JPH0116074B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a vector quantizer that blocks and quantizes three channels of color image signals.

Conventionally, when quantizing color image signals, red,
Green and blue 3-channel image signals (hereinafter R,
G, B) were quantized in parallel for each channel, one sample at a time. FIG. 1 shows the configuration of a conventional scalar quantizer for color image signals.

In the figure, 1 is a red image input signal sequence, 2 is a green image input signal sequence, 3 is a blue image input signal sequence, 4 is a scalar quantizer, 5 is a red image output signal sequence, 6 is a green image output signal sequence, 7 is a blue image output signal sequence.

The scalar quantizer 4 scalar quantizes the R, G, and B input signals in parallel for each channel in units of one sample. In this case, a uniform quantization characteristic as shown in FIG. 2 is used based on the color balance of the three channels.

However, the signals of the R, G, and B channels have a strong correlation between adjacent samples on the screen of the same channel or samples corresponding to the same pixel of mutual channels, and the scalar quantizer shown in Fig. 1 has an extremely large quantization loss. .

This invention was made in order to eliminate these drawbacks, and enables highly efficient encoding by vector quantizing a plurality of R, G, and B channel signal sequences as one block. The object of the present invention is to provide a vector quantizer for color image signals.

First, the principle of vector quantization will be explained. A plurality of samples of the input signal sequence are collectively made into a block, and this is converted into an input vector X={x ₁ , x ₂ ,
...x _K }. Let N partitions of the dimensional signal space S ^K containing all input vectors be S ₁ , S ₂ , ..., respectively.
Let it be S _N. Set Y of output vectors yi which are representative points (e.g. center of gravity) of each division = {y ₁ , y ₂ , ..., y _N }
Then, vector quantization Q is defined as a mapping of S ^K to Y by the following equation.

Q: S ^K →Y Here, Si=Q ^-1 (yi) = {XER ^K :Q(x)=yi} _N ∪ ⁱ⁼¹ Si=S ^K , Si∩Sj=0 (i≠j) Vector Quantization Q can be defined as a continuation of encoding C and decoding D. If encoding C is a mapping of S ^K to an index set J={1, 2, . . . N}, then decoding D is a mapping of J to Y. That is, C: S ^K → J D: J → Y Q=D・C.

Now, we collect M input signal sequences of R, G, and B obtained by scanning the same pixel on the screen in parallel in the color image signal, and calculate R={R ₁ , R ₂ ,
..., R _M }, G={G ₁ , G ₂ ,...G _M }, B={B ₁ , B ₂ ,
...B _M }. Further, with this as one block, the input vector X={R ₁ , R ₂ ,..., R _M , G ₁ , G ₂ ,...,
G _M , B ₁ , B ₂ ,...B _M }={x ₁ , x ₂ ,..., x _K }. Here, K=3.M. Based on the amplitude distribution of this input vector X, a set Y of output vectors yi is determined so as to achieve the minimum distortion. This can be done by using a model of the amplitude probability density of the color image signal to determine the division and set of representative points (for example, center of gravity) that result in the minimum distortion by clustering. At this time, vector quantization is a conversion to an output vector that is closest to the input vector X (resulting in minimum distortion).

FIG. 3 shows the relationship between input vectors and output vectors in the case of a three-dimensional signal space of R ₁ , G ₁ , and B ₁ . 3
When there is a correlation between two signals, the signals are distributed around the principal axis of R ₁ =G ₁ =B ₁ , so the effect of vector quantization in which the output vectors are concentrated and arranged near the principal axis is large.

FIG. 4 shows a block diagram of an encoder which is an embodiment of the vector quantizer for color image signals according to the present invention, and FIG. 5 shows a block diagram of a decoder.

In the figure, 1, 2, and 3 are red, green, and blue image input signal sequences, respectively, 8 is an input vector register, 9 is a code table address counter, and 10
is an output vector code table memory, 11 is an output vector register, 12 is a parallel subtracter, 13 is a parallel absolute value calculator, 14 is a maximum element distortion detector, 1
5 is a minimum distortion output vector detector, 16 is an index strobe signal, 17 is an index signal,
18 is an index latch, 19 is an encoded output signal, 20 is an output vector decomposition register, 21, 2
2 and 23 are red, green and blue image output vectors, respectively.

Next, the operation of the encoder shown in FIG. 4 will be explained. Each R, G, and B image input signal sequence is processed in parallel by M correlated samples, and the input vector X={R ₁ , R ₂ ,...R _M , G ₁ , G ₂ ,
...G _M , B ₁ , B ₂ , ... B _M }={x ₁ , x ₂ , ... x _K } and taken into the input vector register 8. At this point, the code table address counter 9
sequentially counts up to i=1, 2, . . . N, and reads out the output vector set Y from the output vector code table memory 10. Input vector X latched in input vector register 8 and output vector sequentially read out in output vector register 11
yi is compared for each element, and the element distortion D _il = |y _il −x _l | (here, l
=1, 2,...,K) is calculated. The maximum elemental distortion detector 14 detects the maximum elemental distortion Di=Max lD _il between each element. Next, the minimum distortion output vector detector 15 detects an output vector from the set Y of output vectors that gives the input vector X the minimum maximum element distortion Di. That is, the output vector with the minimum distortion D is detected in minimax approximation. Here, D= Min i [ Max l | y _il −x _l |]. When the minimum distortion output vector detector 15 detects the output vector resulting in the minimum distortion, it sends an index strobe signal 16 to the index latch 18. At this time, code table address counter 9
The index signal 17 of the minimum distortion output vector from
is taken into the index latch 18 and output as an encoded output signal 19. This encoded output signal 19 is transmitted or recorded and converted into an output vector by the decoder shown in FIG.

Next, the operation of the decoder shown in FIG. 5 will be explained. The encoded output signal 19 is first taken into the index latch 18. The index latch 18 uses this as the output vector code table address in the output vector code table memory 1.
Read the corresponding output vector from 0. The minimum distortion output vector read out in the above procedure is passed through the output vector register 11 to the output vector decomposition register 20, where the red image output vector 21,
It is decomposed into a green image output vector 22 and a blue image output vector 23. The output vector of each channel becomes a vector quantized output signal sequence by decomposing blocking.

The coding efficiency η of the above vector quantization is as follows: Number of samples to be blocked = 3M, number of output vectors N
= 2 ^t , then η = t/K = t/3M.

As described above, in the vector quantizer for color image signals according to the present invention, the image signals of three channels of R, G, and B are grouped into blocks and vector quantized by a plurality of samples, thereby realizing extremely highly efficient encoding. can.

The above description deals with R, G, and B as the signal components of the three channels of the color image signal, but NTSC
Y, I, Q signal or Y, Y handled by the method etc.
Samples of different ratios (for example, 12:4:4) of the R, Y, and B signals may be collectively blockified and vector quantized.

Furthermore, although minimax approximation is used to calculate the minimum distortion, it is of course possible to use the Euclidean norm or the sum of absolute values.

As described above, in the vector quantizer for color image signals according to the present invention, the signal components of three channels are grouped into blocks and vector quantized by a plurality of samples, so that extremely high efficiency encoding can be realized.

[Brief explanation of drawings]

Fig. 1 is a block diagram of a conventional scalar quantizer for color image signals, Fig. 2 is an explanatory diagram of the quantization characteristics of the conventional scalar quantizer, and Fig. 3 is an input diagram of color image vector quantization according to the present invention. An explanatory diagram showing the relationship between output vectors, FIG. 4 is a block diagram of an embodiment of an encoder of a vector quantizer for color image signals according to the present invention, and FIG. 5 shows vectors for color image signals according to the present invention. FIG. 2 is a configuration diagram of an embodiment of a decoder of a quantizer. In the figure, 1 is a red image input signal sequence, 2 is a green image input signal sequence, 3 is a blue image input signal sequence, 4 is a scalar quantizer, 5 is a red image output signal sequence, 6 is a green image output signal sequence, 7 is a blue image output signal series, 8
is an input vector register, 9 is a code table address counter, 10 is an output vector code table, 11 is an output vector register, 12 is a parallel subtracter, 13 is a parallel absolute value calculator, 14 is a maximum element distortion detector, 15 is a minimum distortion output vector detector,
16 is an index strobe signal, 17 is an index signal, 18 is an index latch, 19
is an encoded output signal, 20 is an output vector decomposition register, 21 is a red image output vector, 22 is a green image output vector, and 23 is a blue image output vector. In addition, in the figures, the same reference numerals indicate the same or corresponding parts.

Claims (1)

[Claims] 1 A total of K color image signals consisting of a signal sequence of 3 channels (K
is an input vector register that blocks the samples of a plurality of input vectors to form an input ^vector an output vector code table memory, a minimax calculator for detecting an output vector that produces a minimum distortion in minimax approximation with the input vector among a set of output vectors sequentially read from the output vector code table memory; an encoder that uses the output vector code table address of an output vector as an encoded output signal; a decoder that reads a corresponding output vector from an output vector code memory for decoding using the encoded output signal as an address signal; A vector quantizer for color image signals, comprising an output vector decomposition register that decomposes an output vector obtained from a decoder into a three-channel output signal sequence.