CA2023440C

CA2023440C - Transform coding apparatus

Info

Publication number: CA2023440C
Application number: CA002023440A
Authority: CA
Inventors: Yoshiaki Kato; Tokumichi Murakami
Original assignee: Mitsubishi Electric Corp
Current assignee: Mitsubishi Electric Corp
Priority date: 1989-08-19
Filing date: 1990-08-16
Publication date: 1994-11-08
Anticipated expiration: 2010-08-16
Also published as: NO903624D0; EP0414074A3; KR910005691A; DE69016880T2; FI903989A0; EP0414074B1; FI98111C; KR930009872B1; US5086488A; CA2023440A1; NO303480B1; FI98111B; DE69016880D1; EP0414074A2; AU6106990A; NO903624L; AU622572B2

Abstract

This invention relates to a transform coding apparatus where linear transform is performed for input signal series such as image signals, and transform coefficients are quantized from lower frequency compo-nents to higher frequency components in the transform region, and then the quantized transform coefficients are coded and outputted. The quantized transform coefficients being zero are counted, and when the count value exceeds the prescribed threshold value, the quantization is terminated so that high speed in the coding is intended and variation of the code generation rate can be prevented.

Description

2r)23440 TRANSFORM CODING APPARATUS
BACKGROUND OF THE INVENTION
The present invention relates to a transform coding apparatus performing band compression of image data using linear transform coding system.
The prior art will be discussed in detail hereinbelow with respect to the drawings.
In order to solve the problems of the prior art, an object of the present invention is to provide a transform coding apparatus wherein the coefficient transmission range is determined corresponding to the quantized level series within the transform block, and the processing time required for the linear transform and the quantization is reduced and the variation of the amount of the coded bits is decreased.
In accordance with one aspect of the invention there is provided a transform coding apparatus wherein a linear transform operation is performed on an input signal series, and transform coefficients are quantized and coded in sequence from the lower frequency components to 20'~3440 the higher frequency components in the transform region, said transform coding apparatus comprising: blocking means for blocking said input signal series; linear transform means for performing a linear transform operation on the blocked signals and for obtaining a series of transform coefficients in sequence from the lower frequency components to the higher frequency components; quantizing means for quantizing in sequence each of said transform coefficients in said series of transform coefficients outputted from said linear transform means; counting means for counting the number of consecutive zero quantized levels in the quantized transform coefficients quantized by said quantization means and for outputting the count values thereof;
lS threshold determining means for determining a threshold value of the number of consecutive zero quantized levels to be transmitted; zero quantized level output means for outputting zero quantized levels when activated; decision means for obtaining at least one additional transform coefficient from said linear transform means when the count value of consecutive zero quantized levels is less than the threshold value and for activating said zero quantized level output means for subsequent transform coef f icients when said count value equals said threshold value; and coding means for coding the outputs of said quantization means.

,5~ ~

In accordance with another aspect of the invention there is provided a transform coding apparatus wherein a linear transform operation is performed on an input signal series, and transform coefficients are quantized and coded in sequence from the lower frequency components to the higher frequency components in the transform region, said transform coding apparatus comprising:
blocking means for blocking said input signal series;
linear transform means for performing a linear transform operation on the blocked signals; scan transform means for rearranging the transform coefficients outputted from said linear transform means in sequence from the lower frequency components to the higher frequency components;
quantization means for quantizing in sequence each of said transform coefficients outputted and rearranged by said scan transform means; counting means for counting the number of consecutive zero quantized levels in the quantized transform coefficients quantized by said quantization means and for outputting the count values thereof; storage means for storing an event set comprising the number of quantized transform coefficients which are non-zero quantized levels and the number of consecutive zero quantized levels counted by said counting means before the non-zero quantized levels;
threshold determining means for determining a threshold value of the number of consecutive zero quantized levels 20~3440 to be transmitted; decision means for quantizing further at least one additional transform coefficient in said quantizing means when the count value of consecutive zero quantized levels counted by said counting means is less than said threshold value and for terminating the quantization processing of subsequent transform coefficients when said count value equals said threshold value; and coding means for allocating codes to the event set stored in said storage means.
In accordance with yet another aspect of the invention there is provided a transform coding apparatus wherein a two-dimensional linear transform operation is performed on an input signal series, and transform coefficients are quantized and coded in sequence from lower frequency components to higher frequency components in the transform region, said transform coding apparatus comprising: blocking means for blocking said input signal series; first transform means for applying a one-dimensional linear transform operation to the blocked input signal series so as to obtain a one-dimensional transform coefficient block; second transform means for applying an orthogonal one-dimensional linear transform operation further to the one-dimensional transform coefficient block so as to obtain a series of two-dimensional coefficients in sequence from the lower frequency components to the higher frequency components;
, . ...

quantizing means for quantizing each two-dimensional transform coefficient outputted from said second transform means; counting means for counting the number of consecutive zero quantized levels in the values of the quantized transform coefficients quantized by said quantizing means and for outputting the count values thereof; storage means for storing an event set comprising the number of non-zero quantized levels and the number of consecutive zero quantized levels counted by said counting means before the non-zero quantized levels; threshold determining means for determining a threshold value of the number of consecutive zero quantized levels to be transmitted; decision means for obtaining further two-dimensional transform coefficients from said second transform means when said value of consecutive zero quantized levels is less than said threshold value and for terminating the second transform and quantizing process for subsequent two-dimensional transform coefficients when said count value equals said threshold value; and coding means for allocating codes to the event set stored in said storage means.

The present invention will be described in detail hereinbelow with the aid of the ~ nying drawings, in whieh:
FIG. 1 is a block diagram showing a transform eoding apparatus in the prior art;
FIG. 2 is an explanation diagram illustrating property of a transform coeffieient bloek;
FIG. 3 is an explanation diagram illustrating alloeation of eodes;
FIG. 4 is a bloek diagram showing another transform coding apparatus in the prior art;
FIG. 5 is a block diagram showing a transform coding apparatus as a f irst embodiment of the invention;
FIG. 6 is a flow chart illustrating operation of the transform coding apparatus shown in FIG. 5;
FIG. 7 is a block diagram showing a transform coding apparatus as a second embodiment of the invention;
FIG. 8 is a flow chart illustrating operation of the transform eoding apparatus shown in FIG. 7;
FIG. g is a block diagram showing a transform coding apparatus as a third f-'~o~l; 1 of the invention;
FIG. 10 is a flow ehart illustrating operation of the transform coding apparatus shown in FIG. 9;
FIG. 11 is a block diagram showing a transform coding apparatus as a fourth embodiment of the invention and FIG. 12 is a flow ehart illustrating operation of the transform coding apparatus shown in FIG. 11.
A

FIG. 1 is a block diagram illustrating a transform coding apparatus in the prior art disclosed, for example, in W.H. CHEN, W.K. PRATT, "Scene Adaptive Coder", (IEEE
Transactions on Communications, Vol. COM-32, No. 3, March, 1984). In FIG. 1, numeral 1 designates a blocking section for blocking input signals, numeral 2 designates a linear transform section for performing two-dimensional linear transform of the block signals, numeral 3 designates a scan transform section for rearranging signal series within the block, numeral 4 designates a quantization, numeral S designates a coding section, numeral 6 designates a transmission buffer, and numeral 7 designates a coding control section.
Next, operation will be described. The blocking section 1 inputs digitized input image signals 101 by one frame, and divides the signals into two-dimensional blocks which have n pixels in the horizontal and vertical directions (n: natural number, for example, n = 4, 8, 16). The linear transform section 2 applies two-d;~-~c;c~n~l linear transform (for example, orthogonal transform such as discrete cosine transform) to the blocked image signals 102, and produces transform coefficient blocks 103 in space frequency region. In this case, the two-dimensional discrete cosine transform, for example, of 8 x 8 pixel blocks f (x, y) (x, y = ), 1, ..., 7) can be given by the following formula.

':-- A

F(u, v) = l C(u)C(v) ~ ~ f (x, y) 4 x=0 Y=0 x cos[TrU(16+1)] CoS[lrV(2y+1)]
for u, v = 0, l, ..., 7, where -- (for u, v = 0) C (u), C (v) =' (for u, v = l, 2, ., 7) where x, y: coordinates in the pixel region, u, v:
coordinates in the transform region.
Property of the transform coefficient blocks F(u, v) will be described based on FIG. 2. Values of F(u, v) indicate what degree become respective space frequency c ~n~ts included in the blocked image signals 102.
The frequency in the horizontal direction becomes high as the value of u becomes large, and the frequency in the vertical direction becomes high as the value of v becomes large. That is, the value of F(0, 0) corresponds to the intensity of the DC component of the blocked image signal 102, and the value of F(7, 7) corresponds to the intensity of the AC component having high frequencies in both the horizontal and vertical directions.
Consequently, for a monotony image block such as the background where variation of values of pixels is little, significant coefficients of non-zero appear only at lower frequency components and zero coefficients appear at most of higher frequency components. On the contrary, for an image block of the edge portion or the like where variation of values of pixels is much, significant coefficients of non-zero appear not only at lower frequency components but also at higher frequency c~ ents.
Next, the scan transform section 3 rearranges the transfer coefficients within the transfer coefficient block 103, for example, in the order of arrow shown in FIG. 2, and outputs transform coefficient series 104. In the foregoing 8 x 8 pixel block, the coefficient series of consecutive 64 coefficients is outputted for one block. The rearr~ t is performed by zigzag scanning from transform coefficients of lower frequency ~ ts where significant coefficients are liable to occur to transform coefficients of higher frequency ~ ~nts where significant coefficients are not liable to occur, so that the significant coefficients continue at front half and the zero coefficients continue at rear half as much as possible. Next, the quantization section 4 quantizes the transform coefficient series 104 in given quantization step size 109, and outputs quantized level series Q (u, v) 105 . The coding section 5 allocates codes to the quantized level series 105, and outputs coded data 106 to the transmission buffer 6.

The two-dimensional variable length coding will now be described as an example of code allocating method.
This is performed in that the number of consecutive zero quantized level (hereinafter referred to as "run" ) and the non-zero quantized levels are combined for the quantized level series 105, and one Huffman code is allocated for the combined event (run level) . For example, in the case of the quantized level series shown in FIG. 3, the event (run level) becomes as follows.
(0, 20), (2, 15), (4, 5), (3, 2), (7, l), EOB
Where EOB is a mark indicating that non-zero quantized level does not exist since then and zero quantized level continue until the end of the block.
Consequently, in the case of the quantized level series, the predetermined Huf fman codes are allocated respectively to the six events including the EOB.
Next, the transmission buffer 6 smoothes the varying the amount of the coded bits and transmits the code 107 at constant rate to the transmission line. The coding control section 7 performs feedback control of quantization step size 109 adaptively among buffer occupancy 108 being data residual quantity in the transmission buffer 6, and outputs it to the quantization section 4. That is, when the buffer occupancy 108 is much, in order to decrease the amount of the coded bits, the quantization step size 109 is made large and the 2o~?344n input image is quantized coarsely. On the contrary, when the buffer occupancy 108 is little, in order to increase the amount of the coded bits, the quantization step size 109 is made small and the input image is quantized finely. Since the number of non-zero quantized levels is significantly different ~lPrPnf~;n~ on whether the quantization step size 109 is large or small, the amount of the coded bits varies significantly from several ten times to several thr~lc~n~l times.
FIG. 4 is a block diagram showing another transform coding apparatus in the prior art. In FIG. 4, numeral 10 designates a decision section allowing only the transform coefficients within the prescribed transmission range to pass, numeral 11 designates a comparison section comparing the transform coefficient with the threshold value, numeral 12 designates a transmission range detPrm;n;n~ section for determining the transmission range of the transform coefficient, numeral 13 designates a threshold setting section for setting the threshold value, and numeral 14 designates a control section for detPrm;n;ng the quantization characteristics and the threshold value.
Next, operation will be described. The blocking section 1 inputs digitized input image signals 101 by one frame, and divides the signals into two-dimensional blocks which have n pixels in the horizontal and vertical 2~23440 directions (n: natural number, for example, n = 4, 8, 16) . The linear transform section 2 applies two-dimensional linear transform (for example, orthogonal transform such as discrete cosine transform) to the blocked image signals 102, and produces transform coefficient blocks 103 in space frequency region.
Next, the comparison section 11 compares the transform coefficient block 103 with the threshold value 113 given from the threshold setting section 13, and outputs the decision results 112 of which is larger. The transmission range determining section 12 totalizes the decision results 112 for each block, and determines the frequency components to be transmitted and outputs the transmission range 114. That is, the transmission range determining section 12 totalizes the decision results 112 in 64 pieces obtained for the 64 coefficients in the 8 x 8 blocks. As a result, if many decision results indicate that the transform coefficient block 103 is less than the threshold value 111, only lower frequency components may be transmitted and therefore the transmission range 114 is small. On the contrary, if many decision results indicate that the transform coefficient block 103 is more than the threshold value 111, not only lower frequency components but also higher frequency components must be transmitted and therefore 2n~3440 the transmission range 114 becomes large. The sean transform section 3 rearranges the transform eoeffieients within the transform coefficient bloek 103, for example, in the order to arrow shown in FIG. 2, and outputs transform eoeffieient series 104. The rearrangement is performed by zigzag seanning from transform eoeffieients of lower frequeney ,_ -ntS where signifieant eoeffieients are liable to oeeur to transform coefficients of higher fre~uency components where significant coefficients are not liable to occur, so that the significant coefficient series continues as long as possible. Next the decision section 10 outputs only the transform coefficients 110 of the transform coefficient series 104 within the given transmission range 114. The ~uantization section 4 ~uantizes the transform coefficients 110 in given quantization characteristics 115, and outputs quantized levels 105. The coding seetion 5 alloeates eodes based on the cluantization eharaeteristies 115 used in the yuantization section 4, the cluantized levels 105 and the transmission range 114, and transmits the eodes 106 to the transmission line or the like. Also the control section 14 controls the quantization characteristics 115 and the threshold value 113 adaptively by the amount of the coded bits of the input image. That is, when the input image is cluantized eoarsely and the amount of the eoded bits is deereased, both the quantization characteristics 115 and the threshold value 113 are outputted in large values. On the contrary, when the input image is quantized finely and the amount of the coded bits is increased, both the quantization characteristics 115 and the threshold value 113 are outputted in small values.
The transform coding apparatus in the prior art shown in FIG. 1 has problems in that the quantization must be performed for any coefficient, and the amount of the coded bits varies significantly dPpPn~ling on whether the quantization step size is large or small and therefore the control of the coding is difficult.
Also the transform coding apparatus in the prior art shown in FIG. 4 has problems in that although the non-zero transform coefficients to be coded and transmitted are limited, all transform coefficients must be obtained in order to estimate the coefficient transmission range, and therefore the processing time required for the linear transform cannot be reduced.
A first embodiment of the invention will now be described based on FIG. 5 and FIG. 6. In FIG. 5, numeral 21 designates a linear transform section for performing linear transform of blocked signals in the predetPrminPd order, numeral 22 designates a counter for counting the number of the consecutive zero quantized levels, numeral 23 designates a decision section for comparing the count value with the threshold value and performing decision whether the linear transform of next coefficient should be performed or not, numeral 24 designates a zero coefficient output ~ A

2Q~,3 -~LO
section for outputting the zero quantized levels of the given number, and numeral 25 designates a threshold setting section for setting the threshold value.
Constitution of the embodiment is similar to that shown 5 in FIG. 4 except that as above described.
Next, operation will be described. In similar manner to the apparatus shown in FIG. 4, digitized input image signals 101 by one frame are divided into two-dimensional blocks by the blocking section 1, and 10 become blocked image signals 102. The linear transform section 2 applies the linear transform to the blocked image signals 102 in the order shown by arrow in FIG. 2 for example, and outputs a transform coefficient 121 (step STll) . The quantization section 4 quantizes the 15 transform coefficients 121 in given quantization characteristics 115, and outputs quantization coefficients 120 (step ST12). The counter 22 counts the number of the consecutive quantized levels 120 having value zero (steps ST13, ST15) . If the non-zero 20 quantized level 120 is outputted from the quantization section 4, the count value 122 is reset and becomes zero (step ST14 ) . The decision section 23 compares the inputted count value 122 with the threshold value 123 given from the threshold setting section, and process 25 is branched to following processing (step ST16).
(I) If count value ~ threshold value, the decision section 23 outputs the decision result 124 indicating 2~2 that next coefficient is estimated by the linear transform to the linear transform section 21. Then the linear transform section 21 performs the linear transform continuously.
(II) If count value = threshold value, the decision section 23 outputs the decision result 124 indicating that the linear transform is not performed from next coefficient to last coefficient to the linear transform section 21, and outputs the number 125 of coefficients where the linear transform is not performed.
The zero coefficient output section 24 for outputting the zero quantized level series 126 of the given number 125 to the coding section 5 (step ST17).
As already described, in general, since the intensity of the transform coefficients becomes weak from lower frequency components to higher frequency components, the quantized levels become zero consecutively with high probability at higher frequency components. Consequently, when the quantization is performed in the order from lower frequency components to higher frequency components and the zero quantized levels are counted, at the small threshold value, since the above-mentioned condition (II) is satisfied at coefficients of relatively lower frequency components, the processing time required to estimate the transform coefficients can be significantly reduced.

2 ~
The coding section 5 allocates codes based on the quantization characteristics 115 used in the quantization section 4, the quantization coefficients 120 and the zero quantization coefficient series 126, 5 and transmits the codes 128 to the transmission line or the like. Also the control section 14 controls the quantization characteristics 115 and the threshold value 127 adaptively by the amount of the coded bits of the input image. That is, in order to decrease the amount 10 of the coded bits of the input image, since the quantization must be effected coarsely, the quantization characteristics 115 are made large and the threshold value 127 is made small so that the transmission to higher frequency components is not performed. On the 15 contrary, in order to increase the amount of the coded bits, since the quantization must be effected finely, the quantization characteristics 115 are made small and the threshold value 127 is made large so that the transmission to higher frequency components is performed 20 and the picture quality of the decoded image is not deteriorated .
Although combination of the two-dimensional linear transform and the quantization has been described in the above embodiment, similar effects can be obtained 25 also in combination of the one-dimensional or three-dimensional linear transform and the quantization.

2023l.~
A second embodiment of the invention will be described based on FIG. 7 and FIG. 8. Numeral 31 designates a quantization section for controlled by decision result 138 and performing quanti2ation, numeral 5 32 designates a threshold setting section for setting the threshold value based on buffer occupancy 134, and numeral 33 designates a decision section for comparing the count value 136 of the counter 22 with the threshold value 137.
Next, operation will be described. I~igitized input image signals 101 by one frame are blocked and then subjected to linear transform. The scan transform section 3 rearranges the transform coefficients, and the quantization section 31 quantizes the transform coefficients 104 by the given quantization step size 135 and outputs the quantization coefficient series 131 (step ST22). The coding control section 7 determines the quantization step size 135 from the buffer occupancy 134 of the transmission buffer 6, and outputs it to the 20 quantization section 31. The threshold setting section 32 determines the threshold value 137 also from the buffer occupancy 134, and outputs it to the decision section 33. The counter 22 counts the number of the consecutive zero values among the quantized level series 25 131, and if the non-zero quantized level is outputted from the quantization section 31, the count value 136 is reset and becomes zero (step ST24) . The decision 202~
section 33 compares the inputted count value 136 with the threshold value 137 given from the threshold setting section 32, and outputs the decision result 138 to the quantization section 31. The quantization section 31 is 5 branched to following processing according to the decision result 138 (step ST26).
(I) If the decision result 138 indicates that count value < threshold value, next coefficient is quantized also in the same quantization step size 135.
(II) If the decision result 138 indicates that count value = threshold value, the quantization is not performed and the zero quantized levels are outputted from next coefficient to the last coefficient.
In the FIG. 3, for example, when the threshold value 15 is set to 4 or 5, the event and the number of the coefficients performing the quantization become as follows .
When the threshold value is 4, the event becomes (0, 20), (2, 15), EOB, and the number of the 20 coefficients performing the quantization becomes eight from Q(0, 0) to Q(2, 1) .
~ hen the threshold value is 5, the event becomes (0, 20), (2, 15), (4, 5), (3, 2), EOB, and the number of the coefficients performing the quantization becomes 18 25 from Q(0, 0) to Q(3, 2).
As already described, in general, since the intensity of the transform coefficients becomes weak 202~
from lower frequency components to higher frequency components, the quantized levels Q (u, v) being a result of the quantization become zero consecutively with high probability at higher frequency components.
5 Consequently, when the quantization is performed in the order from lower frequency components to higher frequency components and the zero quantized levels are counted, at the small threshold value (111), since the above-mentioned condition (II) is satisfied at coefficients of relatively lower frequency components, the number of the operations and the amount of the coded bits to enable the stopping of quantization and coding of the transform coefficients of higher frequency components can be reduced. Consequently, in the apparatus of the prior art shown in FIG. 1, since the amount of the coded bits is controlled only by the quantization step size 109, the amount of the coded bits varied rapidly depending on whether the quantization step size 109 is large or small, however, the threshold setting section 32 controls the threshold value 137 adaptively thereby the rapid variation of the amount of the coded bits can be suppressed. That is, when the buffer occupancy 134 is little, the threshold value 137 is made small and the coefficient transmission range is limited thereby the amount of the coded bit at the small qUantiZatiOn step size 135 can be suppressed. On the contrary, when the buffer occupancy 134 is large, the 2 0 ~ ~ ~L 4 ~
threshold value 137 is set large thereby coefficients to higher frequency components can be transmitted and the deterioration of the picture quality of the decoded image can be prevented.
Also according to this embodiment, since all transform coefficients need not be necessarily quantized, when the quantization is executed using the digital signal processor or the like, the number of the operations can be significantly reduced and further effects can be obtained.
Although combination of the two-dimensional linear transform and the quantization has been described in the above embodiment, similar effects can be obtained also in combination of the one-dimensional or three-dimensional linear transform and the quantization.
As above described, according to this embodiment, since decision whether the subsequent transform coefficients should be quantized and coded or not is performed depending on the number of the consecutive zero quantized levels, effects are obtained in that the processing time and the amount of the coded bit required for the quantization can be significantly reduced without deteriorating the picture quality of the decoded image .
~ext, a third embodiment of the invention will be described based on FIG. 9 and FIG. 10. In FIG. 9, numeral 42 designates an event storage section for 2 Q 2 3 Ll~
temporarily storing an event comprising value of non-zero quantized level among the quantized level 131 and count value 136 at that time, and numeral 43 designates a code allocation section for allocating codes to the event.
Next, operation will be described.
Digitized input image signals 101 by one frame are divided into blocks of N x N pixels, and then sub jected to linear transform. The scan transform section 3 rearranges the transform coefficients, and outputs the transform coefficient series F (n) 104 (step ST31) . The coding control section 7 determines the quantization step size 135 from the buffer occupancy 134 of the transmission buffer 6, and outputs it to the quantization section 31. The threshold setting section 32 determines the threshold value 137 also from the buffer occupancy 134, and outputs it to the decision section 33. As the initial setting, the count value 136 of the counter 22 is made zero, and the storage content of the event storage section 42 is cleared, and the coefficient number i of the transform coefficient series F(n) 104 constituted by the transform coefficients of N2 is made 1 (step ST32). The quantization section 31 quantizes F (i) being one of the transform coefficient series F(n) 104 in the quantization step size 135, and outputs Q(i) being one of the quantized level series Q(n) 131 (step ST34). If the value of Q(i) is not zero, 2~2~0 the event storage section 4 2 stores set comprising the count value 136 of the counter 22 and Q(i) being non-zero quantized level as an event. And then the counter 22 is reset and the count value 136 becomes zero (step ST36). If Q(i) is not the last quantized level, quantization of next transform coefficient is performed continuously (steps ST40, ST41) . If the value of Q(i) is zero, the counter 22 adds 1 to the count value 136 (step 37). Next the decision section 33 compares the inputted count value 136 with the threshold value 137, and outputs the decision result 138 to the event storage section 42 If the decision result 138 indicates that count value (138) = threshold value (137) (step ST38), or if Q(i) is the last quantized level of the quantized level series Q(n) (step ST40), the event storage section 42 outputs the event 142 stored at present, and the processing of the pixel block terminates (step ST42). The code allocation section 43 allocates Huf fman codes and adds EOB to the outputted event 14 2, and outputs the coded data 143 to the transmission buffer 6. On the contrary, if there is no event 142 to be outputted, since it is an insignificant block, code representing the insignificant block is outputted as the coded data 143 to the transmission buffer 6.
On the other hand, the quantization in the quantization section 31 is controlled by the decision result 138 from the decision section 33, and if count 20.~3'1.
value (136) = threshold value (137), the quantlzation processing is terminated for the subsequent transform coef f icients .
In the FIG. 3, for example, when the threshold 5 value is set to 4 or 5, the event stored in the event storage section 42 and the number of the coefficlents performing the quantization respectively become as f ol lows .
When the threshold value is 4, since the four zero 10 quantized levels continue from Q(S) to Q(8), the condition of terminating the quantization is satisfied and the event stored as set of the zero run and the value of the non-zero quantized level is (0, 20), (2, 15), and the number of the transform coefficients 15 performing the quantization becomes eight from Q (1) to Q ( 8 ) .
When the threshold value is 5, since the five zero coefficients continue from Q(14) to Q(18), the stored event is (0, 20), (2, 15), (4, 5), (3, 2), and the 20 number of the coefficients performing the quantization becomes 18 from Q(l) to Q(18) .
As already described, in general, since the intensity of the transform coefficients becomes weak from lower frequency components to higher frequency 25 components, the quantized level series Q(n) 131 being a result of the quantization becomes zero consecutively with high probability as n becomes large. Consequently, 2023~
as the threshold value 137 is made small, the transmission range of the coefficients is limited and the number of the coefficients requiring the quantization becomes little and the amount of the coded 5 bits is also decreased. Consequently, if the threshold value 137 is controlled in feedback adaptively from the buffer occupancy 134 in similar manner to the quantization step size 135, the amount of the coded bits can be smoothed more finely.
According to this embodiment, since all transform coefficients are not necessarily quantized, when necessary transform coefficients only are obtained, the number of the operation to obtain the transform coefficients can be reduced and further effects can be obtained.
Although combination of the two-dimensional linear transform and the quantization has been described in the above embodiment, similar effects can be obtained also in combination of the one-dimensional or three-dimensional linear transform and the quantization.
Next, a fourth embodiment of the invention will be described based on FIG. 11 and FIG. 12. In FIG. 11, numeral 50 designates a one-dimensional linear transform section for performing one-dimensional linear transform of the blocked image signals 102, and numeral 51 designates another one-dimensional linear transform section for performing one-dimensional linear transform 20~3~
of outputs of the one-dimensional linear transform section 50 in other direction.
Next, operation will be described.
Digitized input image signals 101 by one frame are 5 divided into blocks of N x N pixels by the blocking section 1. The blocked image signals 102 are subjected to, for example, the one-dimensional linear transform operation in the row direction by the one-dimensional linear transform section 50, and become one-dimensional 10 transform coefficient block 151 having N x N elements (step ST51) . As the initial setting, the count value 136 of the counter 22 is made zero, and the storage content of the event storage section 42 is cleared, and the coefficient number i of the two-dimensional 15 transform coefficients of N pieces scanned in the scan order shown in FIG. 2 is made 1 (step ST52). Next, the one-dimensional linear transform section 51 performs the one-dimensional linear transform operation in the column direction being orthogonal to the one-dimensional linear 20 transform block 151, and one of the two-dimensional transform coefficient F(i) of the coefficient number i is outputted (step ST53). The coding control section 7 determines the quantization step size 135 from the buffer occupancy 134 of the transmission buffer 6, and 25 outputs it to the quantization section 41. The threshold setting section 32 determines the threshold value 137 also from the buffer occupancy 134, and 2~3~
outputs it to the decision section 33. The quantization section 41 quantizes the transform coefficient F(i) 152 in the quantization step size 135, and outputs the quantized level Q(i) 153 (step ST54).
5 Next, decision is performed regarding whether the quantized level Q(i) is zero or not (step ST55). If the value of Q(i) is not zero, the event storage section 42 stores set comprising the count value 136 of the counter 22 and Q(i) being non-zero quantized level as an event.
10 In this case, the counter 22 makes the count value 22 zero (step ST56). On the other hand, if the quantization coefficient Q(i) is zero in the step ST55, the counter 22 adds 1 to the count value 136 (step ST57).
The decision section 33 compares the count value 136 with the threshold value 137 (step ST58), and outputs the decision result 138. If the decision result 138 indicates that the count value 136 is equal to or more than the threshold value 137, the one-dimensional linear transform section 51 and the quantization section 41 20 stop the processing (step ST59). Set of the number of the zero coefficients among the quantized levels and the value of the non-zero quantized levels stored in the event storage section 4 2 is outputted to the code allocation section 43 (step ST62) . In the comparison of the count value 136 with the threshold value 137 in the decision section 33 as above described, if the count value 136 is less than the threshold value 137 or if the 20~3~
processing in the step ST56 terminates, the event storage section 4 2 performs decision whether the coefficient number i becomes N or not (step ST60). If the coefficient number i is N, i.e., Q(i) is the last 5 quantized level, the event storage section 4 2 outputs the event 154 stored at present (step ST62), and the processing of the pixel block ends. The code allocation section 4 3 allocates Huf fman codes and adds EOB to the outputted event 154, and outputs the coded data 155 to 10 the transmission buffer 6. On the contrary, if there is no event 155 to be outputted, since it is an insignifi-cant block, code representing the insignificant block is outputted as the coded data 155 to the transmission buffer 6. If the coeffic:ient number i is N or less and 15 Q(i) is not the last quantized level, 1 is added to the coefficient number i (step ST61), and the operation and quantization of next two-dimensional transform coefficient F(i) are performed continuously.
According to this embodiment, since all 20 two-dimensional transform coef ficients are not always necessary, at the one-dimensional linear transform operation of the first time, only the one-dimensional transform coefficients necessary to estimate the two-dimensional transform coefficients are estimated, 25 thereby the number of the operation to obtain the transform coefficients can be reduced and further effects can be obtained.

Claims

1. A transform coding apparatus wherein a linear transform operation is performed on an input signal series, and transform coefficients are quantized and coded in sequence from the lower frequency components to the higher frequency components in the transform region, said transform coding apparatus comprising:
blocking means for blocking said input signal series;
linear transform means for performing a linear transform operation on the blocked signals and for obtaining a series of transform coefficients in sequence from the lower frequency components to the higher frequency components;
quantizing means for quantizing in sequence each of said transform coefficients in said series of transform coefficients outputted from said linear transform means;
counting means for counting the number of consecutive zero quantized levels in the quantized transform coefficients quantized by said quantization means and for outputting the count values thereof;
threshold determining means for determining a threshold value of the number of consecutive zero quantized levels to be transmitted;
zero quantized level output means for outputting zero quantized levels when activated;

decision means for obtaining at least one additional transform coefficient from said linear transform means when the count value of consecutive zero quantized levels is less than the threshold value and for activating said zero quantized level output means for subsequent transform coefficients when said count value equals said threshold value; and coding means for coding the outputs of said quantization means.

2. The transform coding apparatus of claim 1, wherein said threshold determining means outputs the threshold value determined from the previous threshold determination.

3. The transform coding apparatus of claim 1, further comprising a transmission buffer for storing codes outputted by said coding means, and control means for determining the size of the quantization step used by said quantization means based on the buffer occupancy of said transmission buffer.

4. The transform coding apparatus of claim 3, wherein said threshold determining means decreases said threshold value when the buffer occupancy of said transmission buffer is at a low value.

5. A transform coding apparatus wherein a linear transform operation is performed on an input signal series, and transform coefficients are quantized and coded in sequence from the lower frequency components to the higher frequency components in the transform region, said transform coding apparatus comprising:
blocking means for blocking said input signal series;
linear transform means for performing a linear transform operation on the blocked signals;
scan transform means for rearranging the transform coefficients outputted from said linear transform means in sequence from the lower frequency components to the higher frequency components;
quantization means for quantizing in sequence each of said transform coefficients outputted and rearranged by said scan transform means;
counting means for counting the number of consecutive zero quantized levels in the quantized transform coefficients quantized by said quantization means and for outputting the count values thereof;
storage means for storing an event set comprising the number of quantized transform coefficients which are non-zero quantized levels and the number of consecutive zero quantized levels counted by said counting means before the non-zero quantized levels;

threshold determining means for determining a threshold value of the number of consecutive zero quantized levels to be transmitted;
decision means for quantizing further at least one additional transform coefficient in said quantizing means when the count value of consecutive zero quantized levels counted by said counting means is less than said threshold value and for terminating the quantization processing of subsequent transform coefficients when said count value equals said threshold value; and coding means for allocating codes to the event set stored in said storage means.

6. The transform coding apparatus of claim 5, further comprising a transmission buffer for storing codes outputted by said coding means, and control means for determining size of the quantization step used by said quantization means based on the buffer occupancy of said transmission buffer, wherein said threshold determining means determines the threshold value based on the buffer occupancy of said transmission buffer.

7. The transform coding apparatus of claim 6, wherein said threshold determining means decreases said threshold value when the buffer occupancy of said transmission buffer is at a low value.

8. A transform coding apparatus wherein a two-dimensional linear transform operation is performed on an input signal series, and transform coefficients are quantized and coded in sequence from lower frequency components to higher frequency components in the transform region, said transform coding apparatus comprising:
blocking means for blocking said input signal series;
first transform means for applying a one-dimensional linear transform operation to the blocked input signal series so as to obtain a one-dimensional transform coefficient block;
second transform means for applying an orthogonal one-dimensional linear transform operation further to the one-dimensional transform coefficient block so as to obtain a series of two-dimensional coefficients in sequence from the lower frequency components to the higher frequency components;
quantizing means for quantizing each two-dimensional transform coefficient outputted from said second transform means;
counting means for counting the number of consecutive zero quantized levels in the values of the quantized transform coefficients quantized by said quantizing means and for outputting the count values thereof;

storage means for storing an event set comprising the number of non-zero quantized levels and the number of consecutive zero quantized levels counted by said counting means before the non-zero quantized levels;
threshold determining means for determining a threshold value of the number of consecutive zero quantized levels to be transmitted;
decision means for obtaining further two-dimensional transform coefficients from said second transform means when said value of consecutive zero quantized levels is less than said threshold value and for terminating the second transform and quantizing process for subsequent two-dimensional transform coefficients when said count value equals said threshold value; and coding means for allocating codes to the event set stored in said storage means.

9. The transform coding apparatus of claim 8, further comprising a transmission buffer for storing codes outputted by said coding means, and control means for determining size of quantization step used by said quantization means based on the buffer occupancy of said transmission buffer, wherein said threshold determining means determines the threshold value based on the buffer occupancy of said transmission buffer.

10. The transform coding apparatus of claim 9, wherein said threshold determining means decreases said threshold value when the buffer occupancy of said transmission buffer is at a low value.