JPH07154606A - Picture encoder - Google Patents

Abstract

PURPOSE:To improve coding efficiency by providing a Markov model control section receiving an edge detection signal and a reference picture element pattern so as to provide an output of a line buffer initializing signal and a select signal of a context generating section to the coder. CONSTITUTION:Coded picture element data of one block are coded by raster scanning. When a coded picture element is accumulated in a serial line buffer 1, a coding object picture element is inputted to a light end of a block of the line buffer 1, a surrounding picture element of the coded picture element is selected from the reference picture element and an output of the reference picture element becomes a context CX by a context generating section 2. A Markov model control section 3 gives a mode signal 108 to an edge detection section 4 by the setting of a block width setting signal BW 104. The edge detection section 4 counts coded picture elements to detect left end and right end picture elements of the block. The Markov model control section 3 receives an edge detection signal 109 to control a select signal 103 of the context generating section 2 thereby executing the edge processing.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to the structure of an image coding apparatus.

[0002]

2. Description of the Related Art Conventional image encoding devices for binary images and color images (ISO, binary encoding standard, JBIG, or image data encoding system, JP-A-63-16457).
5) uses the configuration shown in the block diagram of FIG. 11
Is a line buffer for accumulating the encoded pixel data P (121) already encoded, and 12 is a context generation unit for inputting the pixel data (122) accumulated in the line buffer and outputting a context CX (125). Yes,
An edge detection unit 13 detects that the coded pixel is located at the left end or the right end of the block in which the block width setting signal BW (123) is set, and outputs the left and right edge detection signals (124) to the contest generation unit. is there. In general, the context is given by the reference pixels selected from the peripheral pixels of the coded pixel P, but when the coded pixel P is located at both edges of the block, the reference pixel is out of the block. Since the pixels around the outside of the block are encoded as white pixels, the protruding reference pixels generate a context as white pixels. 14 is the context (125)
Of the transition state signal (128) of the probability estimation table that is input as the address and is output from the arithmetic code calculation unit (16) and a dominant symbol (129) indicating a symbol with a high occurrence probability.
Is a coding state storage unit for storing the probability estimation table R storing the probability estimation value of the pixel for the transition state.
An OM unit 16 is an arithmetic code calculation unit that inputs the coded pixel P (121), the dominant symbol (130), and the probability estimation value (127) from the probability estimation table ROM unit and outputs the code word C.

When the coded pixel data P is given, the context CX is generated by the already coded coded pixels P using the peripheral pixels of the current coded pixel as reference pixels.
From this context CX (125), an estimated value of the probability of occurrence of a dominant symbol having a high probability of occurrence as a coded pixel and a dominant symbol (inferior symbol: a symbol having a low probability of occurring in the context) is obtained, and the arithmetic code operation unit A codeword is output by inputting the coded pixel P (121), the dominant symbol (130), and the probability estimation value (127). The estimated value of the occurrence probability of the dominant symbol or the inferior symbol is stored in the probability estimation table ROM unit (15), and the occurrence probability of the encoded pixel with respect to the context is estimated and output. At this time, the coding state storage unit including a memory such as a RAM does not store the probability of occurrence of symbols, but stores the transition state of the probability estimation table. Therefore, the transition state of the coding state storage unit is input as an address, and the output value of the probability estimation table ROM unit becomes the symbol occurrence probability. Further, since the probability estimation table stores the transition destinations to the next states when the dominant symbol appears and the inferior symbol appears for the coded pixel, the transition state of the coding state storage unit is stored. The occurrence probability is updated by updating. The dominant symbol occurrence probability transitions to a higher occurrence probability when the dominant symbols continuously appear, and transitions to a lower occurrence probability when the inferior symbol appears. As described above, since the image coding apparatus shown in FIG. 2 has the learning effect of updating the occurrence probability by the input coded pixels, it is necessary to prescan the input image in advance to check the occurrence probability of the image. Without this, high efficiency coding can be realized for various images.

[0004]

However, in the above-mentioned conventional technique, the process of encoding an image of A4 size 200 dpi as one block is used as a contrast, and therefore one line is divided into one line.
It is necessary to have several 728-bit line buffers. When a Markov model with 10-bit reference pixels is formed, the number of states in the coding state storage unit (14) is 2 to the 10th power. Therefore, increasing the size of one block increases the hardware amount of the line buffer, and increasing the reference pixels of the Markov model increases the hardware amount of the coding state storage unit.

Further, the image is 32 × 32 bits, 64 × 6.
When divided into small blocks such as 4 bits for encoding,
When a Markov model with 10-bit reference pixels is formed,
The decoding of one block is completed before the symbol occurrence probability is updated by the input data and the learning effect is obtained, and improvement of the compression rate due to the learning effect cannot be expected.

Therefore, the present invention provides a device that solves the problems of the prior art, reduces the hardware scale, and improves the coding efficiency.

[0007]

An image coding apparatus according to the present invention receives a coded pixel data P as input, stores a line buffer 1 for storing the pixel data P, and a pixel data 102 stored in the line buffer 1. A context generation unit 2 that inputs and outputs a context of a Markov model by a select signal 103 from a Markov model control unit 3 in the subsequent stage, a block width setting signal 104, a reference pixel number setting signal 105, a reference pattern setting signal 106, and an edge processing setting. Signal 10
7 is set and the edge detection signal 109 and the reference pixel pattern 111 are input to initialize the line buffer initialization signal 112.
And the Markov model control unit 3 that outputs the select signal 103 of the context generation unit, and the block width mode signal 108 from the Markov model control unit is switched to detect that the coded pixels are located at the left end and the right end of the block and detect the edge. An edge detector 4 that outputs a signal 109, a reference pixel pattern table 5 that stores a reference pixel pattern 111, outputs a reference pixel pattern 111 in response to a reference pixel pattern select signal 110 from the Markov model controller 3, and a context generator. The context of the Markov model from 2 is input as an address, and the transition state 117 of the probability estimation table from the arithmetic code calculator 119 in the subsequent stage is input.
And a coding state storage unit 6 that stores the dominant symbols 115 and 118, and a probability estimation table ROM that inputs the transition state 114 of the probability estimation table from the coding state storage unit 6 as an address and outputs the symbol probability estimation value 116. Part 7
And the arithmetic code operation unit 8 which inputs the coded pixel data 101, the dominant symbol 114 from the coding state storage unit, and the probability estimation value 116 from the probability estimation table ROM unit and outputs the codeword 119. Is characterized by.

[0008]

1 is a block diagram showing the structure of an image coding apparatus according to the present invention. 1 is the encoded pixel data P (101)
Is a line buffer that accumulates 1 bit at a time in a serial format, and 2 is the accumulated pixel data (10
2) is input, and a Markov model context CX (1
13) is a context generation unit that outputs 13).

Reference numeral 3 denotes a block width setting signal BW (104), a reference pixel number setting signal RN (105), a reference pattern setting signal RP (106) and an edge processing setting signal EC (10).
7) is set, and the Markov model control unit outputs the initialization signal (112) of the line buffer and the select signal (103) of the context generation unit by inputting the edge detection signal (109) and the reference pixel pattern (111). Four
Is an edge detection unit that detects that the coded pixel P is located at the left end and the right end of the block by switching the block width mode signal (108) from the Markov model control unit (3), and outputs an edge detection signal (109). is there.

A reference pixel pattern select signal (11) from the Markov model control unit 5 stores the reference pixel pattern.
0) is a reference pixel pattern table for outputting the reference pixel pattern (111), and 6 is the context CX of the Markov model from the context generation unit (2) as an address and the arithmetic code operation unit (8) outputs It is a coding state storage unit that stores the transition state ST (117) and the dominant symbol MS (118) of the probability estimation table.

Reference numeral 7 is a probability estimation table ROM portion which inputs the transition state (114) of the probability estimation table from the coding state storage portion (6) as an address and outputs a symbol probability estimation value (116), and 8 is a probability estimation table ROM portion. Encoded pixel data P (1
01) and the dominant symbol (11
5) and the probability estimation table ROM probability value (1
16) is an arithmetic code calculator that inputs 16) and outputs a code word C (119).

FIG. 3 is a diagram showing a configuration example of a line buffer in which reference pixels are 6 pixels for a block having a block size of 32 bits in width. The encoded pixel data of one block is encoded by raster scanning from the upper left as shown in FIG. 3, for example. As a feature of image data, the correlation between the coded pixel and its peripheral pixels is high. FIG. 3 shows an example in which 6 peripheral pixels of the coded pixel P are selected as reference pixels from the already coded pixels. When the coded pixel P is serially stored in the line buffer, the coded contrast pixel P is input to the bit position 65 of the line buffer, and the reference pixels 1, 32, 33, 34, 63, 64 are selected.
The output of the reference pixel R (1, 32, 33, 34, 63, 64) becomes the context CX of the context generation unit (2). The Markov model control unit (3) gives a mode signal (108) to the edge detection unit (4) by setting the block width setting signal BW (104). The edge detection unit (4) counts the number of encoded pixels and detects the left end pixel and the right end pixel of the block. The Markov model control unit (3) executes the edge processing by inputting the edge detection signal (109) and controlling the select signal (103) of the context generation unit.

A specific example of the edge processing will be described with reference to FIG. When the encoded pixel P is located in the line buffer 65, the reference pixels 32 and 6 are used when encoding the leftmost bit.
Since 3 and 64 are located outside the block,
The reference pixels 32, 63, and 64 are fixed to the white pixel (0) or the black pixel (1) regardless of the actual data, and the context is generated. Since the reference pixel 63 is located outside the block when the leftmost second pixel is encoded, the context is generated by fixing the reference pixel 63 to the white pixel or the black pixel.
Since the reference pixel 34 is located outside the block when the rightmost pixel is encoded, the context is generated by fixing it to the white pixel or the black pixel. At this time, the edge processing setting signal EC (107) is used to select whether the reference pixel outside the block is fixed to a white pixel or a black pixel. For an image with a large proportion of white pixels such as a document image, the compression ratio is improved by fixing the pixels around the block to white pixels. In general, the property of the image differs depending on the cut-out block, and therefore by setting the peripheral pixels for each block, the compression rate is improved as a whole.
Further, when the first line and the second line of the block are encoded, the reference pixels 1, 32, 33 and 34 are located outside the block with respect to the encoded pixel 65 shown in FIG. At this time, by resetting the line buffer to set all the bits to 0, the upper pixels outside the block are set to white pixels. On the other hand, by setting the line buffer and setting all the bits to 1, the upper pixels outside the block are set to black pixels. Initialization signal (112)
Thus, the white pixel and the black pixel described above are set.

FIG. 4 is a diagram showing a configuration example of a line buffer having 6 pixels as reference pixels in a block having a width of 128 bits. The encoded pixel P is input to the bit position 257 of the line buffer, and the reference pixels are 1, 128, 12
9, 130, 255 and 256 are selected. Reference pixel R
Is the context CX of the context generator (2)
Becomes 32-bit width context CX and width 12
Switching the 8-bit context CX selects the output of the bit corresponding to the reference pixel from the output of the line buffer. The output of the line buffer is controlled by the block width setting signal BW, the reference pixel number setting signal RN, and the reference pattern setting signal RP which are inputs to the Markov model control unit to generate a context. At this time, by storing the reference pixel pattern to be selected in advance in the reference pixel pattern table, the desired context can be changed by changing the block width setting, the reference pixel number setting, and the reference pattern setting according to the block to be compared. Can be generated.

Regarding the edge processing described with reference to FIG. 3, the block width setting signal BW (104) and the edge processing setting signal EC (1
According to the setting of 07), the Markov model control unit (3) performs edge processing.

FIG. 5 is a diagram showing an example of the configuration of a line buffer having 6 pixels as reference pixels in a block having a width of 128 bits. The encoded pixel P is input to the bit position 131 of the line buffer, and the reference pixels are 1, 2, 3, 4, 1
29 and 130 are selected. The output of the reference pixel R becomes the context CX of the context generation unit (2). 5 is a modification of the method of selecting reference pixels from FIG.
By selecting the reference pixel as shown in FIG. 4, the line buffer for one line can be reduced as compared with FIG. However, the relationship between the selection method of reference pixels and the compression rate of pixel data differs depending on the characteristics of the image, but when the number of reference pixels is the same, pixels closer to the coded pixel, such as the reference pixels shown in FIG. The compression ratio is improved by selecting as.

FIG. 6 is a diagram showing a configuration example of a line buffer in which a block having a width of 32 bits has four reference pixels. The encoded pixel P is input to the bit position 34 of the line buffer, and reference pixels 1, 2, 3, and 33 are selected. The output of the reference pixel R is the context generation unit (2)
It becomes the context CX of. 6 has a width of 32 bits and is the same as the block size of FIG. 3, but the number of reference pixels is reduced. Since increasing the number of reference pixels increases the number of context states, the characteristics of the image can be finely classified, and the compression rate can be expected to improve. The arithmetic code operation unit (8) according to the configuration of the present invention is an encoding state storage unit (6).
And the probability estimation table is encoded by the probability estimation value output from the ROM section (7), the error between the probability estimation value (116) and the occurrence probability value of the actual pixel with respect to the context CX defined by the reference pixel is small. Highly efficient coding can be realized. The probability estimation value (116) has a learning effect of updating the probability estimation value while encoding without pre-scanning the image, and outputs the probability estimation value adaptively. Therefore, the learning effect works more slowly as the number of states in the context increases, and the learning effect progresses more rapidly as the number of states decreases. Therefore, the smaller the number of reference pixels, the faster the learning effect progresses, and the larger the number of reference pixels, the finer the classification of context states, and the more accurately the characteristics of the image can be expressed.

In the present invention, the block width setting signal BW (1
04), a desired reference pattern is selected from the reference pattern table (5) by the reference pixel number setting signal RN (105) and the reference pattern setting signal RP (106), and arithmetic coding is performed. Therefore, as the line buffer (1), a line buffer having a bit number capable of realizing a reference pixel pattern having a maximum block size as a reference is prepared. Further, the coding state storage unit (6) also has a code state storage unit for storing a power-of-two state of the reference pixel number to be a reference, and the configurations of the line buffer and the coding state storage unit are the reference pixel pattern table. Use it by adaptively switching according to the setting.

[0019]

According to the present invention, a block size suitable for processing can be set for an input image, and the optimum reference pixel number and reference pattern for the block size can be easily selected without changing the hardware. It became possible.

Even when the image is divided into small blocks and encoded, the compression rate does not deteriorate, and the hardware scale can be reduced by dividing into small blocks and encoding.

[Brief description of drawings]

FIG. 1 is a block diagram showing a basic configuration of an embodiment of the present invention.

FIG. 2 is a block diagram showing the configuration of a conventional image encoding device.

FIG. 3 is a diagram showing a relationship between a reference pixel and a line buffer of a block having a width of 32 bits in the embodiment of FIG.

FIG. 4 is a diagram showing a relationship between a reference pixel and a line buffer of a block having a width of 128 bits in the embodiment of FIG.

5 is a diagram showing a relationship between a reference pixel and a line buffer of a block having a width of 128 bits in the embodiment of FIG.

FIG. 6 is a diagram showing a relationship between a reference pixel and a line buffer of a block having a width of 32 bits in the embodiment of FIG.

[Explanation of symbols]

1, 11 ... Line buffer 2, 12 ... Context generation unit 3 ... Markov model control unit 4, 13 ... Edge detection unit 5 ... Reference pixel pattern table 6, 14 ... Encoding State storage unit 7, 15 ... Probability estimation table ROM unit 8, 16 ... Arithmetic code calculation unit 101, 121 ... Encoded pixel data 102, 122 ... Accumulated pixel data 103 ... Select signal 104 , 123 ... Block width setting signal 105 ... Reference pixel number setting signal 106 ... Reference pattern setting signal 107 ... Edge processing setting signal 108 ... Block width mode signal 109, 124 ... Edge detection Signal 110 ... Reference pixel pattern selection signal 111 ... Reference pattern 112 ... Initialization signal 113, 125 ... Context signal 114,117,126,128 ... probability estimation table transition state signal 115,118,129,130 ... MPS 116,127 ... probability estimates 119,131 ... codeword

Claims (1)

[Claims] 1. A line buffer for inputting encoded pixel data to store the pixel data, and inputting pixel data accumulated in the line buffer to select the context of the Markov model by a select signal from a Markov model control unit at a subsequent stage. The output context generator, the block width setting signal, the reference pixel number setting signal, the reference pattern setting signal and the edge processing setting signal are set, and the edge detection signal and the reference pixel pattern are input, and the line buffer initialization signal and context are generated. Model control unit that outputs the select signal of the block, and the edge detection that outputs the edge detection signal by detecting that the coded pixel is located at the left end and the right end of the block by switching the block width mode signal from the Markov model control unit. And the Markov model control unit that stores the reference pixel pattern With the reference pixel pattern select signal of the reference pixel pattern select signal, the context state of the Markov model from the context generator is input as an address, and the transition state and the dominant symbol of the probability estimation table from the arithmetic code calculator are input. An encoding state storage unit that stores the following, a probability estimation table ROM unit that inputs the transition state of the probability estimation table from the encoding state storage unit as an address, and outputs a probability estimation value of a symbol, encoded pixel data, and encoding An image coding apparatus comprising an arithmetic code calculation unit for inputting a dominant symbol from a state storage unit and a probability estimation value from a probability estimation table ROM unit and outputting a code word.