KR100740501B1 - Method for coding ultrasound images using different threshold values - Google Patents

Abstract

A method of encoding a multilevel wavelet image represented by wavelet coefficients according to a plurality of wavelet transforms is provided. Each level includes a low band subband (LL subband) and a border region subband (HL subband, LH subband and HH subband). First, a first threshold and a second threshold smaller than the first threshold are selected, and an important coefficient is selected. If the wavelet coefficient of the low band subband is larger than the first threshold, it is selected as an important coefficient. The wavelet coefficient of the subband is selected as an important coefficient when it is larger than the second threshold. The critical coefficient is selected by searching wavelet coefficients from high level to low level. At the same level, significant coefficients are selected by searching wavelet coefficients in the order of LL subband, LH subband, HL subband, and HH subband.

Ultrasound Image, Wavelet Transform, Wavelet Image, Significance Coefficient, Coding

Description

Wavelet image coding method using differential threshold {METHOD FOR CODING ULTRASOUND IMAGES USING DIFFERENT THRESHOLD VALUES}

1 is a schematic view showing a conventional wavelet decomposition and restoration process.

Figure 2 is a schematic diagram showing the configuration of the subband generated through the conventional one-level wavelet transform.

3A is a schematic diagram showing a hierarchy of subbands generated through conventional three-level wavelet transform;

3B is a schematic diagram showing the hierarchical correlation of pixel values according to a conventional wavelet transform.

4A is a graph showing wavelet coefficient size distributions of LL and HL subbands for a conventional natural image.

Figure 4b is a graph showing the wavelet coefficient size distribution of the LL and HL subbands for the conventional ultrasound image.

5 is a schematic diagram showing a coding sequence for each band of the conventional EZW algorithm.

6 is a schematic diagram illustrating a process of encoding an ultrasound image using EZW according to an embodiment of the present invention.

7 is a schematic diagram showing a tree structure of coefficients in a wavelet transform image according to an embodiment of the present invention.

8 is a flowchart illustrating a critical coefficient determination process in a main encoding process according to an embodiment of the present invention.

9 is a chart showing energy distribution for each subband of a natural image and an ultrasound image;

10 is a schematic diagram showing a coding sequence for each band according to the present invention.

The present invention relates to a wavelet image encoding method, and more particularly, to a wavelet image encoding method using differential thresholds.

An efficient compression method is needed to transmit and display a large amount of images in a short time. MPEG (Motion Pictures Experts Group), H.264, etc., which are typical image compression techniques, perform coding using a discrete cosine transform (DCT) to reduce redundancy in two-dimensional space. However, DCT is a block-by-block transformation, which results in a blocking effect in which the boundary between blocks is clearly seen at a low bit rate, and the image quality deteriorates when the compression ratio is increased.

In order to overcome the shortcomings of DCT and compress a large amount of images, wavelet transformation (hereinafter, referred to as WT) is used. When using WT, spatial and frequency resolutions can be displayed variably, so that the area where pixel value does not change significantly and the sharp change of pixel value such as contours can be processed effectively at the same time. The block boundary does not appear even at low bit rate compression. Image compression using WT having such an advantage is widely used in various fields such as denoising, computer vision, and feature extraction.

Wavelet transform and reconstruction is accomplished by passing a filter bank having a two channel consisting of a low pass filter and a high pass filter as shown in FIG. That is, the analysis filter banks G0 and H0 are applied to an image and decomposed into images of a high band and a low band. Synthetic filter banks G1 and H1 reconstruct an image signal with the signals resolved by analysis filter banks G0 and H0.

On the other hand, if the two-channel analysis filter banks G0 and H0 are applied to the input video signal, the lengths of the two band-divided signals are equal to the length of the first input signal, respectively, and the total length is doubled. To avoid the increase, apply analysis bank filters (G0, H0), then apply down sampling (↓ 2) to remove half of the input signal, and insert a '0' between the signals to reconstruct the down sampled signal again. Upsampling (↑ 2) doubles the frequency spacing of a given signal.

Coding is performed by applying analysis filter banks G0, H0 and downsampling ↓ 2. Wavelet transform coding is an advantage of band division coding in which image band division and image compression are operated independently. There is this.

FIG. 2 shows that LL1, LH1, HL1, and HH1 are generated by applying the analysis filter banks FB1 once in a horizontal direction and a vertical direction with respect to a unit image (band) which is two-dimensional data. LL1 is the subband which passed the low pass filter in the horizontal and vertical direction, LH1 is the subband which passed the high pass filter in the horizontal direction and vertical direction, and HL1 is the subband which passed the low pass filter in the horizontal direction and vertical direction. , HH1 refers to the subband passing through the high pass filter in the horizontal and vertical directions. The number 1 represents the number of wavelet transform repetitions.

Each subband LL1, LH1, HL1, HH1 is divided into directional components of the image. That is, the low-low subband LL ₁ represents the average brightness of the image, and the boundary bands of the LL1, that is, LH1, HL1, and HH1, represent high frequency components of the image corresponding to the horizontal, vertical, and diagonal components, respectively. .

As such, the wavelet transform is a subband subdivision structure that separates a low frequency region containing relatively more meaningful information that has a sensitive effect on human visual perception and a high frequency region containing relatively less meaningful information of an image signal. do.

Wavelet transform coding separates a video signal into different frequency domains through band filter banks, which are separated into frequency domains in two-dimensional space, and the importance of reproduction is reduced as the frequency band increases.

A subband obtained by performing n wavelet transforms is called an n-level subband. FIG. 3A is a schematic diagram illustrating a hierarchy of subbands according to multi-resolution image decomposition through 3-level wavelet transform, and FIG. 3B is a hierarchical structure of pixel values (hereinafter, referred to as wavelet transform coefficients) of the wavelet transform image according to wavelet transform. Schematic diagram showing correlation.

On the other hand, EZW (Embedded Zero-tree Wavelet) coding is used as a compression coding method using the feature of wavelet transform.

EZW encoding is an encoding technique using a zero-tree data structure and a bit plane. The zerotree data structure is set to zero at the root of the tree, i. In other words, the root of the zero tree is set to an end of block (EOB). The bit plane is obtained by decomposing each pixel value constituting an image bit by bit.

EZW encoding is to collect and encode wavelet coefficients related to the same position of the original image. In the case of the wavelet transform using the 2-channel filter bank, since the tree structure of the wavelet transform coefficients increases in proportion to the power of 2 as shown in FIG. 3B, from the upper layer to the lower layer, a large amount of information can be encoded at a time. .

The zero-tree coding method applies a set partitioning sorting algorithm to align bit planes according to coefficients, transmits aligned bit planes, and calculates important coefficients using similarity between different subbands in a wavelet transformed image. Conduct predictive importance checks. In the EZW, the first important information is sent by taking advantage of the fact that the largest coefficient has the largest amount of information. In addition, it transmits from the bit plane corresponding to the most significant bit (MSB). To this end, arbitrary wavelet transform coefficients larger than a threshold value are selected as important coefficients, and the selected significant coefficients are encoded. As such, EZW, which selects and encodes an important coefficient, has very high performance in terms of compression rate and image quality, and has an advantage of sequential image reconstruction.

On the other hand, in the EZW encoding, when the desired bit rate is reached during encoding, the encoding ends regardless of which subband the encoding has progressed to. Therefore, different encoding results can be obtained at the same bit rate depending on what information is sent first. That is, as the first encoding of information for minimizing distortion, image distortion may be reduced.

Meanwhile, the natural image and the ultrasound image have different distributions of wavelet transform coefficients in subbands.

4A and 4B are graphs comparing the size distribution of wavelet transform coefficients in the low-band subband LL and the vertical subband HL of the natural image and the ultrasound image, respectively.

In the case of a natural image, the wavelet transform coefficient distribution in the subbands LL and HL is relatively similar, as shown in FIG. 4A. However, in the ultrasound image, the distribution of wavelet transform coefficients in the subbands LL and HL is different as shown in FIG. 4B.

Unlike the natural image, the ultrasound image has less information on the boundary region subbands (HL, LH, HH) than the low band (LL), causing severe image distortion at a specific bit rate. In particular, as shown in FIG. 4B, the subband HL has a relatively large number of small wavelet transform coefficients in the ultrasound image. Therefore, when the algorithm of the conventional EZW encoding process that selects the critical coefficient by applying the same threshold value to all subbands is applied as it is, when the target bit is reached during encoding, the wavelet transform coefficients in the HL subbands cannot be encoded. There may be a problem.

Meanwhile, in the ultrasound image, speckle information, which is used as important information for diagnosis in the image acquisition step, tends to be concentrated in the LH subband due to mechanical characteristics. The conventional EZW searches sequentially from the high level to the low level, as shown in FIG. 5, from left to right within the same level. Accordingly, since LL subbands, HL subbands, LH subbands, and HH subbands are searched and encoded at the same level, there is a problem in that LH information cannot be encoded when encoding is terminated during HL encoding. .

SUMMARY OF THE INVENTION The present invention solves the above-described problems, and applies different thresholds to the low band subband LL and the boundary subband HL, LH and HH to check the importance of the wavelet transform coefficients at the same bit rate. An object of the present invention is to provide an ultrasound image encoding method using EZW capable of encoding more information.

Another object of the present invention is to provide an ultrasound image encoding method using EZW capable of preserving speckle information, which is a high frequency component used as important information for diagnosing an ultrasound image by changing the encoding order of subbands.

A method of encoding a wavelet image according to the present invention is a method of encoding a multilevel wavelet image represented by wavelet coefficients according to a plurality of wavelet transforms, wherein each level includes a lowband subband and a boundary region subband. Selecting a first threshold and a second threshold less than the first threshold; And selecting an important coefficient, wherein the wavelet coefficient of the low band subband is selected as an important coefficient when the wavelet coefficient is larger than a first threshold value, and the wavelet coefficient of the boundary region subband is selected as an important coefficient when the wavelet coefficient is larger than a second threshold value. Steps.

In the ultrasound image encoding method according to an embodiment of the present invention, as shown in FIG. 6, the ultrasound image is formed by wavelet transforming the ultrasound image based on the ultrasound echo signal, and the dominant pass, the subordinate pass, and the adaptive arithmetic coding are performed. (adaptive arithmetic coding).

According to the wavelet transform, the parent coefficient and the descendant coefficient have a self-similarity. As shown in FIG. 7, four pieces of image information having a coefficient in the HH3 band exist in the HH2 band at the same position. Similarly, in the HH2 band, four pieces of information of an image having one coefficient exist in the HH1 band at the same position. Therefore, 16 pieces of information in the same position as one coefficient in the HH3 band exist in the HH1 band. At this time, the coefficients in the HH3 band are called the parent coefficients, and the four coefficients in the HH2 band and the 16 coefficients in the HH1 band are called descendant coefficients. Similarly, if coefficients in the HH2 band are called parent coefficients, four coefficients in the HH1 band are called descendant coefficients.

In general, the magnitude of the coefficient decreases from the parent coefficient to the descendant coefficient. In addition, when the value of the parent coefficient is large at the same level, the probabilities of the descendant coefficients are also high, and when the value of the parent coefficient is small, the probabilities of the descendant coefficients are also high. This is the autocorrelation of the wavelet resolved image. According to this property, the wavelet coefficient may be expressed as a zerotree.

After wavelet transforming the video signal to a predetermined level, main encoding is performed on the wavelet transformed video signal. The main encoding is a process of determining a significant coefficient based on a threshold value, and includes position information of each coefficient in the main encoding process.

Hereinafter, a main encoding process according to an embodiment of the present invention will be described in detail.

Unlike the natural image, in the present invention, the wavelet coefficient of the boundary region subbands HL, LH, and HH is smaller than that of the lowband subband LL. The critical factor of is selected using different thresholds. That is, a first threshold T1 for selecting a critical coefficient in the low band subband and a second threshold T2 for selecting a critical coefficient in the boundary region subband are set. Preferably, the first threshold value T1 is proportional to 2 ⁿ (n is an integer), and the second threshold value T2 may be set to 2 ⁿ −α smaller than the first threshold value T1 (α> 0).

Then, a zero tree map is formed based on the first threshold value T1 and the second threshold value T2 which are selected to quickly determine the importance of descendants of the boundary band coefficients. The zero tree map is formed by scaling all descendants of the coefficient. T1 / T2 values can be used for scaling. It is preferable to scale the distribution of the descendant coefficients in the zero tree map to be distributed within -T1 to T1.

An important coefficient selection method will be described with reference to FIG. 8.

First, the subband to which the input coefficient belongs is identified. Set the reference value to the first threshold value T1 if the input coefficient belongs to the low-band subband LL, and set the reference value to the second threshold value T2 if the input coefficient belongs to the boundary region subband ((LH, HL, HH)). do.

Next, it is determined whether the absolute value of the input coefficient is larger than the reference value. If the absolute value of the input coefficient is larger than the reference value, the sign of the input coefficient is determined. If the sign is positive, it is classified as POS (Positive Significant). If the sign is negative, it is classified as NEG (Negative Significant).

If the input coefficient is smaller than the reference value, it is determined whether there is a significant descendant coefficient. At this time, the determination of the descendant critical coefficient determines whether the descendant coefficient of the zero tree map formed according to the above-described process is larger than the first threshold value. As described above, when the absolute value of the input coefficient is smaller than the reference value, but there is a value larger than the first threshold value among the descendant coefficients on the zero tree map, it is classified as IZ (Isolated Zero).

When the input coefficient is smaller than the reference value and the descendant coefficient on the zero tree map is also smaller than the first threshold value, the input coefficient is classified as a zero tree root (ZTR).

When input coefficients are classified into POS, NEG, and IZ, each descendant coefficient is determined and classified into POS, NEG, IZ, and ZTR.

If it is determined as ZTR, it no longer encodes descendant coefficients. In other words, when the parent and descendant coefficients related to restoring the same position of the image are considered as one block, the ZTR symbol is a symbol corresponding to the end of block (EOB), in which the descendant coefficients of the block including the current encoding coefficient are all zero. .

Meanwhile, in the encoding process described above, encoding is performed in the order of the high level to the low level, and in the same level, the encoding is performed in the order of the high energy to the low subband. For example, in the case of an ultrasound image having an energy distribution for each subband different from the natural image as shown in FIG. 9, the energy is encoded in the order of high energy, that is, LL subband, LH subband, HL subband, and HH subband. . Accordingly, band-by-band encoding can be performed in the order shown in FIG. 10.

According to the above-described process, the critical coefficients are determined in the main encoding process, and then the dependent encoding is performed. Dependent encoding may be performed according to a method commonly used in the art to which the present invention pertains, and thus a detailed description thereof will be omitted.

Information coming out from the main encoding or the dependent encoding is encoded through an arithmetic encoding process. Arithmetic encoding may also be performed according to a method commonly used in the art to which the present invention pertains, and thus a detailed description thereof will be omitted.

If it is determined that the target bit rate is reached in the arithmetic encoding process, the encoding is terminated by outputting the bit stream. When the target bit rate is not reached, the main encoding process and the dependent encoding process are repeatedly executed while decreasing the first threshold value and the second threshold value by 1/2, respectively.

While the invention has been described and illustrated by way of preferred embodiments, those skilled in the art will recognize that various modifications and changes can be made without departing from the spirit and scope of the appended claims.

As described above, according to the present invention, not only a coding effect higher than that of the conventional EZW can be obtained for an ultrasound image having an energy distribution different from that of the natural image, but also other images having an energy distribution similar to that of the ultrasound image can be obtained. High coding effect can be obtained.

In addition, according to the present invention, the horizontal boundary band LH is preferentially encoded from the vertical boundary band HL among the boundary subbands, thereby preserving speckle information, which is important information for medical diagnosis using an ultrasound image.

Claims (8)

A method of encoding a wavelet image of each level represented by wavelet coefficients according to a plurality of wavelet transforms and including a low band subband and a boundary region subband, Selecting a first threshold value and a second threshold value less than the first threshold value; And Selecting an important coefficient, wherein the wavelet coefficient of the low band subband is selected as an important coefficient when the wavelet coefficient is larger than a first threshold value, and selecting the important coefficient as a significant coefficient when the wavelet coefficient of the boundary region subband is larger than a second threshold value Including, wavelet image encoding method. The method of claim 1, And the significant coefficients are selected by searching for wavelet coefficients from high level to low level. The method of claim 2, The boundary region subband includes an HL subband, an LH subband, and an HH subband. And the significant coefficients are searched and selected in order of energy of the boundary region subbands at the same level. The method of claim 2, The boundary region subband includes an HL subband, an LH subband, and an HH subband. And the significant coefficients are searched and selected in the order of the LL subband, LH subband, HL subband, and HH subband at the same level. The method of claim 4, wherein The step of selecting the critical coefficient, Selecting the first threshold as a reference value when the target wavelet coefficient is a low band subband, and selecting the second threshold as a reference value when the target wavelet coefficient is a boundary subband; Determining whether an absolute value of the target wavelet coefficient is greater than a reference value; If the absolute value of the target wavelet coefficient is larger than a reference value, determining the sign of the input coefficient, classifying it as positive significant (POS) when the sign is positive, and classifying it as negative significant (NEG) when the sign is negative; Determining whether a descendant significant coefficient exists when the target wavelet coefficient is smaller than a reference value; Classifying as an IZ (Isolated Zero) when the absolute value of the target wavelet coefficient is smaller than a reference value but a significant descendant coefficient exists; And And classifying the target wavelet coefficient into a zero tree root (ZTR) when the target wavelet coefficient is smaller than a reference value and a descendant significant coefficient does not exist. The method of claim 5, Forming a zerotree map of the wavelet coefficients based on the first threshold value and the second threshold value, And the descendant significant coefficient is selected by determining whether the descendant coefficient of the zero tree map is greater than a first threshold value. The method of claim 5, And the zero tree map is formed by scaling the wavelet coefficient in a range of a first threshold value to a second threshold value. The method according to any one of claims 1 to 7, The wavelet image is a wavelet image encoding method obtained by wavelet transform the ultrasound image.

Images