JPH06178281A - Bit error detection method - Google Patents

Abstract

PURPOSE:To simplify the bit error detection of compression data by counting the number of EOB codes for each line of macro block layer and checking the relation in the block number and macro block line number. CONSTITUTION:Two-dimension quantization coefficient data being a quantization output of an orthogonal transformation coefficient are converted into a Huffman code while being subjected to zigzag scanning from a low-order (low frequency component) to a high-order (high frequency component) at a variable length coding section 5. In this case, compression picture data in use adopt a hierarchical structure and header attribute information representing the order of macro block line is given to each line of macro block layer and one EOB (end of block) code is inserted without fail to each block independently of whether or not a final quantization coefficient of zigzag scanning is zero. Then an error check section 26 of a compression picture reproduction device detects the EOB code to count the number of EOBs in one line of macro block.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a bit error detecting method for easily detecting a bit error when reproducing compressed image data by high efficiency coding.

[0002]

2. Description of the Related Art Conventionally, there has been known an image compression coding system in which compressed image data has a hierarchical structure composed of four layers. The four layers are a block, a macroblock, a line of macroblock and a picture. Each block contains compressed image data, and has a hierarchical structure that takes into account two-dimensional orthogonal transformation in block units.

[0003]

However, in such a conventional image compression coding method, since compressed image data is contained in each block, if an error occurs even in one bit, this error will occur in the subsequent blocks. However, it is difficult to detect the bit error.

An object of the present invention is to solve the problem that the conventional image compression encoding system is difficult to detect bit errors as described above, and to easily reproduce bit errors when reproducing compressed image data using a hierarchical structure. An object of the present invention is to provide a bit error detection method capable of detecting the error.

[0005]

In order to achieve the above object, the method of the present invention comprises an image data preprocessing step of dividing an image signal into a plurality of blocks, and a variable length coding of each block for each block. To the variable length coding step for obtaining the compressed image data of the hierarchical structure by always adding the EOB code to the
It is characterized by including a compressed data reproducing step of counting the number of codes and judging a bit error according to whether or not the number of the EOB becomes a predetermined number.

Further, in the present invention, preferably, the compressed data comprises layers of a block, a macroblock, a line of macroblock and a picture, and in the compressed data reproducing step, a line of macroblock layer is formed. It can be characterized in that a bit error of the compressed image data is detected depending on whether or not the relationship between the line number of the macroblock in the header attribute information and the number of counted EOB codes is correct.

[0007]

According to the present invention, the EOB code is always added to each block of the compressed image data having the hierarchical structure, and the number of EOB codes is counted for each line of macroblock layer in the reproducing system to determine the number of blocks and the macro. Bit errors are detected by checking the relationship between the number of block lines. Therefore, in the present invention, although the compression rate is slightly lowered, the accuracy of error detection is improved and the image quality is improved as a result.

[0008]

Embodiments of the present invention will now be described in detail with reference to the drawings.

FIG. 1 shows an example of a hierarchical structure of compressed image data used in an embodiment of the present invention. The compressed image data has a hierarchical structure composed of a block (B), a macroblock (MB), a line of macroblock (LMB) and a picture (P). One block is composed of 4 pixels × 4 pixels, one macroblock is composed of 6 blocks containing luminance and color difference information, one line of macroblock is composed of 31 macroblocks, and one picture is composed of 26 macroblocks. It consists of a line of macroblocks of lines. As a result, one picture has 4836 blocks and represents one screen of 208 pixels × 248 pixels.

One header attribute information as shown in FIG. 2 is attached to each picture layer. In FIG. 2, PSC is a picture start code, and P
TYPE is a type information, QBS is a quantized index bit shift, and various factor values, Y / C quantized coefficient table values, etc. are entered in the attribute information.

Further, one header attribute information as shown in FIG. 3 is attached to each line of macroblock layer. In FIG. 3, LMBSC is a line-of-macroblock start code, and LMBN is a line-of-macroblock number containing 1 to 26 indicating which macroblock line. There is no header attribute information in the macroblock layer and block layer.

FIG. 4 shows the functional blocks of the image compression algorithm executed by the image compression encoder according to the embodiment of the present invention. The contents of FIG. 4 will be described along the flow of processing. Generally, the correlation between the pixels of the image is relatively high up to several pixels in the vicinity. Therefore, in order to improve the compression effect, first, in the image data preprocessing unit 1, one screen of the input digital image data
The 12 × 480 pixels are digitized to 248 × 208 pixels and are divided into blocks to obtain image data having a hierarchical structure as shown in FIG.

Next, in the orthogonal transform coding unit 2, 4 ×
Two-dimensional Hadamard transform or two-dimensional discrete cosine transform (D) with coordinate axis rotation for each 4-pixel two-dimensional image block
CT) and the like are converted into the frequency domain, and the conversion coefficients of the respective frequency components that are arranged in order from the highest power and are expressed by linear combination are encoded. At that time, in order to realize high compression coding, high frequency components having small power components are deleted.

Next, in the intra-frame predictive coding unit 3, intra-frame predictive coding of DC components of Y (luminance), Cb (blue color difference), and Cr (red color difference) is performed on the orthogonal transform output. I do. That is, the data of the current pixel of interest is predicted from the data of the previous macroblock, and only the difference between the predicted value and the value of the input signal is encoded. In other words, the inter-pixel distance is coded with a pixel that is out of prediction as a significant pixel. At this time, the data of the head macroblock of the previous screen is distributed to the first macroblock of one screen.

Next, in the quantizer 4, in order to match with the visual system, the predictive coding output is quantized with a different quantization width for each frequency component (non-linear quantization). That is, an adaptive bit that has a small energy component and is visually insensitive and has a high frequency is reduced by increasing the step width, and a portion having a visually sharp sensitivity and a low frequency is reduced by decreasing the step width. Make an allocation.

Next, in the variable length coding unit 5, the two-dimensional quantized coefficient data of the block, which is the quantized output of the orthogonal transform coefficient, is zigzagged from the low order (low frequency component) to the high order (high frequency component). Convert to Huffman code while scanning.

That is, generally, as the order is increased, the number of terms which are considered to be "0" in the converted components increases. This leads to efficient transmission. Therefore, the encoding is performed according to the scanning order, and the last coefficient at which the quantized coefficient is non- “0” is the object of encoding, and the EOB (End of Block) code indicating the end of the abort scanning is added.

The one-dimensional data obtained by the above-mentioned zigzag scanning includes some continuous "0" s and non-zero "0" s, and the continuous number (run length) of "0s" and "0" s. Two-dimensional Huffman coding that generates one code is performed on a two-dimensional code (symbol) that is combined with non-data (called a significant conversion coefficient). In Huffman coding, shorter symbols are given to symbols with higher occurrence probability, which minimizes the average code length and improves transmission efficiency. Since the amount of data after Huffman coding varies depending on the design, it is stored once in the buffer memory and then read at a constant speed.

Finally, the video frame encoding unit 6 adds the header attribute information as shown in FIG. 2 and FIG. 3 to format the frame structure.

The compressed image data thus data-compressed is recorded on a recording medium such as a floppy disk, or parallel-serial converted and transmitted to a distant device via a transmission path.

FIG. 5 is a block diagram of the variable length coding unit 5 described above.
The procedure of converting the data of the quantized coefficient of the block into the Huffman code is shown by using a specific numerical example. First, as shown by the arrow in FIG. 5A, the quantized coefficient QINDX in the 4 × 4 block is zigzag-scanned.
The result of scanning the quantized coefficient is shown in FIG. Then, as shown in FIG. 5C, by inserting EOB immediately after the final non-zero coefficient of the scan, the transmission efficiency of the subsequent zero coefficient is improved. The quantization coefficient is (D) in FIG.
A two-dimensional code (R, L) in which a run length R of zero coefficient and a level L of non-zero coefficient following it are combined is used as shown in FIG. R of the two-dimensional code (R, L) represents the number of zero components preceding the non-zero component by an unsigned integer of 4 bits (bits), and L represents the level of the non-zero component by a signed integer of 8 bits. Furthermore, Huffman coding is applied to the event of the two-dimensional code. During Huffman coding, EOB is also counted as one of the occurrences, and a specific bit code is assigned to it. Maximum Huffman code length is 14
up to bits.

FIG. 6 shows the result of the zigzag scan described above.
The handling according to the present invention when the last quantized coefficient is a significant conversion coefficient when "0" is shown. The result of the zigzag scanning of the quantized coefficient of the block as shown in FIG. 6A is as shown in FIG. In this way, when the last quantization coefficient is not "0", the EOB code is not added in the prior art as shown in FIGS. 7 (A) and 7 (B). In the present invention, since this EOB is used as a check code for bit error detection at the time of reproducing image data as described later, as shown in FIGS. 6C and 6D, the last quantization coefficient is “0”. If not, add EOB. That is, in the present invention, EOB (end
Of block) code is inserted.

Next, the reproduction processing of the compressed image data in the embodiment of the present invention will be described with reference to FIG. The above-mentioned compressed data recorded or transmitted on the recording medium is temporarily stored in a buffer memory (not shown) and then supplied to the compressed image reproducing device shown in FIG. In this compressed image reproducing device, the variable length code decoding unit 11, the dequantization unit 12, the prediction code decoding unit 13 and the inverse Hadamard transform unit 14 are used in FIG.
The image compression encoder of FIG. First, the variable length code decoding unit 11 converts the Huffman code into one block of quantized coefficient data by performing an operation opposite to that of the variable length coding process as described later.
Next, the inverse quantization unit 12 transforms the data of the quantization coefficient QINDX according to the value of the quantization coefficient QINDX of the input data to realize the non-linear quantization. Next, the predictive code decoding unit 13 obtains data in a state before predictive coding by adding the input data and the predicted value. Next, the inverse Hadamard transform unit 14 performs a 4 × 4 two-dimensional inverse Hadamard transform on the 1,0-bit Hadamard index from the inverse quantization unit 12 by pipeline processing to obtain actual image data.

FIG. 9 shows a detailed configuration of the variable length code decoding unit 11 of FIG. First, the header attribute information extraction unit 21
Image coded data input in (compressed image data)
The following header attribute information in is extracted and a predetermined process is performed. That is, a 19-bit PSC (picture start code) in the header attribute information of the picture layer is detected, and the octet counter of the picture header is initialized. Also, Y, C factors, Y, C quantization table information are extracted from the picture attribute information, and the picture attribute information is output at the timing of a predetermined timing chart in synchronization with the data output of the quantization coefficient QINDX. . 19-bit LMBSC of header attribute information of line of macroblock layer
(Line of Macroblock Start Code)
And LMBN (line of macroblock number) are detected, and the variable length code decoding unit (VLCD unit) 11 is initialized for each macroblock line.

Decoding of the Huffman code into a two-dimensional code is performed by drawing the Huffman ROM table inside the Huffman code decoding unit 22 at a cycle of 34.92 ns. First, by operating only the Huffman code decoding unit 22 for 2k cycles without moving the two-dimensional code decoding unit 24, 128 words × 12 can be obtained in any combination of Huffman codes.
A 12-bit two-dimensional code (see FIG. 5D or FIG. 6D) is stacked until the bit FIFO (fast-in / fast-out memory) 23 is full. When the FIFO 23 becomes full, the operation of the Huffman code decoding unit 22 is temporarily stopped, the two-dimensional code decoding unit 24 is started to operate in a burst cycle of 139.68 ns, and one screen of data is processed without stopping thereafter. Keep doing The two-dimensional encoding / decoding unit 24 operates, and the FIFO
When 23 becomes available, the Huffman code decoding unit 22 continues to operate until the FIFO 23 becomes full again.

In the two-dimensional code decoding unit 24, the two-dimensional code (R, L) in which the run length of the zero coefficient: R and the level of the following non-zero coefficient: L are combined is used to scan the quantized coefficient (see FIG. (B) of FIG. 6 and (B) of FIG. 6).

The output of the two-dimensional code decoding unit 24 is supplied to the intra-block scan order conversion unit 25. This conversion unit 25
Then, the zigzag scan order in the block is returned to the scan order for each row as shown in FIG. 10, and the scan order in each row direction and column direction is adapted to the inverse Hadamard transform unit 14 in a high-speed pipeline operation. The scan order is converted into a scan order for and output as a quantized coefficient QINDX.

Part of the header attribute information detected by the header attribute information extracting section 21 is also supplied to the error detecting section 26. The error detection unit 26 detects an error in the compressed image data by constantly monitoring the relationship between the number of LMBN (line of macroblock number) and EOB (end of block) in the data.

That is, as described above, the compressed image data used in the present invention has a hierarchical structure, and the number of macroblock lines is inserted as header attribute information for each line of macroblock layer. For each block, as shown in FIGS. 5 and 6, there is always one EO regardless of whether or not the last quantization coefficient of the zigzag scan is zero.
B code is inserted. The number of blocks in one line of macroblock is constant (31 × 6 in this embodiment).
Since one EOB code is always inserted in each block as described above, the number of EOB codes existing in each line of macroblock is constant. However, when a bit error occurs due to transmission, the data shifts and the relationship between the number of EOB codes and LMBN changes.

Therefore, the error detecting section 2 of the compressed image reproducing device
6 detects this EOB code and counts how many EOBs are present in one line of macroblock. For example, in the case of this embodiment, 31 × 6 EO
A line of macroblock header comes for each B code, and the error detection unit 26 determines whether the relationship with the macroblock line number (LMBN) in the header attribute information is correct. Detects bit errors.

In this way, when a bit error is detected, the error detection section 26 sends an error detection signal to that effect to a control section (not shown). According to the error detection signal, the control unit performs a predetermined control operation such as discarding the corresponding screen, receiving the compressed data of the same screen again, or stopping the compression reproduction process.

The present invention is not limited to the image compression / decoding processing method described in the above embodiment, but various hierarchical structures,
It can be applied to orthogonal transform method, predictive coding method, quantization method, and variable length coding method.

[0033]

As described above, according to the present invention,
An EOB code is always added to each block of the compressed image data having a hierarchical structure, the number of EOB codes is counted for each line of macroblock layer in the reproduction system, and the relationship between the number of blocks and the number of macroblock lines is calculated. Since bit error detection is performed by checking, it is possible to easily detect bit errors in compressed data with simple hardware, and the compression rate is slightly reduced, but error detection accuracy is low. Is improved, resulting in a better image quality.

[Brief description of drawings]

FIG. 1 is a conceptual diagram showing an example of a hierarchical structure of compressed image data used in an embodiment of the present invention.

2 is a format diagram showing a configuration of header attribute information attached to each one picture layer of FIG. 1. FIG.

FIG. 3 shows one line-of-macroblock of FIG.
It is a format figure which shows the structure of the header attribute information added for every layer.

FIG. 4 is a block diagram showing functional blocks of an image compression algorithm executed by the image compression encoder according to the embodiment of the present invention.

5 is an explanatory diagram showing a conversion procedure of an embodiment of the present invention when converting one block of quantized coefficient data into a Huffman code in the variable length coding unit of FIG.

FIG. 6 is an explanatory diagram showing a conversion procedure of the embodiment of the present invention when the final quantized coefficient is not zero.

FIG. 7 is an explanatory diagram showing a conversion procedure in a conventional example provided for comparison with FIG.

FIG. 8 is a block diagram showing a configuration of a compressed image reproducing device according to an embodiment of the present invention.

9 is a block diagram showing a detailed internal configuration of a variable length code decoding unit in FIG. 8. FIG.

FIG. 10 is an explanatory diagram showing a scan order conversion operation in the intra-block scan order conversion unit of FIG.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Image data preprocessing unit 2 Orthogonal transformation unit 3 Intra-frame prediction coding unit 4 Quantization unit 5 Variable length coding unit 6 Video frame coding unit 11 Variable length code decoding unit 12 Inverse quantization unit 13 Prediction code decoding unit 14 Inverse Hadamard transform unit 21 Header attribute information extraction unit 22 Huffman code decoding unit 23 FIFO memory 24 Two-dimensional code decoding unit 25 In-block scan order conversion unit 26 Error detection unit LMB line of macroblock LMBN line of macroblock Number EOB End of Block

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁵ Identification code Office reference number FI technical display location H04N 7/137 A

Claims (2)

[Claims] 1. An image data preprocessing step of dividing an image signal into a plurality of blocks, and an EOB code is added to each block without fail when variable length encoding of the block is performed to obtain compressed image data having a hierarchical structure. A long encoding step and a compressed data reproducing step of counting the number of the EOB codes when reproducing the compressed image data and judging a bit error according to whether the number of the EOB is a predetermined number or not. A method for detecting a bit error, comprising: 2. The compressed data comprises layers of blocks, macroblocks, line-of-macroblocks and picture layers, and in the compressed-data reproducing step, macros in header attribute information of the line-of-macroblock layer. 2. The bit error detection method according to claim 1, wherein the bit error of the compressed image data is detected depending on whether or not the relationship between the line number of the block and the number of counted EOB codes is correct.