JPH07203450A - Image data compression method - Google Patents

Abstract

PURPOSE:To transmit high frequency band information reproduced accurately by a expansion reproduction side at a high compression efficiency in the image data compression method in which a picture signal is divided into sub ands and high frequency band information compressed by using inter-sub-band prediction is coded and sent together with low frequency band information. CONSTITUTION:An image signal is divided into sub bands by an LPF 1 and an HPF 2 to obtain a predict signal He(j) for a high frequency band signal H(j) obtained by applying HPF 5 to a low frequency band signal L(i) and its inverted signal -He(j). Difference arithmetic operation sections 7, 9 obtain a difference signal H(j)+ or -He(j) between the signal and the high frequency band signal H(j), an absolute value comparator section 10 compares the absolute value of the difference signal H(j)+ or -He(j) and that of the high frequency band signal H(j) and a signal selection section 11 selects the minimum absolute value signal as a transmission signal. Furthermore, an identification IB representing the selected type is added to the selected transmission signal.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image data compression method, which is applied to a digital video tape recorder (D-VTR), an image data filing system, etc., and uses a correlation between subbands to generate a high band signal. The present invention relates to a method of highly efficiently compressing and transmitting such information.

[0002]

2. Description of the Related Art In a data transmission system such as a D-VTR, image data is compressed / decompressed according to the JPEG (Joint Photographic Experts Group) system, but it is increasingly required to increase data processing and transmission speed. Becomes,
In the encoding of image data, it is an important issue to improve the compression efficiency so as to reduce the code amount after data compression as much as possible.

With respect to the problem, conventionally, when a method of dividing an image signal into sub-bands and encoding and transmitting is adopted, each sub-band is independently coded under the assumption that there is no correlation between the sub-bands. However, on the contrary, while verifying that there is actually a correlation between subbands,
Various proposals have been made on the possibility of realizing efficient data compression by utilizing the correlation. For example, "Subband image coding method based on local structure of image" (IEICE Technical Report IE92-73, 1992-11, Sawazawa / Hamada / Matsumoto; hereinafter referred to as "Reference 1") and "Wavelet transform" Of Bandwidth Prediction in 1992 "(IEICE Spring Conference D-33
0, pp. 7-72, Takeda and Otake; hereinafter referred to as "Reference 2"), a study on signal prediction based on correlation between subbands is made. Note that Document 2 is considered to be a method of subband coding that imposes a predetermined condition on the division / synthesis filter, and thus can be included in the subband coding.

Then, in Reference 1, the fact that the factors that create the cross-correlation between subbands is that the local structure such as edges and lines of an image and that the filter does not have an ideal band limiting characteristic is verified. Then, the digital image signal is sub-band divided into a low band signal and a high band signal using a low pass filter (hereinafter referred to as "LPF") and a high pass filter (hereinafter referred to as "HPF"), and A method is given in which the value obtained by multiplying the adjacent value difference of the low band signal by a constant (cross-correlation coefficient) is used as the predicted signal level value of the high band signal.
Therefore, if the high band signal is coded and transmitted together with the low band signal as the information expressed using the cross-correlation coefficient, the data compression efficiency is higher than that in the case of independently coding the high band signal. Transmission becomes possible.

[0005]

By the way, the following problems are pointed out in the data compression method according to Document 1. Prior to pointing out the problem, a signal prediction model using the correlation between subbands will be described with reference to FIGS. 10 to 13. First, FIG. 10 is a circuit diagram when a digital image signal is divided into subbands. The input signal I (n) is divided into subbands by LPF1 and HPF2.
The outputs are thinned out by 2: 1 by decimators 3 and 4, respectively, and a low band signal L (i) and a high band signal H (j) are obtained. Also, LPF1 and HPF
2 is the SSKF (Sub-
band Coding of Digital Images Using Symmetric Shor
t Kernel Filters) and their transfer functions are shown in Equation 1 below.

[0006]

[Equation 1]

On the other hand, FIG. 11 shows the changing state of the one-dimensional input signal I (n) at the sampling points S1 to S9.
Shows the changing states of the low band signal L (i) and the high band signal H (j) corresponding to the input signal I (n) of FIG. 11, respectively.
The signal passed through each filter 1 and 2 is 2: 1 in decimator 3 and 4.
Since the thinning is performed, the low band signal L (i) is S2,
Appearing at S4, S6, S8, the high band signal H (j) is S1, S3, S5, S
Appears in 7, S9. If the image signal is a luminance signal, 0 to 25 bits represented by 8 bits are used as the input signal.
It may be set to 6 quantization levels (positive value), but in such a case, the central value of 128 quantization levels is subtracted.

The inter-subband prediction is performed by predicting the high band signal H (j) of FIG. 13 from the low band signal L (i) of FIG. Here, from FIG. 11, the input signal I (n) does not change between I (1) to I (3) and I (8), I (9), and the original signal of the image is I (4) and I (4). Between (5) and I (7) and I (8)
It is easy to see that there is an edge between L (4) and L (6) and the original signal of the low band signal L (i) is between L (4) and L (6). It is understood to have an edge between L (6) and L (8). On the other hand, referring to FIG. 13, the high band signal H (j) has a relatively large level value at H (5) and H (7). Therefore, it is presumed that a strong correlation exists between the low band signal L (i) and the high band signal H (j) at the edge portion of the original signal.

Next, for inter-subband prediction, first, the original signal decimated from the low band signal L (i) in FIG. 12 is estimated. In that case, an estimated original signal is created by doubling the sampling frequency by zero-order interpolation, but in that case as well, the following two types of zero-order prediction methods [A],
There is [B].

In the 0th-order prediction method [A], as shown in FIG. 14, as the estimated original signal La (5) at the sampling point S5 of the low band signal L (i), L at the sampling point S4 immediately before it is used. (Four)
Is used as the estimated original signal La (7) at the sampling point S7, and L (6) at the immediately preceding sampling point S6 is used. Then, the predicted high band signals He (5) and He (7) at the sampling points S5 and S7 are obtained from the above-mentioned estimated original signals La (5) and La (7). As shown by the chain double-dashed line in FIG. 10, the 0th-order predictor 20 and HPF5 are provided on the output side of the low band signal L (i).
And it is enough to install the decimator 6. Here, HPF5
Has the same transfer function as HPF2. In this case, the predicted values He (5), He (7) of the high band signal are La (5) = L (4), La
Since (7) = L (6), they are given by the following equations. He (5) = [-L (4) + 2.La (5) -L (6)] / 4 = [L (4) -L (6)] / 4 (b) He (7) = [- L (6) + 2.La (7) -L (8)] / 4 = [L (6) -L (8)] / 4 (B)

Therefore, the same result can be obtained even if the transfer function FH2 (Z) of HPF5 is given by the following formula 2 except for the 0th-order predictor 20 and the decimator 6.

[0012]

[Equation 2]

On the other hand, in the 0th-order prediction method [B], as shown in FIG. 14, as the estimated original signal Lb (5) at the sampling point S5 of the low band signal L (i), the sampling point S6 immediately after it is used. L (6) at the sampling point S7 is used as the estimated original signal Lb (7) at the sampling point S7. Therefore, in this prediction method [B], the predicted values He (5), He (7) of the high band signal obtained as a signal that has passed through the HPF 5 and the decimator 6 are Lb (5) = L (6) , Lb (7) = L (8), the following equation is obtained. He (5) = [-L (4) + 2.La (5) -L (6)] / 4 =-[L (4) -L (6)] / 4 (C) He (7) = [ -L (6) + 2.La (7) -L (8)] / 4 =-[L (6) -L (8)] / 4 (D)

As described above, the predicted high band signal He
(5), He (7) can be obtained by the 0th-order prediction methods [A] and [B], as seen in the results of (a) and (c) and the results of (b) and (d). Although positive and negative will be obtained as opposite predictive values, they are likely to be equally reliable. However, looking at the change in the high band signal H (j) in FIG.
For (5), the results obtained in (c) are valid, and conversely for He (7), the results obtained in (b) are assumed to be valid. 11 input signal I (n)
Is also supported by the changes in. In other words, He (5) does not succeed in prediction according to the 0th-order prediction method [A], and He (7)
For 0, the prediction will not succeed according to the 0th-order prediction method [B].

Then, the data compression method according to Document 1 is
As described above, it is not taken into consideration that the predicted high band signals He (5) and He (7) are inverted between positive and negative depending on how to select the 0th-order prediction method. The value obtained by multiplying the adjacent value difference of the low band signal explained in Literature 1 by a constant (cross-correlation coefficient) is H
Although e (5) and He (7) will be given as approximations to either positive or negative values shown above, the 0th-order prediction method is not switched step by step between A and B. However, the prediction of the high band signal may not be successful. Note that, in Document 2 as well, the predicted value of the high band signal is created using the output obtained by applying the low band signal to the HPF, but the same can be said. Therefore, although the data compression based on Document 1 can achieve higher data compression than the case where the high band signal is independently encoded, the data is expanded and reproduced by using the cross-correlation coefficient on the receiving side. Sometimes it makes an error.

Therefore, an object of the present invention is to provide an image data compression method which makes it possible to transmit data with high compression efficiency without causing an error in prediction while using the inter-subband prediction method. Was created as.

[0017]

According to the present invention, an image signal is sub-band divided into a low band and a high band, and the high band signal is predicted from the low band signal obtained by the sub band division. It is applied when an image data compression method is adopted in which high band information generated by using the predicted signal is encoded and transmitted together with low band information.

In a first aspect of the present invention, when the image data compression method is adopted, the low band signal is subjected to high pass processing by HPF to obtain a prediction signal related to the high band signal. Creating the positive / negative inversion signal, the level value of the prediction signal and its positive / negative inversion signal, or a value obtained by multiplying the level value by a constant coefficient and the high band signal at the sampling point corresponding to the prediction signal A difference between the level value is obtained, a signal having the smallest absolute value is selected as a transmission signal from the difference signals and the high band signal, and identification information indicating the selection type is given to the selected transmission signal. The present invention relates to an image data compression method characterized in that

In a second aspect of the present invention, when the image data compression method is adopted, the low band signal is subjected to high pass processing by HPF to obtain a prediction signal related to the high band signal. Creating the positive / negative inversion signal, the level value of the prediction signal and its positive / negative inversion signal, or a value obtained by multiplying the level value by a constant coefficient and the high band signal at the sampling point corresponding to the prediction signal The difference between the level value is obtained, and the encoded data amount in a state where the identification information indicating the selection type when each signal is selected for the respective difference signal and the high band signal is added correspondingly, The image data compression method is characterized in that a signal having the smallest data amount is selected as a transmission signal.

A third aspect of the present invention is the image data compression method according to the first or second aspect of the present invention, wherein a low-pass band signal of (m × n) pixel block units is subjected to high-pass processing by HPF, and Each of the difference signals is created by using the result of sequentially selectively multiplying the level value of the positive / negative inversion signal by a plurality of different coefficients as a prediction signal, and in the image data compression method of the first invention, in the pixel block unit, A signal having a minimum absolute value among the difference signal and the high band signal is calculated for each coefficient, and a signal having a minimum absolute value is selected as a transmission signal from among the calculated signals. As for the image data compression method of (1), the signal having the minimum encoded data amount is obtained for each coefficient in the pixel block unit, and the signal having the minimum encoded data amount is transmitted among the obtained encoded data amounts. When a differential signal is selected for the entire pixel block as a signal, an image in which identification information, pixel block information, and coefficient information corresponding to the selected signal are added to the transmission signal in a block unit This relates to a data compression method.

[0021]

[Action]

Regarding the first invention; a signal obtained by subjecting a low-frequency band signal to high-pass processing by an HPF gives a prediction signal related to the high-frequency band signal, but as described with reference to FIG. There are two methods to obtain the estimated original signal of the low band signal by the next prediction, and the estimated signal based on the estimated original signal obtained by each method is
The signal becomes a signal with positive and negative inversion as in the relationship of (c) and the relationship of equations (b) and (d). Therefore, since one of the estimated original signals by each method has an accurate level value as a prediction signal, the level value of the prediction signal and its positive / negative inversion signal and the sampling point corresponding to the prediction signal When the difference with the level value of the high band signal is obtained, any difference should be the difference between the accurate prediction signal and the high band signal, and its absolute value should be 0 or a minute value.

When transmitting a signal, generally, the smaller the absolute value of the signal level, the smaller the amount of data after encoding. Therefore, if the signal having the smallest absolute value is selected as the transmission signal from each of the differential signals and the high band signal, the information related to the high band signal can be transmitted with the minimum data amount. However, identification information indicating the selection type is added to the transmission signal. It should be noted that the case where the absolute value of the original high band signal is also selected as a comparison target is included in that the prediction signal is only related to prediction, and it is more important than the case of transmitting the difference signal. This is because the amount of data may decrease.

On the other hand, on the decompression playback side, when the transmission signal is received, the identification information is first discriminated to confirm what signal is selected, and the low band signal transmitted at the same time is used to detect the signal. It is possible to reproduce the high band signal by executing an algorithm reverse to the procedure on the data compression side of.

Regarding the second invention; When transmitting encoded data, the transmission efficiency is naturally proportional to the amount of data after encoding. Therefore, originally, it is ideal to minimize the amount of data after encoding.

By the way, in the above-mentioned first invention,
The signal having the smallest absolute value among the differential signals and the high band signal is selected as the transmission signal, but if the selected transmission signal is encoded by adding identification information indicating the selection type, the code is In terms of the amount of data to be converted, the amount of data relating to the transmission signal is not always the minimum.
That is, the identification information has a minimum of 2 to indicate three kinds of selection conditions.
Although a bit amount is required, in view of the transmission data amount, the data amount of another signal may be smaller than that of the signal selected under the condition that the absolute value becomes the minimum, depending on how to allocate the bit.

Therefore, the present invention is similar to the first invention until each differential signal is obtained, but at the final stage of selecting a transmission signal, identification information is provided to each differential signal and high band signal. The encoded data amount in the state of corresponding addition is obtained, and the signal having the minimum data amount is selected as the transmission signal.

Regarding the third invention; In the first and second inventions, the difference between the level value of the predicted signal and its positive / negative inverted signal and the high band signal is obtained, and each difference signal is used for the high band information. It is used as a candidate signal. However, in a local structure such as a sharp edge on the image, the high-frequency band signal may have a large amplitude with respect to the prediction signal and take a value far apart, in which case no matter which signal is selected, Since the absolute value of the signal level and the coding amount have relatively large values, the effect of data compression decreases.

Therefore, in the present invention, the image data is (m
Xn) divided into pixel blocks, and the results obtained by sequentially multiplying each of the signals corresponding to the prediction signal and the positive / negative inversion signal thereof in the pixel block unit by a plurality of different coefficients are shown. Used as a prediction signal. Therefore, as described above, the level of the differential signal created from the prediction signal multiplied by each coefficient and the high band signal has an absolute value or coded data amount that differs depending on the coefficient value.

According to the present invention, since the differential signal having the smallest absolute value or coded data amount is the selection candidate in the pixel block unit, the differential signal selected as the transmission signal depending on the setting method of the coefficient value. Can be suppressed from increasing due to the local structure on the image, and the amount of transmission data can be reduced. However, in the present invention, pixel block information and coefficient information related to the selected signal must be added to the transmission signal in addition to the identification information indicating the signal selection type. However, it suffices to add corresponding information to each pixel block, and the addition of information does not become a factor for increasing the amount of transmission data.

[0030]

Embodiments of the image data compression method of the present invention will be described below in detail with reference to FIGS. Embodiment 1; This embodiment corresponds to the first invention described above, and FIG. 1 shows a system circuit diagram of an apparatus for carrying out the method. In the figure, the input signal I (n) is set to LP
F1 and HPF2 are divided into sub-bands, and their outputs are thinned out by 2: 1 data by decimators 3 and 4, respectively, to generate a low band signal L (i) and a high band signal H (j), Further, the prediction signal He (j) relating to the high band signal H (j) is obtained by subjecting the low band signal L (i) to high pass processing with HPF5 as in the case of FIG. It is the same. The transfer function of HPF5: FH2 (Z) is given by Equation 2, and the zero-order predictor 20 and the decimator 6 are unnecessary.

In the present embodiment, the difference calculator 7 obtains the difference between the high band signal H (j) and the prediction signal He (j): H (j) -He (j), and also the prediction signal He (j). Is inverted by the inversion circuit 8 and the difference calculation unit 9 obtains the difference: H (j) + He (j) between the high band signal H (j) and the inverted signal: -He (j), and the absolute value comparison is performed. In the section 10, the absolute value of each differential signal: H (j) ± He (j) and the high band signal H (j) are compared.

Then, the absolute value comparison unit 10 determines that the signal having the minimum absolute value is the transmission signal, and sends the selection signal SS based on the determination to the transmission selection unit 11, which then selects the selection signal SS. The signal X (j) [differential signal: H (j) −He (j) or differential signal: H (j) + He (j) or high band signal H (j)] corresponding to the above is transferred to the encoding unit 12. To do. Further, the transmission selection unit 11 transfers the selection signal SS to the identification bit generation unit 13, and the identification bit generation unit 13 generates the identification bit IB related to the selection type indicated by the selection signal SS and sends it to the encoding unit 12. Forward. In this embodiment,
The case of transmitting any of the differential signals is represented by 2 bits [“11” and “10”], and the case of transmitting the high band signal H (j) is represented by 1 bit [“0”]. There is.

On the other hand, the encoding unit 12 receives the low band signal L (i) from the decimator 3 and the signal X (j) selected above and the identification bit IB, encodes them and transmits them to the transmission line. Output to. As a result, the absolute value of the high band signal H (j) together with the low band signal L (i), the differential signal: H (j) -He (j), and the differential signal: H (j) + He (j) The signal with the minimum value will be selectively transmitted through the transmission path.
(1) Each differential signal corresponds to the level difference between the original high band signal and the predicted signal, and one of the values has an extremely small absolute value, and (2) the absolute value. Since the signal with the minimum condition corresponds to the signal with the smallest data amount when variable length coding is performed, the high band signal H
It becomes possible to highly efficiently compress / encode the information related to (j) and transmit it.

Although variable length coding is assumed in (2) above, requantization of the signal level is usually performed in order to keep the coded data amount below a certain amount. Since a signal with a large absolute value is quantized in a large quantization step, and quantization noise increases, which causes deterioration of the image, transmitting the signal with the minimum absolute value maintains the image quality. Effective above. When the absolute value of the original high frequency band signal H (j) is minimized, the signal H (j) is transmitted as it is, which is La (5), Lb (5), La (7), and Lb (7) are just the estimated low-frequency band signal L (i), and the prediction signal He (j) based on them is always highly accurate in the high-frequency band signal H. This is because the predicted value of (j) is not always given, and the absolute value of the high band signal H (j) may be the minimum and the coded data amount may be the minimum.

On the other hand, the decompression reproducing section for receiving the transmission signal via the transmission line can reproduce the signal in the following procedure. First, the transmitted encoded signal is decoded by an inverse encoding algorithm, the compression selection type of the high band signal H (j) is confirmed by detecting the identification bit, and the signal selectively transmitted based on the confirmation. Recognizing the compression algorithm of X (j) and reproducing the high band signal H (j) using the predicted signal He (j) obtained from the signal X (j) and the low band signal L (i). Become.

By the way, in this embodiment, a filter circuit similar to that shown in FIG. 10 is constructed. For example, the transfer functions of LPF1 and HPF2 for subband division are given by the following mathematical expression 3,

[0037]

[Equation 3]

Prediction signal He (j) from low band signal L (i)
The transfer function of HPF5 for obtaining can be expressed by the following formula 4.

[0039]

[Equation 4]

In this case, the predicted high band signal He (5) corresponding to (a) above is He (5) = [-L (4) + 4.La (5) -3.L (6)]. / 8 = 3 · [L (4) −L (6)] / 8. Also, LPF1 and HPF2 for subband division
The transfer function of HPF5 may be set to Equation 1 regardless of.

Embodiment 2; This embodiment corresponds to the above-mentioned second invention, and FIG. 2 shows a system circuit diagram of an apparatus for carrying out the method. In the figure, LPF
1, HPF2, HPF5, decimator 3, 4, 6, difference calculation unit 7,
The circuit configurations related to 9, the inversion circuit 8, the transmission selection unit 11, the encoding unit 12, and the identification bit generation unit 13 are almost the same as in the case of the first embodiment (FIG. 1). The feature of this embodiment is that a coding amount comparison unit 14 is provided in place of the absolute value comparison unit 10 of the first embodiment.

For the coding amount comparison unit 14, the difference between the high band signal H (j) and the prediction signal He (j): H (j) -He (j) is given from the difference calculation unit 7. The difference between the high band signal H (j) and the prediction signal He (j): -He (j): H (j) + He (j) from the difference calculator 9 is the high band signal from the decimator 4. Although H (j) is input, the coding amount comparison unit 14 obtains the coding data amount including the respective identification bits IB added correspondingly when the respective signals are selected, and the data amount is obtained. Make a comparison. That is, no matter which signal is selected as the transmission signal, the signal and the identification bit IB added corresponding thereto are transferred to the encoding unit 12 and encoded for transmission. The comparison unit 14 compares the encoded data amounts of the signals when the identification bits IB are added to the respective signals.

Next, the coding amount comparison unit 14 selects the signal having the smallest coded data amount as the transmission signal based on the comparison result, and sends the selection signal SS corresponding to the selection to the transmission selection unit 11. . Then, the transmission selection unit 11 based on the selection signal SS, the differential signal: H (j) -He (j) / the differential signal: H (j) + H.
Either e (j) / high band signal H (j) is selected, the identification bit IB relating to the selection type of the identification bit generation unit 13 is associated and transferred to the encoding unit 12.

Therefore, in the case of the first embodiment, the identification bit IB
The encoded data amount of the transmission signal selected depending on the allocation method may be larger than that when other signals are selected, but according to the present embodiment, the high band signal H (transmitted by the encoding unit 12 is The transmission data amount of the information related to j) is always suppressed to the minimum, and as a result, highly efficient compression transmission can be realized. The reproduction process on the decompression reproduction side is the same as in the case of the first embodiment.

Third Embodiment: This embodiment relates to an improvement of the first embodiment or the second embodiment, wherein the absolute value of the difference signal: | H (j) ± He (j) | If it is smaller, the absolute value comparison unit 10 or the encoding amount comparison unit 14 selects the high band signal H (j) as a transmission signal without performing comparison, and in that case, the high band signal H (j). Also, the identification bit IB indicating that () is selected is not added.

For example, the portions I (1) to I (3) in FIG. 11 are portions in which the change in the image is small, and the low band signals L (2) and L (4) corresponding to those portions in FIG. Has hardly changed. The levels of the corresponding high band signals H (1) and H (3) in FIG. 13 are almost zero. Therefore, the prediction signal He (j) obtained by performing the high-pass processing on the low band signal L (i) by HPF5 also becomes almost 0 in the above-mentioned portion, so that it does not become an inappropriate prediction value. However, the high band signal H (j) in that part is almost 0
Therefore, the effect of using the prediction signal He (j) as the prediction value cannot be expected so much.

In the first embodiment, as described above, the case of transmitting any of the difference signals as the identification bit IB is represented by 2 bits ["11" and "10"], and the high band signal H Since the case of transmitting (j) is represented by 1 bit [“0”], even if the signal level of the high band signal H (j) is close to 0, at least the identification bit IB Since one bit is added, the compression efficiency is reduced. That is, in the portion where the change in the image signal is small, the effect of reducing the data amount by creating the differential signal: H (j) ± He (j) using the prediction signal He (j) is not so great, On the contrary, the identification bit IB causes a decrease in compression efficiency.

Therefore, when the above-mentioned improvements are made and the absolute value of the difference signal: | H (j) ± He (j) | is smaller than a predetermined threshold value, the high band signal H (j) is used as the transmission signal. By selecting and not adding the identification bit IB, the compression efficiency is improved. On the other hand, on the decompression playback side, if the received low band signal L (i) is subjected to high pass processing by applying the same HPF as in the circuit of FIG. 1, a prediction signal He (j) can be obtained, and the prediction The absolute value of the difference signal: | H (j) ± He (j) | is calculated between the signal He (j) and the transmitted high band signal H (j), and if the level value is smaller than the above-mentioned constant threshold value. The signal related to the received high band signal H (j) can be expanded and reproduced from the first bit as the original high band signal H (j).

Fourth Embodiment: This embodiment is also an extension of the first embodiment or the second embodiment, and a signal [difference signal: H (j) -He (j)] related to the high band signal H (j). / Differential signal: H (j) + He
(j) / high band signal H (j)] is divided into a virtual pixel block composed of (m × n) pixels, and the original high band signal H (j) with respect to the previous pixel signal in the block is divided. When selected, the block bit BB indicating the fact is added corresponding to each block, and the identification bit IB is not added to each pixel signal in the block one by one.

In order to enable the method according to the present embodiment, as shown in FIG. 3, a storage unit 15 for storing at least the signal related to the high band signal H (j) for one pixel block is provided. deep. The algorithm of this embodiment is shown in FIG. 4, and at the stage where the pixel signals for each pixel block are stored in the storage unit 15, each differential signal: H (j) ± He for each pixel.
(j) is compared with the absolute value of the high band signal H (j), and the absolute value of the high band signal H (j) is minimized for all one pixel block, and the high band signal H (j) is transmitted. When selected as a signal, one block bit BB is associated with the transmission signal of the one pixel block.

As described above, in the first embodiment, the identification bit IB indicating the selection type is individually added to the signal selected as the transmission signal. The same applies to the third embodiment when a certain condition is not satisfied. By the way, although it is also related to the signal change state of the third embodiment, when it is observed in the entire image or a partial region thereof, the image signal changes only relatively gently in that range, and the high band signal H (j ) May be almost 0. In that case, the condition of | H (j) ± He (j) |> H (j) is satisfied for most of the pixels, and the high band signal H (j) is selected as the transmission signal. Individual high band signal H
An identification bit will be added to (j).

On the other hand, in the above case, the image signal is
When viewed in pixel block units consisting of (m × n) pixels, all pixels are | H (j) ± He (j) |> depending on the pixel block.
It is highly possible that the condition is H (j). Therefore, in such a case, the high band signal H is added to each pixel signal.
The data amount can be significantly reduced by correspondingly adding the block bit BB indicating the fact for each pixel block rather than correspondingly adding the identification bit IB indicating that (j) is selected. That is, compared with the case where 1 bit of the identification bit IB is added to each pixel signal, if 1 bit of the representative block bit BB is added in block units, (m × n−1) bits The amount of data can be reduced by that amount.

Incidentally, on the decompression playback side, by detecting the block bit BB, it is determined that all the signals of the corresponding pixel block are the original high band signal H (j) and decompression playback is performed. Become.

[Embodiment 5] In this embodiment, as in the case of Embodiment 4, the signal related to the high band signal H (j) is divided into virtual pixel blocks consisting of (m × n) pixels. The low band signal L (i) is multiplied by HPF5 to perform high-pass processing, and the prediction signal He (j) and its positive / negative inverted signal: -He (j) are sequentially multiplied by a plurality of different coefficients K, and after the multiplication, The difference between each signal value and the high frequency band signal H (j) is obtained, and together with the high frequency band signal H (j), each differential signal: H (j) −K · He (j), H (j) + K · He Let (j) be a transmission candidate signal related to the high band signal H (j). When the transmission candidate signals are selected in the entire pixel block, the identification bit IB and the coefficient information bit indicating the coefficient K related to the selected signal are added to each block together with the identification bit IB. There is.

The sharpness of the change in the edge portion of the image signal may change depending on the content of the image or the area in the image. For example, there is a large difference between the in-focus portion and the in-focus portion of the camera lens. At a steep edge, a high band signal H (j) having a relatively large amplitude tends to appear with respect to the absolute value of the prediction signal He (j), while at a steep edge, the opposite tendency is observed. .

In this embodiment, as shown by the chain double-dashed line in FIG. 3, multipliers 16 and 17 for multiplying the prediction signal He (j) and its positive / negative inverted signal: -He (j) by a coefficient K are provided. The absolute value comparison unit 10 is configured so that the multiplication coefficient K of the multipliers 16 and 17 can be changed. For example, the value of coefficient K is (5/4), 1, (3/4), (1/2)
Prepare as follows. Then, while changing the coefficient K for each pixel block, each differential signal: H (j) −K ·
He (j), H (j) + K · He (j) is calculated to obtain the high band signal H (j)
It is stored together with the storage unit 15.

As a result, the storage unit 15 stores each differential signal in pixel block units corresponding to each coefficient K: H (j) -K.He (j),
H (j) + K · He (j) is stored, but in the absolute value comparison unit 10, each difference signal and high band signal H for each value of coefficient K are stored.
If the absolute values of (j) are compared and | H (j) ± K · He (j) | ≦ | H (j) | is adopted and any one of the differential signals is adopted as the transmission signal, the coefficient The minimum value of the absolute values of the difference signal: | H (j) ± K · He (j) | that differ depending on the value of K is finally selected as the transmission signal. Further, the identification bit IB indicating the signal selection type, the block bit BB indicating the pixel block, and the coefficient information bit KB of the coefficient K related to the finally selected difference signal are added to the transmission signal.

Therefore, even when the high-frequency band signal H (j) of large amplitude appears at a steep edge, the absolute value of the difference signal can be reduced by selecting the value of the coefficient K, and the difference in the image signal can be reduced. Adaptive data compression is possible for pixel blocks where the edge changes sharply or gently, and efficient transmission is achieved by suppressing redundant transmission data due to the local structure of the image. it can. Then, on the decompression reproduction side, the value of the coefficient K used from the coefficient information bit KB is determined by detecting the identification bit IB, the block bit BB, and the coefficient information bit KB, and related to each pixel of the corresponding pixel block. Difference signal: H
(j) ± K · He (j) is expanded and reproduced to a high band signal H (j). Although the coefficient information bit KB is transmitted in this embodiment, the value of the coefficient K may be transmitted as it is. In that case, the coefficient K of the adjacent pixel block is transmitted.
Redundancy can be suppressed by transmitting the difference value from the value of. Further, in this embodiment, the case where the absolute value of the differential signal: H (j) ± K · He (j) and the high band signal H (j) is compared based on the method of the first embodiment has been described. It can also be applied to the case where the transmission signal is selected by comparing the encoded data amounts as in the second embodiment.

[Embodiment 6] This embodiment is a further extension of each of the above embodiments and relates to a method for increasing the data compression efficiency by dividing an image signal into two-dimensional subbands. First, the configuration shown in FIG. 5 is adopted as a filter circuit for an image signal, and FIG. 6 shows a signal division model diagram in the horizontal direction and the vertical direction. That is, the image signal I (n)
Is sub-band-divided in the horizontal direction and the vertical direction, and the sub-band in the horizontal low-range and the vertical low-range is further sub-band divided in the horizontal and vertical directions. Therefore, in FIG. 6, HL1 is a horizontal high band and a vertical low band, LH1 is a horizontal low band and a vertical high band, and HH1 is a horizontal and vertical high band. And LL2, LH2, HL2, H
H2 is a subband of the horizontal low band and the vertical low band, and LL2 is the horizontal low band of the horizontal low band after the subdivision, and LH2 is the horizontal low band after the redivision. , HL2 corresponds to a horizontal high band and a vertical low band after re-division, and HH2 corresponds to a horizontal and vertical high band after re-division.

The arrows shown in FIG. 6 and FIG.
The prediction signal He (j) and its positive / negative inversion signal obtained by performing the high-pass processing by multiplying the signal L (i) corresponding to the low band side by HPF every time the sub-band prediction is executed in the following order:- He
Difference signal between (j) and the signal H (j) corresponding to the high band side: H
(j) ± He (j) is calculated, and the absolute value of each difference signal: | H (j) ± He
(j) | and the absolute value of the signal corresponding to the high band side: | H (j) | are compared, and the signal having the minimum absolute value is selected as the transmission signal. Further, with respect to those selections, an identification bit indicating the selection type is attached. As described above, in this embodiment, means for two-dimensionally expanding the method of the above-described first to fifth embodiments is provided to further improve the data compression efficiency.

Note that Document 1 discloses a subband prediction method in the order shown in FIG. 9, but the horizontal low band vertically high band and the horizontal high band vertically. No prediction has been made with respect to the higher subbands. Since the data compression efficiency is improved by dividing the horizontal high band in the vertical subband, a higher data compression rate can be obtained in the present embodiment than in the method of Document 1.

Furthermore, for two-dimensional data compression based on inter-subband prediction in the order shown in FIGS. 6 and 7,
As shown by arrows and'and arrows and 'in FIG. 8, a sub band LH2 (LH1) of a horizontal low band and a vertical high band and a sub band HL2 (HL1) of a horizontal high band and a vertical low band are formed. Subband H with horizontal high range and vertical high range from both sides
A method of compressing data while predicting H2 (HH1) can also be adopted. According to this method, the absolute value of the difference signal by the subband prediction in the other arrow direction may be smaller than the absolute value of the difference signal by the subband prediction in one arrow direction, and the difference signal is When selected, the data compression efficiency can be improved.

[Embodiment 7] In the present embodiment, when the image data compression method according to the above embodiment is adopted, when the compressed image data is encoded and requantized, the contents of the image data are changed. The present invention relates to a method of preventing deterioration of image quality by adaptively changing a requantization step in correspondence with the requantization step.

When the image data is finally encoded and transmitted, the transmission amount is restricted by the transmission system, so that it is necessary to control the code amount so that the encoded data amount is constant. The code amount control is performed by controlling the requantization step of the pixel signal and not transmitting the visually insignificant signal component. Generally, when requantizing a pixel signal, it is better to make the interference due to requantization noise less visually noticeable and to roughly quantize the signal in the less noticeable part of the image, which is subband-divided. In the signal coding, by roughly quantizing the signal on the high band side, it is possible to construct an image in which quantization noise is not noticeable when comparing decoded images having the same coded data amount. Further, in the case of dividing the signal on the high band side into a virtual pixel block composed of (m × n) pixels and determining the requantization step for each block, a signal with a large amplitude in the block. The larger the number, the less the quantization noise of the decoded image when the coarsely quantized data has the same encoded data amount.

By the way, as in the above-described embodiment, the inter-subband prediction is performed and the differential signal: H (j) -He (j) / differential signal: H (j) + He (j) / high band signal H. When any one of (j) is selected as the transmission signal, the signal with the smaller amplitude and the signal with the smaller amplitude are selected, so that the general method for determining the requantization step cannot be directly applied. . For example, for a certain pixel block, the difference signal: H
When (j) ± He (j) is selected as the transmission signal, the amplitude becomes small due to the difference signal even though the prediction signal He (j) has a large amplitude, May be coarsely requantized, but it will be requantized in fine steps based on the signal level, and as a result an excessive amount of code will be generated and the effect of data compression will be impaired. On the contrary, a signal of a pixel block to be requantized in fine steps may be requantized in coarse steps.

In the present embodiment, the coding unit 12 in FIG. 1 or 2 has a built-in storage unit, and a signal [difference signal: H (j) -He] relating to the selected high band signal H (j). (j) / difference signal: H (j) + He (j) / high band signal H (j)] is temporarily stored in the storage unit. Then, regarding the stored signal (m
× n) the pixel is divided into virtual pixel blocks, and the number of pixel signals whose absolute value is greater than or equal to a predetermined absolute value is counted for the signal level in each pixel block, and each pixel block is calculated based on the counted number. The class related to the quantization step of is determined.

Next, a large block (a partial area of the screen or one screen or a plurality of screens) in which a plurality of the above pixel blocks are collected is captured, and a requantization step corresponding to the class is performed for each pixel block. Is applied to re-quantize, and the coding amount is set to a predetermined value in the entire large block. Specifically, while changing the basic quantization step related to a large block sequentially, the class is used as a coefficient for determining the requantization step, and the requantization step of each pixel block is changed. The code amount after requantization of the block is detected, and the basic quantization step that gives the predetermined code amount is determined. Here, since the class is an index indicating the degree of the amplitude of the pixel signal in each pixel block, the association of the coefficient with the class is a rough requantization for the pixel block corresponding to the class with a large count number. The quantization step, on the contrary, can be applied to the fine requantization step for a pixel block corresponding to a class with a small count number. In addition, the encoding unit 12 adds information regarding the finally selected basic quantization step and information regarding the class of each pixel block to the large block and each pixel block as requantization condition bits to add the image. Transmit data. Therefore, on the decompression reproduction side, the requantization step of each pixel block can be confirmed by detecting the requantization condition bit, and the confirmation information can be used to reproduce the high band signal H (j). it can.

As a result, according to the present embodiment, as in each of the above-described embodiments, inter-subband prediction is performed and the difference signal: H
(j) -He (j) / difference signal: H (j) + He (j) / High band signal H (j) is selected as the transmission signal, the transmission code amount is kept constant. The requantization step corresponding to the image state can be applied in each pixel block unit, and it becomes possible to transmit the decoded image as compressed image data in which quantization noise is not noticeable.

[0069]

The image data compression method of the present invention has the following effects by having the above-mentioned configuration. According to the invention of claim 1, an image signal is sub-band divided into a low band and a high band, and a low band signal obtained by the sub-band division is subjected to HPF to create a prediction signal of the high band signal. In that case, depending on the state of the image signal, the prediction signal does not uniquely give prediction of the high band signal, and the prediction signal or its positive / negative inversion signal is appropriate as the prediction value, and generally the encoded data amount is Considering that the smaller the absolute value of the signal, the smaller the absolute value of the signal, the difference between the level value of the prediction signal and its positive / negative inversion signal and the level value of the high band signal to be predicted is calculated. Since the signal with the smallest absolute value among the signals is selected as the transmission signal related to the high frequency band signal, it is possible to adaptively and efficiently perform data compression on the image signal and at the receiving side. That allows the reproduction of accurate high-frequency band signal. According to the invention of claim 2, in the invention of claim 1, what signal is selected as the transmission signal related to the high frequency band signal is shown by associating with the identification signal, but the transmission including the identification signal is performed. When the signal is encoded, the transmission signal related to the high band signal selected in the invention of claim 1 may not be the minimum encoded data amount, so the encoded data amount including the identification signal is the minimum. Is selected as the transmission signal related to the high band signal so that the information of the high band signal having the minimum coded data amount can always be transmitted. According to a third aspect of the present invention, the image data is divided into (m × n) pixel blocks, and a plurality of pixel blocks are provided for each signal corresponding to the prediction signal and the positive / negative inversion signal thereof according to the first or second aspect of the invention. Since the results obtained by selectively multiplying different coefficients in sequence are used as the prediction signal, it is possible to select the one with the smallest absolute value or coded data amount from the multiple difference signals using each coefficient. Enables adaptive data compression according to the structure.

[Brief description of drawings]

FIG. 1 is a system circuit diagram of an image data compression apparatus for carrying out the image data compression method of the present invention (the invention of claim 1).

FIG. 2 is a system circuit diagram of an image data compression apparatus for implementing the image data compression method of the present invention (the invention of claim 2).

FIG. 3 is a diagram showing a main part of a system circuit adopted in the case of the fourth or fifth embodiment.

FIG. 4 is a diagram showing an algorithm according to a fourth embodiment.

FIG. 5 is a configuration diagram of a filter circuit in the case where an image signal is two-dimensionally divided into subbands in the sixth embodiment.

FIG. 6 is a signal division model diagram when an image signal is divided horizontally and vertically by the filter circuit of FIG. 5 (Example 6).
It also shows the direction and order of inter-subband prediction in.

FIG. 7 is a signal division model diagram similar to FIG. 5, showing the direction and order of inter-subband prediction following FIG. 6;

FIG. 8 is a signal division model diagram similar to FIG. 5, showing another direction and order of inter-subband prediction.

FIG. 9 is a signal division model diagram showing the direction and order of conventional inter-subband prediction.

FIG. 10 is a filter circuit diagram when an image signal is divided into subbands.

FIG. 11 is a graph showing a level change state of the input image signal I (n).

FIG. 12 is a graph showing a level change state of a low band signal.

FIG. 13 is a graph showing a level change state of a high band signal.

FIG. 14 is a graph showing a level change state of a low band signal and an estimated original signal of the low band signal by each of the zero-order prediction methods [A] and [B].

[Explanation of symbols]

1 ... LPF (low pass filter), 2,5 ... HPF (high pass filter), 3,4,6 ... decimator, 7,9 ... difference calculator, 8 ...
Inversion circuit, 10 ... Absolute value comparison unit, 11 ... Transmission selection unit, 12 ... Encoding unit, 13 ... Identification bit generation unit, 14 ... Encoding amount comparison unit,
15 ... Storage unit, 20 ... Zero-order predictor, BB ... Block bit (pixel block information), H (j) ... High band signal, He (j) ... Prediction signal, H (j) ± He (j), H (j) ± K · He (j) ... difference signal,
I (n) ... Input image signal, IB ... Identification bit (identification information indicating signal selection type), K ... Coefficient, KB ... Coefficient information bit (coefficient information), L (i) ... Low band signal, SS ... Selection signal, X (j)
… Signals that are selectively transmitted.

Claims (3)

[Claims] 1. An image signal is sub-band divided into a low band and a high band, the high band signal is predicted from the low band signal obtained by the sub band division, and the predicted signal is used. In an image data compression method of encoding and transmitting the created high band information together with the low band information, the low band signal is subjected to high pass processing by a high pass filter, and the high band signal is concerned. The positive and negative inversion signal is obtained together with obtaining the prediction signal, and the difference between the level value of the prediction signal and the positive and negative inversion signal thereof and the level value of the high band signal at the sampling point corresponding to the prediction signal is calculated. Of the differential signals and the high band signal, the signal having the smallest absolute value is selected as a transmission signal, and the selected transmission signal is associated with identification information indicating the selection type. A method for compressing image data, characterized in that the image data is added. 2. An image signal is sub-band divided into a low band and a high band, the high band signal is predicted from the low band signal obtained by the sub band division, and the predicted signal is used. In an image data compression method of encoding and transmitting the created high band information together with the low band information, the low band signal is subjected to high pass processing by a high pass filter, and the high band signal is concerned. The positive and negative inversion signal is obtained together with obtaining the prediction signal, and the difference between the level value of the prediction signal and the positive and negative inversion signal thereof and the level value of the high band signal at the sampling point corresponding to the prediction signal is calculated. Of each differential signal and the high band signal, the encoded data amount in a state where the identification information indicating the selection type when each signal is selected is added correspondingly, An image data compression method characterized in that a signal having a minimum data amount is selected as a transmission signal. 3. The image data compression method according to claim 1 or 2, and a signal obtained by subjecting a low band signal in (m × n) pixel block units to high pass processing by a high pass filter, and a signal thereof. The differential signal is created by using a result obtained by sequentially and selectively multiplying the level value of the positive / negative inversion signal by a plurality of different coefficients as a prediction signal, and the image data compression method according to claim 1, 3. The image data according to claim 2, wherein a signal having a minimum absolute value is obtained for each of the differential signal and the high band signal, and a signal having a minimum absolute value is selected as a transmission signal among the obtained signals. Regarding the compression method, a signal having a minimum coded data amount is obtained for each coefficient in the pixel block unit, and a signal having a minimum coded data amount among the obtained coded data amounts is transmitted. When a differential signal is selected for the entire pixel block, the image in which identification information, pixel block information, and coefficient information corresponding to the selected signal are added to the transmission signal in a block unit Data compression method.