KR100200621B1 - Method for compressing video signal through wavelet transform using human visual system - Google Patents

Abstract

Disclosed is an image compression method according to the present invention.

The method for dividing an input video signal using the wavelet transform according to the present invention, and quantizing the divided data to variably encode the video signal to compress the video signal includes the characteristics of the bandpass frequency and the human visual system of the wavelet transform. Obtaining an integral ratio with a modulation transfer function (MTF) using a spatial frequency characteristic, and determining a quantization step size based on an inverse value of the integral ratio.

Therefore, according to the present invention as described above, by using the quantization coefficient obtained by the integral ratio of the frequency response of the modulation transfer function of the human visual system to quantize the wavelet transformed signal and variable length-encoding the result, the compression efficiency is increased Has an effect.

Description

Wavelet transformed video signal compression using human visual system.

The present invention relates to a video signal compression method, and more particularly to a wavelet transform video signal compression method using the characteristics of the human visual system.

Transform coding technique uses energy redundancy in the original image to divide the correlated signal into the minimum uncorrelated spectral coefficient as much as possible to obtain the energy concentration effect. Although transcoding shows excellent performance, block effects occur in reconstruction due to discontinuities between blocks. In order to prevent this phenomenon, various methods, such as low pass filter, block overlapping, Lapped Orthogonal Transform (LOT), and Wavelet Transform, are used in the transformation stage of an image processing system.

When implementing an image processing system, an object is to maintain the required image quality at the minimum cost or to obtain the maximum image quality under a given resource. Image information is ultimately recognized by human vision, so you only need to pay attention to information that can be detected with your eyes. In addition, since image processing equipment such as a camera or a monitor has inherent frequency characteristics, it cannot represent all spatial frequencies of an image. Therefore, it would be uneconomical to allocate resources (generally bits) to represent image information that cannot be detected by human vision or represented by image processing equipment. Therefore, development of an optimal image processing system using a human visual system is required.

SUMMARY OF THE INVENTION The present invention was created to meet the above-mentioned requirements, and is quantized by the quantization level of each subband by the integral ratio of the frequency response of the wavelet transform and the modulation transfer function of the human visual system, and the result is variable-length encoded. It is an object of the present invention to provide a wavelet transformed image compression method using a human visual system that increases efficiency.

1 is a view showing a wavelet transform image compression and reconstruction apparatus using a human visual system according to the present invention.

2A to 2B are diagrams for explaining decomposition and synthesis of two-dimensional signals by wavelet transform.

3 is a diagram illustrating a band divided by a wavelet function.

4 is a diagram illustrating a band divided by a scaling function.

5 is a diagram illustrating frequency response characteristics of respective resolutions of the Vetterlli 22-tap filter.

6A to 6D are diagrams illustrating the frequency transfer characteristics of a modulation transfer function and a wavelet basis in a frequency domain to obtain an integral ratio quantization step interval.

In order to achieve the above object, according to the present invention, a method of dividing an input video signal using a wavelet transform, and quantizing the divided data to variably encode the video signal to compress the video signal includes a band pass of the wavelet transform. Obtaining an integral ratio with a modulation transfer function (MTF) using spatial frequency characteristics among the characteristics of the frequency and the human visual system, and determining a quantization step size based on the inverse of the integral ratio.

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

In the present invention, the wavelet transform can increase the coding efficiency by combining an octave band scale function in the spatial frequency domain. Since this process is similar to the human visual system, the coding efficiency can be improved by combining the wavelet transform and the Modulation Transfer Function (MTF) of the human visual system (HVS). The quantization level of each subband is determined from the integral ratio of the wavelet coefficient and the frequency response of the modulation transfer function.

1 is a view showing a wavelet transform image compression and recovery apparatus using a human visual system according to the present invention.

In FIG. 1, reference numeral 10 denotes an encoder, reference numeral 12 denotes a decoder, and reference numeral 14 denotes a quantization coefficient table for determining a quantization level during quantization and inverse quantization in the encoder 10 and the decoder 12, respectively. Indicates. In the present invention, encoding and decoding are performed by wavelet transform, and the transformed video signal determines the quantization level of each subband based on the integral ratio of the frequency response of the modulation transfer function of the human visual system.

In addition, the encoder 10 encoding the input video signal is composed of a wavelet transform unit 101, a quantization unit 102 and a variable length encoder 103, the decoding unit for decoding the compressed video signal is variable It consists of a long decoding unit 121, an inverse quantization unit 122 and an inverse wavelet transform unit 123. In addition, the quantization unit 102 and the inverse quantization unit 122 use a quantization coefficient table 14 that determines the quantization interval of each subband by the integral ratio of the frequency response of the modulation transfer function of the human visual system. In the present invention, since the decoding method is almost an inverse transform of the encoding method, the coding will be mainly described.

In the case of compressing or decompressing an image using wavelet transform, an input image decomposition process in the encoder and a signal synthesis process in the decoder are required.

2A to 2B are diagrams for explaining decomposition and synthesis of two-dimensional signals by wavelet transform, and FIG. 2A is a view for explaining wavelet decomposition of two-dimensional signals, and FIG. 2B illustrates wavelet synthesis of two-dimensional signals. It is a figure for demonstrating.

In order to split the 2D input image by wavelet transform, a filter having a pyramid structure as shown in FIG. 2A is provided. The input image is read in the horizontal direction, passed through the low pass filter (g) and the high pass filter (h), and the output passed back through the low pass filter (g) and the high pass filter (h) again in the vertical direction. In other words, one wavelet transform is performed. You can repeat this process several times to continuously convert the wavelets. Here, the low pass filter g shows the function of the scaling function Φ (x), and the high pass filter h performs the function of the wavelet transform function Ψ (x). In order to obtain high frequency resolution in the low frequency band, wavelet analysis can be performed by repeating the same filter structure for the low frequency band while leaving the high frequency band intact, that is, by constructing an iterated filter bank.

In order to reconstruct the divided image, the reverse process may be performed using a filter similar to that of dividing the image, as shown in FIG. 2B.

The wavelet transform functions to separate the input image into partial images having different resolutions corresponding to different frequency bands.

The wavelet transform function is derived by the movement and expansion contraction of one wavelet prototype. The wavelet transform function Ψ is shown in Equation 1 below.

[Equation 1]

Ψa, b (t) =

Here, a denotes a variable of a scale function of stretching or expanding a circular wavelet, and b denotes a moving variable representing movement. In other words, if the value of a is large, it becomes the base circle extension, and this becomes the basis of the low frequency signal. On the contrary, if the value of a is small, it is the basis for the high frequency signal. The continuous wavelet transform function is expressed as Same as 2.

[Equation 2]

end If the space created by In The closure of is as shown in Equation 3 below.

[Equation 3]

Becomes

Then the total space Is all subspace It can be expressed as a direct sum of. That is, the following equation (4).

[Equation 4]

Also, elements of all functions f (x) May be decomposed as shown in Equation 5 below.

[Equation 5]

If wavelet function Ψ is orthogonal basis, all subspaces Also orthogonal. Now, the direct sum of Equation 4 is expressed as the following Equation 6 by orthogonal intersection of subspaces.

[Equation 6]

The division of the frequency domain according to equations (4) and (5) is shown in FIG.

Also, other subspaces Is defined as in Equation 7.

[Equation 7]

Subspace with no overlap in space Unlike subspace Are superimposed and satisfy the property of Equation 8 below.

[Equation 8]

A subspace that satisfies the property shown in Equation 8 is shown in FIG. 4.

Subspace Basis function to generate Is called a scaling function, and is defined as in Equation 9 below.

[Equation 9]

Subspace Is generated by the wavelet function Ψ Is generated by the scaling function.

Various modulation transfer functions have been proposed to use the visual response characteristics of spatial frequencies in image coding. Under the condition of isotropic conditions, the human visual system is expressed by Equation 10 below.

[Equation 10]

In Equation 10, f represents a spatial frequency, and a, b, c, and d represent constants, respectively. Mannos and Sakrison proposed a modulation transfer function model as shown in Equation 11 through experiments.

[Equation 11]

The modulation transfer function has bandpass characteristics in the frequency domain. As a result of common research, human vision shows the maximum response sensitivity at 4.7 to 8 (cycle per degree), and at 0.1 and 30 to 40 (cycle per degree), it shows only about 3% of the maximum response. It is also known that the reactivity decreases exponentially in the high frequency region.

In the present invention, Equation 11 is used as the modulation transfer function. In the modulation transfer function of Equation 11, the maximum responsiveness is about 8 (cycle per degree). The modulation transfer function expressed on the plane of the spatial frequency and the visual responsiveness needs to be transformed into a normalized spatial frequency domain. For this conversion, the spatial frequency of the modulation transfer function can be expressed by the following equation (12).

[Equation 12]

here, Means the normalized spatial frequency, Represents the number of pixels included in one degree of time. In the present invention 64 This is equivalent to viewing an image at a distance of about four times the screen size.

A transformation of the modulation transfer function expressed in spatial frequency into a normalized frequency domain is shown in FIG. 5. Normalized Coefficients are obtained by applying the wavelet basis function to the modulation transfer function. Wavelet function Space to subspace It can be expressed as Equation 13 below.

[Equation 13]

In Equation 13, when the wavelet basis function Φ is an orthogonal function, all subspaces are also orthogonal to each other, and each space corresponds to the rental width at the j th stage of the iterative filter bank.

The signal obtained by applying the wavelet basis function Φ to the discrete signal is limited to the rental width of (π / 2,) in the entire interval (0, π) of (). The application of the continuous application function in the iterative filter bank is the process of dividing the rental width in half and the rental width obtained by applying the j th filter. Is represented by Equations 14 and 15 below.

[Equation 14]

Vetterli's 22 tap filter is an orthogonal filter that satisfies perfect reconstruction conditions. Wavelet segmentation is the application of a wavelet function to a video as a filter. The FIR filter may be implemented as a moving average filter, and the transfer function of the filter in the Z-conversion region is expressed by Equation 16 below.

[Equation 16]

The transfer function of the j th filter in the iterative filter bank is expressed as in Equation 17 below.

[Equation 17]

The band pass characteristics of the wavelet filter and the normalized modulation transfer function are used to determine the quantization interval of the quantizer, which is obtained from the integral ratio of the filter pass band and the modulation transfer function (MTF). The stepsize of the quantizer is obtained as the inverse of the integral ratio. Equation 18 below shows the integral ratio ( ) And Equation 19 is the step size of the quantizer ( ).

Equation 18

[Equation 19]

6A to 6D show frequency response curves of the modulation transfer function (MTF) and the wavelet basis function in the frequency domain in order to calculate the integral ratio and the step size. FIG. 6A shows the frequency response of the first step filter, FIG. 6B shows the frequency response of the third step filter, FIG. 6C shows the frequency response of the fifth step filter, and FIG. 6D shows the frequency response of the sixth step filter. Indicates.

As described above, the present invention can be used in a general still image and video signal compression system such as a digital VCR, a high definition television (HDTV), a multimedia, a filmless camera, and the like.

As described above, according to the present invention, the wavelet transformed signal is quantized using the quantization coefficient obtained by the integral ratio of the frequency response of the modulation transfer function of the human visual system, and the result is variable-length encoded, thereby increasing the compression efficiency. Have

Claims (1)

A method of compressing an image signal by dividing an input image signal using a wavelet transform and quantizing the divided data and variably encoding the divided data, the method comprising: a space between a band pass frequency characteristic of the wavelet transform and a characteristic of a human visual system Obtaining an integral ratio with a modulation transfer function (MTF) using frequency characteristics; And determining a quantization step size based on an inverse value of the integral ratio.