JPS60146363A - Vector encoder between action compensating frames - Google Patents

Abstract

PURPOSE:To encode signals to a low bit rate at a high efficient level and to realize an easy control of quantity of occuring information by quantizing application vectors after action compensating prediction error signals have been decided whether they are significant or not by block. CONSTITUTION:Mean values of prediction error picture signals 123 are separated and normalized at every block through a mean value separator circuit 23 and a normalizing circuit 24, and an input vector 129 is transmitted to a distortion calculating circuit 28. At this time an output vector address counter 26 counts up sequentially and reads out an output vector from an output vector code table 27. The distortion calculating circuit 28 calculates distortion between input/ output vectors, transmits distortion 132 between respective output vectors to a minimum distortion detector 29 and transmits a strove signal 133 so that an output vector address 134 of the counter 26 will be fetched into an index latch 30. On the other hand, in a significance decision circuit 25, a mean value and fluctuation gains when a significance decision discriminating index is significant are outputted in accordance with a mean value and fluctuation gains of a block of the prediction error picture signal 123.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a motion compensated interframe encoding device that removes redundancy and realizes highly efficient encoding by utilizing intraframe and interframe correlations of moving image signals.

First, the principle of motion compensated interframe vector encoding applied to conventional devices of this type will be explained. In FIG. 1, it is assumed that the object moves from position A to position B in the period from the f-1th frame to the fth frame by one step. At this time, a block s% of the image signal at the position where the motion vector detector was located is calculated from the position vector R in the f-1 frame on the moon, which is a block s% that collects a plurality of grid-like samples of the image signal centered on the position vector R in the f-th frame. Block S”
(R-r) is equal to zo. As shown in Figure 2, position vector R-(m, n) and motion vector) A/ y +e
If s (u, v), then s = sf' (R-r)
It is. now.

The image at Rw (m, n) is S (R) + w (8 (m-2,, n-2), -j8 (m
, n), -, S (m-1-2, +1+2) :I, then 5f(nt to s'' (R-r) to M([L(
u, v)'15
It is defined as m+g-u, n-)h-v]b. At this time, the motion vector T is. In other words, S and 5f-1(R-r) have similarity L(u, v
) should be minimized as a result of block matching.

To introduce motion compensation in inter-frame predictive coding, at the time when the block S (R) of the image signal at position B of the f-th frame is given as input to the inter-frame predictive encoder, The block with the minimum L (nsv) among the blocks of the image signal is designated as f-1.
If the signal is inserted from the image signal block of the frame and used as a prediction signal, the prediction error signal power is minimized and coding efficiency is improved.

FIG. 3 shows the bottom side of a conventional device of this type.

In the figure, uu is /. converter, (2) is a raster/block scan converter, (3) is a frame memory, (4) is a motion vector detector, (5) is a variable delay circuit, (6) is a subtractor, and (7) is a scalar. The embellisher, (8) is the adder, (9
) is a variable length encoder.

Next, the operation will be explained. First, an analog image input signal (101) is digitized by the digital converter 11+, and an image signal series (102) is output in accordance with a raster scanning sequence. The Lascoux scanning digital image signal series (102) passes through the raster/block scanning converter (2) and converts the time-series output signal WA of the two image signals into block scanning, and scans one screen from top to bottom from left to right. This results in a block scan image input signal (10!l) arranged in grid-like block units (the inside of the block is raster scanned).

From the frame memory (3), the reproduced image signal (104) of the previous frame is reproduced in the interframe DPCM loop.
is read out. The motion vector detector (4) performs block matching between the current block scanned image input signal (105), the one-frame previous reproduced image signal (104), and the image signal, and performs block matching of the current block-scanned image input signal (105), one-frame previous reproduced image signal (104), and the one-frame previous image signal with the minimum similarity. The motion vector (104) of the image signal (104) is r's(u, v
). The motion vectors (11, v) U each correspond to pixel shifts of the block in the horizontal and vertical directions of the reproduced image signal (104) one frame before. Based on the motion vector (104), the variable delay circuit (5) converts all the image signals closest to the current block scan image input signal (103) by one frame previous reproduced image signal (1o4).
) km vector depth and outputs it as a predicted image signal (10(S)).The subtracter (6) converts the block scanning image input signal (103) and predicted image signal (106) into blocks.
) and calculate the pixel-by-pixel difference of the prediction error image signal (1o7).
It is output to the k scalar smoother (7). The prediction error image signal (107) whose power has been reduced by the motion compensation is processed by a scalar generator (7) having the error conversion characteristic shown in FIG.
), the prediction error is converted into an image signal (1o8) in which the quantization level is reduced in pixel units. The predicted error quantized image signal (108) and the predicted image signal (1o6) are added in an adder (8), and a reproduced image signal (109) containing the scalar digitization error is sent to the frame memory (3). The frame memory (3) stores the current reproduced image signal (109
) executes a one-frame delay operation.

1iL in the above motion compensated inter-frame DPCM loop
ii iQ large input B (1os) i sf (m * n
) #Predicted image signal (106) iPf (m, n),
Prediction error signal (107)'iεf(m, n), scalar garbled noise t-qs(m, n).

If the prediction error quantized signal (10B) is ε(m, n), the reproduced image signal (109) i8 (m, n), and the reproduced image signal one frame before (104) i8 (m, n), then gf (n+,n)ms'(m-,n)-Pf(m,n)
Liao (m, n) - ε (m, n) + Q (m, n) Sf (m
, n)-P (m, n) 1ε (In*n)-8(m
, n) 10Q (m,n)'g'(m,n)-Bf(m,n)Z', where 2 indicates the delay gt of one frame.

Pf(fII, n) is determined by motion compensation based on g'' (m, n) using the following equation.

Pf(m,n)-8' 1(mu-u,n-v)Motion vector detector to perform motion interpolation (&+41
It's a cypress tree.

In the figure, ill is a similarity division circuit, [111 is a motion area line memory, U is a line memory control circuit, Q1 is a similarity ratio IVi circuit, and U is a motion vector latch.

In the motion vector detector (4), the current uniform input signal (
103) Load the sequence into a plurality of blocks into a 5(R)td similarity calculation circuit H. At this point, frame memory (
The reproduced image signal (104) one frame before 3) is 8' (
Only lines corresponding to the tracking range of the motion region R) are stored in the motion region line memory tlz. The line memory control circuit a sequentially calculates the similarity iti of the blocks S (R+r) in the vicinity of the plurality of blocks S (R+r) of the raw image signal (104) sequentially one frame from the motion area line memory Q2. It is sent to circuit a1. The similarity calculation circuit Q (9 calculates the similarity L(u, v)'fI of the blocks near S (R result Sf-''(R-r)) and calculates the similarity L(u, v)'fI: In order to correspond to the address shift of the horizontal and vertical motion area line memory till, the motion detection strobe (m) is sent from the similarity comparison circuit a3 to the motion vector latch (14) at the point in time when the similarity is the minimum, and the motion vector Take in address (112)'i. Motion vector latch Q4il: Displacement rf motion vector (105) with respect to S (R) of 5f1 (R-r) where similarity L (+1.V) is minimum The variable delay circuit shown (
5) and the variable length encoder (9).

The variable length encoder (9) in FIG. 3 encodes the motion vector (
105) and the prediction error quantized signal (108) are variable-length encoded to reduce information fth of the image signal. The motion compensated interframe encoded output (110) can be transmitted at a low bit rate through the above processing.

Since the conventional motion-compensated interframe encoder is configured as described above, motion compensation is calculated on a block-by-block basis, and inter-frame DPCM is calculated on a pixel-by-pixel basis. For this reason, picture m1
The effect of distinguishing between minute fluctuations and noise cannot be obtained, and it is difficult to perform variable length encoding of the @ vector and the prediction error quantized signal. Furthermore, since it is difficult to fully control the changes in information generated due to fluctuations in movement stitching, there is a large loss when transmitting through a transmission line having a constant transmission capacity. Furthermore, the prediction error F±child signal is encoded pixel by pixel, which is inefficient. Since the motion compensation R' method is vulnerable to transmission path errors, if a transmission path error occurs, it is necessary to reset the frame memory and retransmit it, but in that case it also takes a long time to recover.

In order to solve all of these drawbacks, this invention performs adaptive vector quantization after determining the significance of the motion compensation prediction error signal on a block-by-block basis, thereby achieving highly efficient coding at a lower bit rate and making it easier to perform variable length coding with motion vectors. It is an object of the present invention to provide a motion compensated interframe vector encoder that can easily control information generation.

One drawing will be explained in detail below.

FIG. 6 is a block diagram showing an example of a motion compensated interframe vector encoder according to the present invention.

In h and w are multi-access frame memo! I and h are adaptive vector quantization encoders, and (c) is an adaptive vector quantization post-coder. In the figure, parts with the same number as the third part indicate the same or equivalent parts.

FIGS. 7 and 8 are block diagrams showing one embodiment of an adaptive vector appendix encoder and an adaptive vector appendix No. 13 coded word according to the present invention.

In the figure, (2) is an average 1-direction separation circuit, and (c) is a normalization circuit.

(c) is a significance determination circuit, (a) is an output vector address counter, @ is an output vector code table, and ＼ha is a distortion i
It calculation circuit, (2) is a minimum distortion detector, OI is an index latch, 0υ is an average value correction circuit, (2) is an amplitude regeneration circuit, and C (31 is a reproduction vector)/register. In the drawings, parts with the same numbers indicate the same or equivalent parts.

Next, the operation of the motion compensated frame-to-frame vector encoder according to the present invention will be explained with reference to FIG.

The block scan image input signal (1O3) from the raster/block scan converter (2) is sent to the motion vector detector (4).
Further, a block scan image input signal (120) delayed by several lines so as not to overlap with the motion compensation area is sent to the calculator (6) for frame same vector encoding. The motion vector detector (4) receives the image input signal (1
The motion vector (105) is determined through the same procedure as the motion vector detection described above based on the reproduced image signal (121) of one frame before and the variable delay circuit (5) The prediction quotient (122) corresponding to the position of the input signal (120) is controlled to be output to the subtractor fGl.The prediction error image signal (123) which is the output of the subtractor (6) is The vectors are divided into blocks and combined into blocks, and are vector-chilled through an adaptive vector quantization encoder (128) and an adaptive vector quantization decoder (129).The vector is a chilled prediction error image signal (125). becomes a predicted error vector mutated image signal (125) containing vector mutated noise, and the predicted image signal (122) is added in an adder (8) to form a reproduced image signal (126). Reproduced image signal ( 126) is written in an area that does not overlap with the motion compensation execution area of the multi-access frame memory (a).

Since the above-mentioned interframe encoding process is executed in block units, each block is defined as a pixel vector. The image input signal (120) is converted into sX, and the predicted image signal (122)
t-bow, the prediction error signal (123) is εt, the prediction error vector needle correction signal (125) is ? It is assumed that the vector quantization noise is oX, the reproduced image source signal (,126) is included, and the reproduced image signal (121)^f-1 of one frame before is St. 1 indicates the block sequence grant. At this time, the frame direction encoding performance 'J4-tfi is expressed by the following equation.

11 ξt-+ll+Ql j tη0Pt-5t-1-Q＾f-1Afgo l Predicted image signal (122) iJ: Motion compensation processing? Accepted 1
It is formed by cutting out all the VCC ant blocks from Flemert Danrei No. 1B (122) so that the degree of similarity is small. The blocks of the predicted image signal (122) cut out and formed by motion compensation do not need to match the boundaries and block sizes of the blocks handled by frame vector coding. That is, motion compensation is performed by sliding block matching, and frame co-vector encoding is performed using fixed blocks.

Next, we will explain the operation of the adaptive vector generator that converts the prediction error -1 (g (123)) into a 4-corresponding vector while constructing the prediction error in block units to achieve efficient coding. The vector chiller is constituted by a cascade connection of an adaptive vector encoder Qυ and an adaptive vector encoder Qυ.

First, the principle of adaptive vector correction of a prediction error signal according to the present invention will be explained.

Prediction error image signal Ku [ε
ε ..., ε1] is an image t 11 as shown in Figure 9.
21 Block P2 of predicted image signal from block SZ of force signal
can be obtained by converting each yuan. The prediction error image signal et is subjected to the above conversion in order to perform significance determination and adaptive vector obfuscation.

, to μ1-Kj, Ej δ-x 1 ε-μm l j t-1 xj-(εj-μt)/δ7 xt-[xl, x2. , xk) That is, the average value μ is separated and normalized by the amplitude δ□ to obtain the human vector xt. The input vector Xt is vector quantized in a 2-dimensional signal space and mapped to an output vector y1 with minimum distortion. Output vector y-Cy, ym・
.

lj−δt“yij ten μt 81 = [C4, C2, ..., ε] Also, if μ T and δ2≦−T2, then f,=.

IT1 and T2 are set as the level determination threshold (tZa) of C2. In other words, the significance or non-significance is determined in block units from the prediction error image signal, and! Only when there is a change in the image signal level between frames at a predetermined level even after the compensation is performed, the prediction error image signal is pH amplitude normalized by 0 average value and converted into a vector B-child.

Below a predetermined level, it means that there is no motion or that all images can be reproduced with motion compensation. If block-6 of the prediction error image signal has a significance determination identification code ν, then μz<T, and if C2<T2, shear 0μ7>T,
Or, if C2>T2, if t = 1-1, the block can be encoded with 1 bit. Also T...
Threshold value of T2 (12B) 'i If controlled, information generation 1tt
- Can be kept constant.

Next, the principle of vector digitization in which the input vector of the prediction error image signal is encoded with ultra-low pit-ray ()f and high efficiency will be explained.

Input vector X−[Xl in K-dimensional signal space aK
s x2 #”・1%]K, RK(1) N
95 divisions R1°R2, . . . , RN. Output vector y = (y y ・
...yYlk) set i i1'x2' Y-(yy ...a THE also.y's in1
# Let the 2-step index set be the knee [1, 2j...s'').

At this time, vector suture conversion Vq is expressed as a cascade connection of encoding C and decoding.

vo(X+ y i i t x RIC: X-*
i if d(x,y')(d(x,y) for
al, l jl j D:1→y The mechanical strain temperature a(x, y,) represents the path separation between the input and output vectors in a two-dimensional signal space, and in the absolute value type measure, d(x, y
) -X l:lc, -YIJll j=1. At this time, the data rate of index 1, which is the vector numeric encoding output, is K tog2N bits/
It is a pixel. That is, vector stitching realizes highly efficient encoding by encoding the input vector X and the index l of the output vector y that has the minimum distortion min d (x, yi). For decoding, it is only necessary to convert to the output vector y corresponding to index 1. output vector y
□) Y may be obtained from the actual input vector X through clustering training, or may be determined from a predetermined input vector probability model. Figure 10 shows the relationship between input and output vectors. This adaptive vector encoding output has a significance judgment identification index ji and an average value μ.

is the amplitude gain δ6. and output vector index 1.

The operation of the above-mentioned adaptive vector stitching encoder will be explained with reference to FIG.

The prediction error image signal (125) is average-separated and normalized block by block through the mean-value separation circuit (c) and the normalization circuit, and then sent to the input vector (129) or the distortion meter circuit. At this point, the output vector address counter (a) sequentially counts up and reads the output vector (131) from the output vector code table (c). The distortion calculation circuit (to) calculates the distortion a(x, y□) between the input and output vectors and calculates the distortion (132) with each output vector.19. The signal is sent to the minimum distortion detector (a). In the minimum distortion detector (c), when the distortion between the input and output vectors becomes the minimum, at that point the output vector address (134) of the output vector address counter e corresponding to the index of the output vector is taken into the index latch (to). strobe signal (15
3) is sent. The index latch (to) outputs the index of the output vector that produces the minimum distortion with respect to the input vector. The ``1'I'' constant circuit (c) outputs the average value and amplitude gain in the case of the significance determination identification index latch based on the average value and amplitude gain of the ``previously off-shore'' erroneous ground image signal (flock CI of 12Q). The adaptive vector aggregation encoder output signal (124) is a significance determination identification index,
The prediction error in the case of significance is the average value of the image signal, the amplitude gain, and the output vector index.

In the adaptive vector merging encoder shown in FIG.
The output vector code table (c) corresponding to the output vector index is read out and a predicted error vector quantized image signal when significant is calculated by the amplitude reproduction circuit clIl and the average value correction circuit C33. In the reproduced vector register (to), if the significance determination identification index is insignificant, the contents of the register are reset to zero, and the predicted error vector needle-shaped image signal (125) i is finally output.

As shown in FIG. 11, the above interframe vector encoding processing and motion compensation processing are performed using a vector detector (4) with a multi-frame access frame memory mtm and a variable delay circuit ll\is (51) so that they do not overlap in time. It is desirable that the

As described above, the encoded data subjected to the motion compensation interframe adaptive vector needle are the motion vector, the significance determination identification index, the prediction error image signal average value in the case of significance, the amplitude, and the output vector index. These are variable-length encoded by the variable-length encoder (9) in FIG. 6 and sent to the transmission section as a motion-compensated interframe vector encoded output (127). Note that it is desirable that the block size for executing block matching for motion compensation and the block size for vector quantization be equal in the horizontal and vertical directions, or have a relationship of integral multiples. At this time, there is an advantage that the motion vector and the adaptive vector encoding output can be variable-length encoded together in block correspondence. Furthermore, the generated information is more responsive to movement, so encoding is easier. Also,
The significance determination value of adaptive vector quantization can be transmitted at a constant rate when a signal occurs by observing the increase or decrease in the variable length encoded output and performing feedback control.

As described above, according to the present invention, similar blocks are searched in the horizontal and vertical directions between frames and motion compensation is used as a prediction signal, and prediction error signals whose predictions are not correct are subjected to threshold processing on a block-by-block basis. By using adaptive vector quantization, it is resistant to noise, and the precise control of alarm generation is easy.
This method has the advantage of being able to efficiently encode high-quality images at low bit rates.

[Brief explanation of the drawing]

Figure 1 is an explanatory diagram that defines the change due to #I between frames of an image signal as a motion vector, Figure 2 is an explanatory diagram of the correspondence between the motion vector and the pixel array of a motion block, and Figure 3 is an illustration of the conventional motion A block diagram of an embodiment of a compensation interframe encoder, and FIG. 4 is a diagram of the input/output characteristics of the scalar encoder
5 is a block diagram showing an embodiment of a motion vector detector, FIG. 6 is a block diagram showing an embodiment of a motion compensated interframe vector encoder according to the present invention, and FIG. 7 is a block diagram showing an embodiment of a motion vector detector according to the present invention. FIG. 8 is a block diagram showing an embodiment of the adaptive vector character post-coder according to the invention; FIG.
The figure is an explanatory diagram of a prediction-stroke j event signal which is an input signal for adaptive vector quantization according to the present invention. FIG. 10 is an explanatory diagram showing the relationship between input and output vectors of vector quantization, and FIG. 11 is an explanatory diagram showing the temporal relationship between on-screen processing of motion compensation and adaptive vector quantization processing. In the figure, fi+ is the negative group, (2) is the raster/block scan converter, (3) is the frame memo 'J, (
41 is a motion vector detector, (5) is a variable delay circuit, (
6) is a subtractor. (7) is a scalar quantizer, (8) is an adder, (9) is a variable length encoder, - is a similarity calculation circuit, all is a motion area line memory, θ2 is a line memory control circuit, 09 is a similarity In this comparison circuit, 0 (a) is a motion vector latch, (4
) is a multi-access frame memory, I211 is an adaptive vector stitch encoder, @ is an adaptive vector chill decoder, @ is an average value separation circuit, (goods) is a normalization circuit, (c)
is a significance judgment circuit, (A) is an output vector address counter, @ is an output vector code table, (C) is a distortion calculation circuit, 4 is a minimum distortion detector, (to) is an index launch, aυ is an amplitude recirculation circuit, 0 is the average value correction circuit. (2) C is a raw vector register. In the drawings, the same or corresponding parts are designated by the same reference numerals. Agent Masuo Oiwa Figure 3 Figure 4E 5vA Figure 6 Figure 71! ! ! 1 Figure 8 Figure 9 Sf. Figure 10 χ2 Figure 11 1 --- ÷ Ice direction

Claims (1)

Scope of Claims: A frame memory that stores past image signals for at least one frame, an image input signal block that is formed into a block by setting Kt for each current image input signal, and one frame that is read from the frame memory. A motion vector detection unit that moves a previous image signal block, which is obtained by combining a plurality of previous image signals that are one or more minutes ahead, in the horizontal and vertical directions and outputs the movement of the most similar previous image signal block as a motion vector. , a variable delay circuit that reads out a previous image signal block most similar to the current image input signal block from the contents of the frame memory based on the motion vector with a predetermined delay and sets it as a predicted image signal block;
! A subtracter that calculates a prediction error image signal block by calculating $C between the il image input signal block and the predicted image signal block in corresponding pixel units, and inputs the average value separated from the prediction error image signal block and normalized by an amplitude gain. form a vector. Search for an output vector with the minimum distortion from an output vector code table that stores a set of output vectors generated from a probability model of input vectors in advance, set its address as an output vector index, and set the average value and the amplitude gain to a predetermined value. Adaptive vector needle coding that determines whether a prediction error image signal block is significant or insignificant by comparing it with a threshold value, and outputs a significant identification index, the above-mentioned average value, amplitude gain, and output vector index in the case of presence. receiving the output of the adaptive vector quantization encoder, reading an output vector corresponding to an output vector index from an output vector code table having the same contents as the adaptive vector quantization encoder;
After multiplying the output vector by the amplitude gain, the average value is added to reproduce the prediction error vector quantized image signal block, and if the significant identification index is not significant, output the prediction error vector quantized 1liII image signal block as zero. An adaptive vector quantization decoder. Adding the predicted error vector quantized image signal block, which is the output of the adaptive vector quantization encoder, and the predicted image signal block, the frame memory delays the result by one frame and uses it as a past image signal. Adder that calculates the reproduced image signal for. A motion compensation frame characterized by comprising: a variable length encoder that collectively variable length encodes the motion vector, the significant identification index, the average value, the amplitude gain, and the output vector index of the predicted incorrect image signal block; Interval vector encoder.