JPH05130418A - Picture processor - Google Patents

Abstract

PURPOSE:To obtain the picture processor preventing deterioration in picture quality and realizing the control of compressed data quantity. CONSTITUTION:The processor is provided with a conversion means (DCT section 3) applying frequency conversion to input picture data, a 1st quantization means (quantization section A) quantizing a transformation coefficient transformed by a transformation means, a 1st coding means (Huffman coding section A) coding the transformation coefficient quantized by the 1st quantization means and a 2nd coding means (Huffman coding section B) coding a quantization error by the 1st quantization means.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing apparatus for effectively compressing input image data.

[0002]

2. Description of the Related Art A so-called ADCT (Adaptive Discrete Cosine Transform) compression method has been proposed as a compression technique for multi-valued images, which mainly targets natural images. This compression method is 3
The primary color (RGB) signal is converted into three components of Y, U, and V, and the luminance Y signal has the same resolution as it is, and the chromaticity component U and V signals have the resolution reduced by subsampling in some cases. And compress. In the compression step 1, the DCT is performed for each block of 8 × 8 pixels for each component.
And converted into an 8 × 8 frequency space to obtain DCT coefficients. In step 2, a luminance component (Y) and a chromaticity component (U, V), two types of quantization tables are prepared, and DC
The T coefficient is linearly quantized (divided) by an 8 × 8 quantized value obtained by multiplying each element of the 8 × 8 quantization table by a quantization factor for each component to obtain a quantized coefficient. In step 3, this quantized coefficient is Huffman-encoded using the Huffman encoding method which is a variable length encoding method.

[0003]

However, when the compression is performed by the conventional compression method, the restored image is deteriorated due to the quantization error. In particular, when compression is performed on an image in which a natural image, a character image, a CG image, and the like are mixed, the quantization error of an image portion that is not suitable for the quantization table used becomes large, and the restored image of that portion deteriorates. It had the drawback of becoming large.

Further, since the Huffman coding system is a variable length coding system, it has a drawback that it is difficult to compress it to a target data amount.

Therefore, it is an object of the present invention to provide an image processing apparatus which prevents deterioration of the image quality of a restored image and realizes control of the amount of compressed data.

[0006]

In order to solve the above-mentioned problems, the image processing apparatus of the present invention quantizes conversion means for frequency-converting input image data and the conversion coefficient converted by the conversion means. A first quantizing means; a first coding means for coding the transform coefficient quantized by the first quantizing means; and a first coding means for coding a quantization error by the first quantizing means. 2 encoding means.

[0007]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The following embodiments of the present invention are intended to prevent the deterioration of image quality by gradually reducing the quantization error due to the quantization by providing a multi-step quantization part and a Huffman coding part. Is. Further, by operating the compressed data at each stage, it is possible to realize fixed length compression that makes the amount of compressed data constant.

FIG. 1 is a diagram showing a specific configuration for implementing the present invention. Input RGB image data by the following 3 × 3 linear matrix conversion,

[0009]

[Outer 1] Conversion to YUV is performed using the color conversion unit 1 indicated by. Y
Represents a luminance component, and U and V represent chromaticity components. Taking advantage of the sensitivity characteristic of human eyes that the luminance component (Y) is more sensitive than the chromaticity component (U, V), the sub-sampling unit 2 performs sub-sampling, and Y:
U: V = 4: 1: 1 or 4: 2: 2. Next, in the DCT unit 3, Y, U,
V, DCT (discrete cosine transform) is performed for each 8 × 8 block, and frequency conversion is performed for each 8 × 8 block. This converted coefficient is hereinafter referred to as a DCT coefficient.

The DCT coefficient is written in the temporary buffer 4. The data is quantized in multiple stages as shown below.

In the first stage, the quantizing unit A5 reads data for each color component (Y, U, V) from the buffer 4 for each block, and quantizes the data using the quantization table A8. The result is called a quantized coefficient, is encoded by the Huffman encoding unit A11, and is transferred to the address decoding unit 14. The quantization error generated at that time, that is, the residual data A is sent to the quantization unit B6.

Next, in the second stage, the quantizer B6 quantizes the transmitted residual data A using the quantization table B9. The quantized coefficient B is encoded by the Huffman encoding unit B12 and transferred to the address decoding unit 14. Further, the quantization error generated at that time, that is, the residual data B is sent to the quantization unit C7.

Further, in the third stage, the quantizer C7 converts the residual data B sent thereto into a quantization table C10.
To quantize. The quantized coefficient C is encoded by the Huffman encoding unit C13 and transferred to the address decoding unit 14.

The address decoding unit 14 writes the sent Huffman code data in the compression memory 15 at each stage, and writes information indicating which stage of the code data is in the segment information buffer 16.

In the conventional method, since the quantization is performed in only one step, the deterioration of the information amount due to the quantization error is large. However, as described above, by performing the quantization in multiple steps, two steps, three Since the deleted information amount is saved at the stage, it is possible to prevent the deterioration of the image quality significantly.

For the quantization in each step, for example, in the first step, a quantization table of coefficients that saves the information amount in the low frequency space is used, and in the second step, the information amount in the intermediate frequency space is used. By using a coefficient quantization table for storing the above, and a coefficient quantization table for storing the information amount of the high frequency space in the third step, the character of the quantization at each step can be changed. .. Therefore, it is possible to apply variously to the image mainly including the space below the intermediate frequency depending on the purpose such as using up to the second stage.

Furthermore, by changing the quantization table at each stage, the amount of compressed data at each stage can be changed, and thereby fixed length compression can be realized.
For example, the compressed data amount of the first, second and third stages is set to 3: 2: 1.
If the total compressed data amount is smaller than the target compressed data amount, all stages of compressed data are used, and the total compressed data amount is 10% of the target compressed data amount. If it exceeds, the target compressed data amount can be realized by deleting the compressed data of the third stage.

In this embodiment, common encoding means is used for each of the Y, U and V components. By doing so, the common circuit portion can be omitted and the configuration can be simplified.

Regarding decompression of compressed data, the Huffman coding units A, B and C are Huffman decoding units A, B and C, and the quantizing units A, B and C are inverse quantizing units A, B, C and quantum. The quantization tables A8, B9, and C10 are inverse quantization tables A and B, respectively.
B and C, and the DCT part becomes an inverse DCT part.

The address decoding unit 14 transfers each coded data from the compression memory 15 to the Huffman decoding units A, B and C according to the segment information buffer 16.
The Huffman decoding units A, B and C respectively decode the coded data and transfer them to the dequantization units A, B and C. Inverse quantizer A,
In B and C, the obtained quantized coefficients are dequantized and transferred to the adder 17. The adder 17 adds the results of the inverse quantizers A, B, and C to each color component for each pixel, and transfers the result to the buffer 4. The inverse DCT unit reads data from the buffer and performs inverse DCT conversion. Sub-sampling unit 2
Then, when the sub-sampling ratio is 4: 1: 1, Y
1, Y2, Y3, Y4, U1, V1 ... to Y1, Y2, Y
3, Y4, U1, U1, U1, U1, V1, V1, V
1, V1, ..., and when 4: 2: 2, Y1, Y2, U
, 1, V1, ... Are converted to Y1, Y2, U1, U1, V1, V1. The color conversion unit 1 may perform the inverse conversion of Expression (1.0).

(Other Embodiments) FIG. 2 is a diagram showing a modification of the above-mentioned embodiment. Components having the same functions as those in FIG. 1 are designated by the same reference numerals, and the description thereof will be omitted. In FIG. 2, by providing the remainder buffer 20, multi-stage quantization can be performed by the quantization tables A8, B9, C10, one quantizing unit 18 and the Huffman coding unit 19, and the multi-stage quantizing of FIG. The circuit configuration of the quantization / encoding unit can be simplified. The process will be described below. In the first stage, the quantizing unit 18 quantizes using the quantization table A8, and the Huffman coding unit 19 quantizes the result.
Huffman coding. The quantization error at that time, that is, the remainder data is written in the remainder buffer 20. In the second stage, the remainder buffer 20 in the first stage is set to the quantization table B.
9, the quantizer 18 quantizes the result, and the Huffman encoder 19 Huffman-encodes the result. Further, the quantization error at that time, that is, the residual data is written in the residual buffer 20 again. Thereafter, similar processing is performed to realize multi-step quantization.

To decompress compressed data, the quantizer 18 is used as an inverse quantizer, the quantization tables A8, B9, and C10 are used as inverse quantization tables A, B, and C, and the Huffman encoder 19 is used for Huffman decoding. Part. The dequantization unit selects the dequantization tables A, B, and C for each stage of the quantized coefficients obtained from the Huffman decoding unit, dequantizes them, and transfers the result to the adder 17.

In the embodiment and the other embodiments described above, only three stages, A, B and C, are mentioned as the quantization stages.
This may be performed by adding D, E, F ...

Also, instead of the buffer 4, a double buffer as shown in FIG.
The efficiency of DCT and quantization can be improved by alternately using.

[0025]

As described above, according to the present invention, it is possible to prevent the deterioration of the image quality when the image is compressed, and it is possible to realize the control of the compressed data amount.

[Brief description of drawings]

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

FIG. 2 is a block diagram showing another embodiment of the present invention.

FIG. 3 is a diagram showing a double buffer used in another embodiment of the present invention.

[Explanation of symbols]

 1 Color Converting Section 2 Sub-Sampling Section 3 DCT Section 4 Buffer 5 Quantization Section A 6 Quantization Section B 7 Quantization Section C 8 Quantization Table A 9 Quantization Table B 10 Quantization Table C 11 Huffman Code Section A 12 Huffman Code part B 13 Huffman code part C 14 Address decoding part 15 Compression memory 16 Segment information buffer 17 Adder 18 Quantization part 19 Huffman code part 20 Residual buffer

Claims (7)

[Claims] 1. A transforming means for frequency-converting input image data, a first quantizing means for quantizing a transform coefficient transformed by the transforming means, and a transform quantized by the first quantizing means. First encoding means for encoding the coefficient, and second encoding means for encoding the quantization error by the first quantizing means.
An image processing device, comprising: 2. A second quantizing means for quantizing the quantization error by the first quantizing means, wherein the second coding means is quantized by the second quantizing means. The image processing apparatus according to claim 1, wherein the result is encoded. 3. The image processing apparatus according to claim 2, wherein the second quantization means quantizes the residual data obtained by the first quantization means. 4. The image processing apparatus according to claim 2, wherein the first and second quantization tables are different from each other. 5. The image processing apparatus according to claim 1, wherein the coded data by the first and second coding means is operated according to a target data amount. 6. The image processing apparatus according to claim 1, further comprising means for temporarily storing the residual data by the first quantizing means. 7. The image processing apparatus according to claim 1, further comprising a plurality of means for temporarily storing the spatial frequency component.