data to generate a decoded image, and band limiting means for performing band limitation on the decoded image at the cutoff frequency and changing the cutoff frequency to obtain a final decoded image. An image decoding device characterized by the above-mentioned.