KR0185845B1 - A layered encoder using wavelet transform - Google Patents

Abstract

The present invention relates to a hierarchical encoder using a wavelet converter, comprising: a frame memory (31) for storing a predetermined image frame; A three-dimensional wavelet converter 32 for receiving an output of the frame memory and converting the three-dimensional wavelet; A first band separator 33 for separating low frequencies from the output of the three-dimensional wavelet converter 32; A first quantizer 34 for quantizing the output of the first band separator 33 and outputting a bit stream of the base layer; A second band separator 35 for separating the high range at the output of the three-dimensional wavelet converter; The second quantizer 36 quantizes the output of the second band separator 35 and outputs a bit stream of the second layer, thereby efficiently encoding the video signal without using motion compensation.

Description

Hierarchical Coder Using Wavelet Converter

1 is a block diagram showing a conventional video encoder.

2 is a block diagram illustrating a hierarchical encoder using a wavelet transformer according to the present invention.

3 is a diagram illustrating the concept of band division of an output of a three-dimensional wavelet converter.

* Explanation of symbols for main parts of the drawings

10: subtractor 12: discrete cosine converter

14 quantizer 16 variable length encoder

18 inverse quantizer 20 inverse discrete cosine converter

22: adder 24: frame memory

26: motion estimation unit 28: motion compensation unit

31: frame memory 32: wavelet converter

33: first band separator 34: first quantizer

35: second band separator 36: second quantizer

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a video encoding apparatus, and more particularly, to a hierarchical encoding apparatus using a wavelet transform for efficiently performing under measurement code without performing motion estimation and compensation in a wavelet transform space.

In general, motion compensation coding (Motion Compensated Coding) in the video encoding apparatus is a representative motion image coding technique for reducing the temporal correlation of the video signal through the motion estimation, the process of estimating and compensating the motion and the motion-compensated difference signal Performed by encoding. In general, since the current image may be regarded as moving only a portion of the previous image, the current image may be configured by tracking the direction in which the arbitrary block moves in the previous shape. The direction in which an arbitrary block moves in the previous image is called a motion vector, and a process for searching for this motion vector can be expressed by using Equation 1 below using a minimum absolute error.

In Equation (1), A and B are the size of a block, f (i, j) is the minimum value of the current image, and g (i + x, j + y) is the pixel value of the moved previous image.

FIG. 1 is a block diagram showing the structure of a daily video encoder, and is used in many standardized encoders such as H.261, MPEG-1, MPEG-2, and the like.

Referring to FIG. 1, the subtractor 10 sets a difference image between frames by subtracting a motion compensation image of an original image of a current frame and a reconstructed image of a previous frame, which are input in units of frames.

The discrete cosine transforming unit (DCT) 12 outputs the discrete cosine transform coefficients by performing discrete cosine transforming of the inter-frame low image into a block of 8x8 pixels, for example, to remove the correlation between pixels.

The quantizer Q 14 quantizes the discrete cosine transform scheme of the interframe difference image output from the discrete cosine transforming unit 12 at a predetermined quantization interval and outputs the quantized interval.

The variable length encoder (VCL) 16 variably encodes the inter-frame difference image quantized by the quantizer 14 and transmits the variable length coder to a decoder (not shown) through a buffer (not shown). For example, among the signals represented by 8 bits, the more frequent data is represented by fewer bits, and the less frequently data is represented by more bits, thereby reducing the total number of bits representing the difference image.

An inverse quantizer (IQ) 18 is connected to the output terminal of the quantizer 14 and restores the quantized interframe difference image to a state before being input to the quantizer 14.

The inverse discrete cosine transforming unit (IDCT) 20 is connected to the output terminal of the inverse quantizer 18 and before the inverse quantized interframe difference image is input to the discrete cosine converting unit 12 by the inverse quantizer 18. Restore to state

The adder 22 adds the inter-frame difference image and the motion compensation image reconstructed by the inverse discrete cosine transform unit 20 and stores the reconstructed image of the previous frame in the frame memory 24.

The motion estimation unit (ME) 26 generally uses a block matching algorithm, estimates a similar portion between the input current image and the reconstructed image of the previous frame stored in the frame memory 24, and moves the result of the position shift. Output as a vector.

The motion compensator (MC) 28 outputs the motion compensation image compensated by the motion vector to the subtractor 10 and the adder 22, respectively, by compensating the motion position of the reconstructed image of the previous frame stored in the frame memory 24, respectively. .

However, in the video encoding apparatus as described above, since the above-described motion estimation method requires a large amount of computation along with a discrete cosine transform, etc., not only does it require a lot of time for encoding, but also there is a problem in that hardware for implementing this is complicated.

Accordingly, an object of the present invention is to provide a hierarchical encoding apparatus using 3D wavelet transform for efficiently performing hierarchical encoding without performing motion estimation and compensation in 3D wavelet transform space. There is this.

In order to achieve the above object, the device of the present invention comprises a frame memory for storing a predetermined image frame; A three-dimensional wavelet converter which receives the output of the frame memory and converts the three-dimensional wavelet; A first band separator separating the low range at the output of the three-dimensional wavelet converter; A first quantizer for quantizing the output of the first band separator to output a bit stream of a base layer; A second band separator separating the high range at the output of the three-dimensional wavelet converter; And a second quantizer configured to output the bit stream of the second layer by quantizing the output of the second band separator.

That is, according to the present invention, after converting a 4-frame image into a three-dimensional wavelet, the low frequency band is separated into a low band and a high band, and a high band is quantized by a first quantizer having a small step size, and a high band is quantized by a second quantizer having a large step size and a base layer. And to the second layer.

Hereinafter, with reference to the accompanying drawings will be described in detail with respect to the present invention.

First, the wavelet transform will be described in general for easy understanding of the present invention.

Since a large amount of data is required to store or transmit a digital image, compression methods for expressing the image with a minimum amount of information have been studied.

Representative methods proposed so far to compress image data include predictive coding, transform coding, subband coding, vector quantization, and combination coding thereof (hybrid coding). ).

Recently, wavelet transform encoding has been studied as a new transform encoding method.

The basic concept of wavelet transform is to express an arbitrary function as a linear superposition of a wavelet basis function that is simultaneously localized in the time-frequency domain. The base functions in the wavelet transform are created by stretching / expanding the prototype wavelet function ψ (x) as in equation (2) and translating.

Here, a is a scale variable for stretching / expanding the circular wavelet, and b is a transition variable. Discrete expressions are possible by adjusting the coefficients as shown in equation (3).

Wavelet decomposition is as shown in the following equations (4) and (5).

At this time, when a ₀ = 2 and b ₀ = 1, it is possible to select Ψ in which _{m, n} constitute a normal orthogonal basis. In multi-resolution analysis, the basis function actually has two functions, one is a wavelet function and the other is a scaling function, as shown in Equation (6).

The wavelet function has a bandpass filter characteristic, and the scaling function has a lowpass filter characteristic, respectively. The resolution is that the approximated function of 2 ^m can be expressed as a _{m, n} (f) = Φ _{m, n} f) using the scaling function _{, where} c _{m, n} (f) = Ψ _{m, n} f is Shows the missing information when approximating a function with a resolution of 2 ^m to a function with a resolution of 2 ^m-1 . If this is implemented as a filter, it is the same as Formula (7), Formula (8), and Formula (9).

h _n and g _n are the low pass filter and the high pass filter, respectively. Synthesis after decomposition in this way is as shown in the following equation (10).

Instead of normal orthogonal wavelets with nonlinear phase characteristics, the linear phase characteristics can be obtained using bi-orthogonal wavelets with different base functions in decomposition and synthesis. If the wavelet transform is extended in two dimensions, the two-dimensional scaling function Φ (x, y) exists as in one dimension, and if the one-dimensional wavelet function Ψ (x) is for the one-dimensional scaling function Φ (x) The dimensional wavelet function is defined as equation (11).

Here, H, V, and D represent horizontal, vertical, and diagonal directions of the two-dimensional wavelet function, respectively. The two-dimensional wavelet transform may be interpreted as splitting the signal into independent, directional frequency bands. The original image is called R _m and the wavelet transform reduces the resolution by half This happens, Converting to wavelet again reduces the resolution in half This occurs. Where the superscript indicates directionality and the subscript indicates resolution.

Subsequently, the hierarchical quantizer using the three-dimensional wavelet converter according to the present invention includes a frame memory 31 for storing a predetermined image frame, as shown in FIG. A three-dimensional wavelet converter 32 that receives the output of the frame memory 31 and converts the three-dimensional wavelet; A first band separator (33) for separating low frequencies from the output of the three-dimensional wavelet converter (32); A first quantizer (34) for quantizing the output of the first band separator (33) to output a bit stream of the base layer; A second band separator (35) for separating high frequencies at the output of the three-dimensional wavelet converter (32); And a second quantizer 36 for quantizing the output of the second band separator 35 and outputting a bit stream of the second layer.

In FIG. 3, the frame memory 31 is capable of storing four frames, and the output of the frame memory 31 is three-dimensional wavelet transformed by the three-dimensional wavelet converter 32.

The wavelet converter 32 may output a moving picture signal input from the frame memory 31 and the video signal output by the wavelet transform on a three-dimensional space as shown in FIG. ) Is separated by the first band separator 33 and quantized by reducing the step size in the first quantizer 34 and output as a bitstream of the base layer, and the high band L2 is the second band separator 35. ), The step size is increased, quantized, and output as a bit stream of the second layer.

As described above, the video encoding apparatus according to the present invention uses a wavelet transform that is easy to give priority to each division band, thereby satisfying the user's demands compared to the conventional encoding apparatus using the discrete cosine transform. There is an advantage that it is easy to transmit the video.

In addition, by eliminating the motion estimation and compensation circuit by the three-dimensional wavelet transform, not only the time required for encoding can be shortened but also the hardware implementation can be simplified.

Claims (2)

A frame memory 31 for storing a predetermined image frame; A three-dimensional wavelet converter 32 that receives the output of the frame memory and converts the three-dimensional wavelet; A first band separator (33) for separating low frequencies from the output of the three-dimensional wavelet converter (32); A first quantizer (34) for quantizing the output of the first band separator (33) to output a bit stream of the base layer; A second band separator 35 for separating high frequencies from the output of the three-dimensional wavelet converter and a second quantizer 36 for quantizing the output of the second band separator 35 to output a bit stream of a second layer A hierarchical encoder using wavelet transform. The layer encoder of claim 1, wherein the frame memory is configured to store image data of four frames.