JPS6326951B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an interframe encoding device that highly efficiently encodes an image signal (reduces the bit rate) by utilizing the phase difference between successive screens. A conventional interframe encoding device of this type was constructed as shown in FIG. In FIG. 1, 1 is a subtracter, 2 is a scalar quantizer, 3 is an adder, and 4 is a frame memory. Now, the sample sequence of the image input signal 5 of the f-th frame that is sequentially raster scanned from left to right from the top to the bottom of the screen is S ^f _K (here, K is the sample sequence number in raster scanning), and the above S ^f The predicted signal 6 composed of sample values of one frame before _K is S^ ^f-1 _K , and the image input signal 5 and the predicted signal 6 are
The inter-frame difference signal 7 is X ^f _K , the scalar quantized signal 8 for each sample of the inter-frame difference signal 7 is X^ ^f _K , and the reproduced image signal 9 is the sum of the scalar quantized signal 7 and the prediction signal 6 Let be S^ ^f _K. At this time, in the encoder of FIG. 1a, the image signal is encoded with high efficiency through the following processing. X ^f _K =S ^f _K −S^ ^f-1 _K X^ ^f _K =X ^f _K +q ^f _K S^ ^f _K =S^ ^f-1 _K +X^ ^f _K =S^ ^f-1 _K +X ^f _K +q ^f _K That is, the inter-frame difference signal X ^f _K whose signal power decreases with respect to S ^f _K is expressed as its amplitude probability distribution P (X ^f _K )
High efficiency encoding is achieved by encoding the scalar quantized signal X^ ^f _K by reducing the number of levels so that the distortion is minimized based on . q ^f _K is the quantization noise of the system caused by the decrease in the scalar quantization level. The decoder shown in Figure 1b performs the calculation S^ ^f _K = S^ ^f-1 _K +X ^f _K +q ^f _K based on the transmitted scalar quantized signal 8, and eliminates the quantization noise ^f _K A reproduced image signal 9 containing S^ ^f _K is obtained. As described above, the conventional interframe encoding device scalar quantizes the difference signal between the image signal of the current frame and the image signal of the previous frame corresponding to the same position by reducing the number of levels for each sample. According to this method, since the image signal includes Gaussian noise, not only does the interframe difference signal 7 exist at a predetermined level even in areas with no movement, but also the amount of information generated regarding images with a lot of movement increases, so the transmission rate remains constant. It is difficult to perform high-efficiency encoding while controlling This invention was made to eliminate the above-mentioned drawbacks, and is resistant to disturbances caused by Gaussian noise.
The object of the present invention is to provide an interframe encoding device that vector quantizes an interframe difference signal by setting a threshold so that the encoding efficiency does not drop significantly even for moving images. . Hereinafter, the invention will be explained in detail with reference to the drawings. Now, as shown in Figure 2, samples on four temporally consecutive scanning lines of a certain f-1th frame (in the case of television, one frame consists of two interlaced scanning fields) 4×
The four blocks are grouped together into a block, which is expressed as one vector, S _f-1 = [S ₁ , S ₂ . . . S ₁₆ ] _f-1 . On the other hand, in the f-th frame one frame later, S _f-
Let S _f = [S ₁ , S ₂ , . . . , S ₁₆ ] _f be a sample block corresponding to the same position on the screen as ₁ . Furthermore,
From the f-th frame line signal vector S _f to the f-
It is assumed that an input vector X _f of an inter-frame difference signal obtained by subtracting the image signal vector S _f-1 of one frame for each element (sample) = [X ₁ , X ₂ , . . . , X ₁₆ ] _f .
High efficiency encoding is realized by vector quantizing this interframe difference signal X _f in a 16-dimensional Euclidean signal space. FIG. 3 shows a block diagram of an embodiment of a vector quantization interframe coding apparatus according to the present invention. In the figure, 10 is a raster/block scan converter;
11 is a vector quantization encoder, 12 is a vector quantization decoder, 13 is a block scan frame memory, and 14 is a block/raster scan converter. The operation of this interframe encoding device is as follows. First, in the encoder of FIG. 3A, the image input signal 5 of the f-th frame is converted on the screen from raster scanning to block scanning as shown in FIG. 2 by a raster/block scan converter 10. In block scanning, each sample of the image signal is [S ₁ , S ₂ ,...
..., S ₅ , S ₆ , ..., S ₉ , S ₁₀ , ..., S ₁₃ , S ₁₄ ,
S ₁₅ , S ₁₆ ] _f are output in this order, and this block is scanned moving from left to right on the screen and then from top to bottom. This block scanning image signal 15
is the block scan predicted image signal 16 [S^ ₁ , S^ ₂ , ..., S^ ₁₆ ] _f corresponding to the same position of the f-1th frame which is similarly block scanned and read out from the block scan frame memory 13. _-1 can be subtracted.
The subtracter 1 converts the inter-frame difference signal 17 into X ₁ , X ₂ ,
..., output in the order of X ₁₆ . The inter-frame difference signals 17 obtained in this way are combined into blocks and sent to the vector quantization encoder 11 to form an input vector X _f =[X ₁ , X ₂ , ..., X ₁₆ ] _f , It is transformed (mapped) into an output vector y _i =[y _i1 , y _{i2 ,} ^.
The index i of this output vector is the output vector encoded signal 18 of the vector quantization encoder 11.
Sent as . Further, the output vector encoded signal 18 is input to the vector quantization decoder 12 and converted into an output vector y _i , and the vector quantized inter-frame difference signal 19 is read out in the order of y _i1 , y _i2 , ... y _i16 becomes. This vector quantized inter-frame difference signal 19 is the block scan predicted image signal 1.
6 and are added for each sample at the corresponding position on the screen, and the reproduced image signal 20 is S^ _f = [S^ ₁ , S^ ₂ , . . .
S^ ₁₆ ] Obtain _f . The above processing is expressed by the following equation. X _f = S _f − S^ _f-1 y _i = x _f +q ^f S^ _f = S^ _f-1 + y _i = S^ _f-1 + x _f + q ^f , where q ^f is the 16-dimensional signal space is the vector quantization noise in . In the decoder shown in FIG. 3b, the vector quantization decoder 12, the adder 3, and the block scanning frame memory 13 calculate S^ _f = S^ _f-1 + x _f + q ^f , and the vector S^ _f =[S^ ₁ , S^ ₂ , ..., S^ ₁₆ ] _f is obtained as the reproduced image signal 20 including quantization noise q ^f . This can be converted to standard raster scan through block/raster scan converter 14. Next, the principle of vector quantization applied to the interframe difference signal 17 will be briefly explained. Now, K information source input signal sequences are put together into an input vector X = [x ₁ , x ₂ , . . . , x _K ]. In this case, let the predetermined divisions of the K-dimensional _Euclidean signal space R ^K = ( X ∈ R ^K ) be R ₁ , R ₂ , . = [ y
₁ , y ₂ , ..., y _N ]. Representative point y _i = [y _i1 ,
y _i2 ..., y _iK ] is divided into the input vector _X included in R _i
The output vector corresponding to is called vector quantization. Vector quantization Q is defined by the following equation. Q: R ^K →Y Here, R _i =Q ^-1 ( y _i )={ X ∈R ^K :Q ( x )= y
_i } _N U ⁱ⁼¹ R _i =R ^K , R _i ∩R _j =0 (i≠j) The above vector quantization Q is expressed as a cascade connection of encoding C and decoding D. The encoding C is a set of output vectors in R ^K Y = [ y ₁ , y ₂ , ...,
y _N ) index set I={1, 2, ...,
N}, and the decoding D is mapping from I to Y
This is mapping to. That is, C: R ^K → I, D: I → Y Q=D・C. Since the encoded output I is transmitted or recorded during vector quantization, extremely efficient high-efficiency encoding can be realized. Vector quantization is particularly efficient when the amplitude probability distribution P( X ) of the input vector X is biased. The extraction of the set Y of output vectors consists of processing a large number of input vectors generated from a model of the input signal source based on the amplitude probability distribution P ⁽
Clustering may be performed to converge so that the sum of Min d( x , y _i ) is minimized. FIG. 4 shows the arrangement relationship of the input vector X and the output vector y _i on a two-dimensional plane. As can be inferred from Figure 4, in vector quantization, the distortion caused by Gaussian noise is averaged out in the K-dimensional signal space because it is mapped to the output vector that causes the minimum distortion in the K-dimensional signal space. There is. Furthermore, when an output vector close to the origin occurs in the K-dimensional signal space, it is possible to control the amount of information generated by masking it and making it a zero vector. FIGS. 5 and 6 show block diagrams of an encoder and a decoder which are an embodiment of the vector quantizer for interframe difference signals according to the present invention. In the vector quantization encoder shown in FIG. 5, 22 is an input vector register, 23 is an address counter, 24 is an output vector code table, 25 is an output vector register, 26 is a parallel subtracter, 27 is an absolute value calculator, 28 is a maximum element distortion detector, 29 is a minimum distortion output vector detector, 30 is an index latch, and 31 is a threshold circuit. In the vector quantization decoder shown in FIG. 6, 32 is an output vector shift register;
The same reference numerals as those in the figures indicate the same or equivalent parts. Next, the operation of the vector quantization encoder shown in FIG. 5 will be explained. First, the inter-frame difference signal 17 of the f-th frame
is grouped in units of 16 and becomes the input vector X _f =
[X ₁ , X ₂ , . . . , X ₁₆ ] is latched into the input vector register 22 as _f . At this point, the address counter 23 sequentially counts up to i=1, 2, ..., N, and outputs the output vector y _i from the output vector code table 24 as y ₁ , y ₂ , ...,
Read in the order of y _N , output vector register 25
are sequentially latched. The output vector code table 24 contains 16 vectors that result in the minimum distortion based on the amplitude probability distribution P( X _f ) of the input vector X _f in advance.
A set of output vectors in the dimensional signal space are written in order of proximity to the origin. That is, in the 16-dimensional signal space, the output _vectors are arranged at addresses i=1, 2 , . y _i = of each output vector register sequentially read out and latched in the output vector register 25
Regarding [y _i1 , y _i2 , ..., y _i16 ], the input vector X _f and the absolute value of the difference for each element are calculated by the parallel subtractor 26
is calculated by the absolute value calculator 27. The maximum value of each element distortion is determined by the maximum element distortion detector 28 as the distortion d( x _f , y _i ) between the input and output vectors. Here, d( x _f , y _i )=Max[|x _K −y _iK |]. (K=1, 2, . . . , 16) Next, the minimum distortion output vector detector 29 detects the output vector that minimizes the distortion d( x _f , y _i ). The minimum distortion D of the minimum distortion output vector and the input vector is D = Mind ( x _f , y _i ) = Min [Max | X _K − y _iK |]
becomes. This maximum distortion D is vector quantization noise. The minimum distortion output vector detector 29 detects the output vector with the minimum distortion between the input and output vectors and sends a strobe signal to the index latch 30. In the index latch 30, the output vector y _i that provides the minimum distortion for each input vector X _f1
The index i corresponding to the address of is latched. A threshold level is set for this index i, and if the threshold level reaches a predetermined level in the threshold circuit 31, the value of i is sent out, and if it does not exceed it, a special code is sent out as a zero vector.
The output of this threshold circuit 31 is sent as an output vector encoded signal 18 to a vector quantization decoder. In the vector quantization decoder shown in Fig. 6,
The output vector encoded signal 18 consisting of the index i of the output vector and a special code corresponding to the zero vector is decoded into a predetermined output vector.
Index latch 31 latches index i of the output vector and reads output vector y i corresponding to index _i from output vector code table 24 filled with the same set of output vectors as in the encoder of FIG. . This output vector y _i = [y _i1 , y _i2 , ..., y _i16 ] is latched in parallel to the output vector shift register 32, and then vector quantized frames are converted into vector quantization frames in the order of y _i1 , y _i2 , ..., y _i16 . The difference signal 19 is output in series. Also, when a special code is sent,
It is reset to the output vector shift register 32 and outputs zero in sequence 16 times as a zero vector. As described above, the threshold level of the threshold circuit 31 of the vector quantization encoder shown in FIG. If this is provided in the communication path, the amount of information generated can be controlled by controlling the threshold level. Furthermore, since vector quantization itself performs quantization in a multidimensional signal space, it is resistant to Gaussian noise, and has the effect of suppressing the amount of information generated by averaging the noise components superimposed on the interframe difference signals of individual samples. As described above, the vector quantization type interframe coding device according to the present invention vector quantizes a plurality of interframe difference signals at once, and suppresses the output vector close to the origin in a multidimensional signal space by setting a threshold level. With this configuration, a highly efficient interframe encoding device can be realized with a simple configuration. The device can be applied to high-efficiency encoding of commercial television, variable portion still image transmission, or conference television. Although the case where vector quantization is performed on 16 samples of the inter-frame difference signal at once has been described above, the method of configuring the samples to be blocked may be arbitrary. Furthermore, before the vector quantization encoder, the sample values of each block are normalized by the standard deviation or the median of the absolute values of the deviations, and after the vector quantization decoder, they are restored by multiplying by a normalization constant.
It is also possible to adaptively control the followability to the amplitude change of the image signal between frames. In this case, the set of output vectors consists of normalized output vectors. As described above, in the vector quantization type interframe coding device according to the present invention, a plurality of interframe difference signals of image signals are quantized in a multidimensional signal space, and a vector close to the origin is quantized to a threshold level. Since the configuration is configured such that the vector is set and controlled to be a zero vector, it is possible to realize an interframe encoding device that is simple and capable of transmitting product quality images with high efficiency.

[Brief explanation of the drawing]

Fig. 1 shows a conventional interframe encoding device, in which a is a block diagram of an encoder, b is a block diagram of a decoder, and Fig. 2 is a pixel arrangement between frames of a raster-scanned image signal. An explanatory diagram showing the relationship, and FIG. 3 is a block diagram showing an embodiment of the vector quantization type interframe encoding device according to the present invention, where a is a diagram showing an encoder and FIG. 3 is a diagram showing a decoder. , Fig. 4 is an explanatory diagram showing the relationship between input and output vectors in vector quantization, Fig. 5 is a block diagram showing an embodiment of a vector quantization encoder, and Fig. 6 is an illustration of an example of a vector quantization decoder. FIG. 2 is a configuration diagram showing an example. In the figure, 1 is a subtracter, 2 is a scalar quantizer, and 3
is an adder, 4 is a frame memory, 10 is a raster/block scan converter, 11 is a vector quantization encoder, 12 is a vector quantization decoder, 13
is a block scan frame memory, 14 is a block/raster scan converter, 22 is an input vector register, 23 is an address counter, 24 is an output vector code table, 25 is an output vector register, 26 is a parallel subtracter, and 27 is a parallel absolute value. 28 is a maximum element distortion detector, 29 is a minimum distortion output vector detector, 30 is an index latch, 3
1 is a threshold circuit, and 32 is an output vector shift register. Note that the same reference numerals in the figures indicate the same or equivalent parts.

Claims (1)

[Claims] 1. A frame memory that always stores at least one frame of an input image signal, and the latest image signal obtained by grouping the input signal series into K blocks (K is plural) is input. a subtracter that reads out a block-formed one-frame previous prediction signal at a position corresponding to the same position on the screen one frame before from the storage unit and calculates an inter-frame difference signal sequence; A set of output vectors selected in advance so as to have an optimal arrangement, that is, a predetermined number of representative points in a K-dimensional signal space R ^K that includes all K vectors, is set as an input vector. The input vector is encoded into the identification code of the output vector that minimizes the distortion (distance in the K-dimensional signal space) between the input and output vectors, and the minimum distortion output vector is closer to the origin than the set value in the K-dimensional signal space. If the distance is close to a vector quantization decoder that sequentially outputs blocks of output vectors as vector quantized inter-frame difference signal sequences; A vector comprising an adder that sequentially adds predicted signal sequences to reproduce the latest image signal including vector quantization noise, and outputs the reproduced image signal sequence so as to be written in the frame memory. Quantization interframe coding device. 2. During vector quantization, a set of output vectors is written into the output vector code table memory in order of proximity to the origin of the K-dimensional signal space ^RK , and the difference between each element of the output vector and input vector is a minimum distortion output vector calculation unit that determines an output vector with a minimum absolute value as a minimum distortion (shortest distance) output vector; and an address in the output vector code table memory of the minimum distortion output vector as an identification code of the minimum distortion output vector. and a vector quantization encoder that outputs an identification code of a zero vector as an encoded output when the address of the minimum distortion output vector is less than or equal to a predetermined value. The vector quantization interframe encoding device described above.