JP2960446B2 - Image processing method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing system for compressing, transmitting or recording a still image, and then expanding the image.

[Conventional technology]

“Baseline S” to standardize the natural image coding system
Various international standardization schemes such as "ystem" and "Extended System" have been proposed.

Fig. 5 shows the “Baseline System” of the international standard system
It is a schematic diagram showing the processing procedure of. This system divides an input image into blocks of 8 × 8 pixels, and performs discrete cosine transform (DCT) for each block.
mation) (process P1), and the obtained DCT coefficient is 8 × 8
The quantization is performed by dividing by each threshold value of the quantization matrix including the threshold values (process P2). Quantized DCT
The DC component of the coefficient takes the difference from the DC component quantized in the previous block, the number of bits of the difference is Huffman coded,
The AC component is converted into a one-dimensional sequence by performing a zigzag scan in the block, and then two-dimensional Huffman coding is performed on the number of consecutive zeros (ineffective coefficients) and the number of bits of the effective coefficients (processing P3 and P4). ).

The data thus compressed is decompressed by a process reverse to the processes P1 to P4. That is, Huffman decoding in process P4 ', decoding of DC and AC components in process P3', inverse quantization in process P2 ', and inverse DCT (IDCT) in process P1'.

[Problems to be solved by the invention]

By the way, in the above-described processing procedure, the value of the output image data is input due to a quantization error at the time when the inverse DDCT is performed, an arithmetic error resulting from a finite arithmetic word length at the time of DCT and inverse DCT operation, and the like. Overflow or underflow occurs beyond the range that can be represented by the number of bits of image data, and there is a disadvantage that a correct reproduced image cannot be obtained.

SUMMARY OF THE INVENTION It is an object of the present invention to provide an image processing method capable of preventing an overflow or an underflow caused by a quantization error or an arithmetic error.

[Means for solving the problem]

According to the image processing method of the present invention, an input image is represented by N × N
N × N obtained by performing discrete cosine transform on each block by dividing into a plurality of blocks
Data transformation step of performing quantization by dividing the number of transform coefficients by each threshold value of a quantization matrix composed of N × N threshold values, and multiplying the quantized transform coefficient by each threshold value of the quantization matrix. In an image processing method that includes a data decompression step of performing an inverse discrete cosine transform, a maximum pixel data MAX and a minimum pixel data MIN are obtained from pixel data of an image obtained by performing an inverse discrete cosine transform,
For the maximum value max and the minimum value min that can be represented by the number of bits representing each pixel data, the maximum value pixel data MAX
And the maximum value max, and normalize each pixel data of an image obtained by performing an inverse discrete cosine transform so that the minimum value pixel data MIN matches the minimum value min. The data is output as output image data.

[Action]

The image processing method according to the present invention performs quantization by normalizing each pixel data of an output image when image data compressed by processing such as discrete cosine transformation is reproduced by expansion processing such as inverse discrete cosine transformation. An overflow or an underflow caused by an error or a calculation error is prevented.

In this normalization process, the maximum pixel data MAX and the minimum pixel data MIN are obtained from the pixel data constituting each output image, and the maximum value represented by the number of bits representing the pixel data is obtained.
The calculation is performed by performing the following equation based on max and the minimum value min.

In this case, G (i, j) represents pixel data constituting each output image, i = 1,2, ..., m, j = 1,2, ..., n, and m is the number of horizontal pixels of the output image. , N respectively represent the number of vertical pixels of the output image.

Therefore, according to the present invention, the pixel data G (i, j)
Is converted into pixel data G ′ (i, j) that changes within the maximum range represented by the number of bits, and is output. Therefore, it is possible to prevent overflow and underflow due to the effects of quantization errors and arithmetic errors. Can be.

〔Example〕

FIG. 1 is a schematic diagram showing a processing procedure of an image processing system according to the present invention. The same parts as those in FIG.

First, the input image has N dots in the horizontal direction and N dots in the vertical direction.
The line is divided into N × N pixels, for example, a block composed of 8 × 8 pixels, and a discrete cosine transform (DC
T) is performed (process P1). The obtained DCT coefficient is divided by each threshold of a quantization matrix composed of 8 × 8 thresholds to perform quantization (process P2). FIG. 2 shows a quantization matrix for a luminance signal.

A difference between the DC component of the quantized DCT coefficient and the DC component quantized in the previous block is obtained, and the number of bits of the difference is Huffman-coded (processing P3 and P4). Further, the AC component is subjected to zigzag scanning within the block and converted into a one-dimensional sequence, and then two-dimensional Huffman coding is performed on the number of consecutive zeros (ineffective coefficients) and the number of effective coefficient bits (processing P3 and P4). FIG. 3 shows a table indicating the order of the zigzag scan.

At the time of the quantization process P2, each threshold value of the quantization matrix is multiplied by a certain coefficient (scale factor) and quantization is performed. The coefficient adjusts the image quality and compression ratio of the compressed image.

Also, Huffman coding does not use the quantized coefficient value itself for both the DC and AC components, and the number of bits required to represent the value is input to Huffman coding. Then, separately from the Huffman code, the value of the number of bits is added as additional information. For example, if the quantized coefficient is 2
(Decimal number), if expressed in binary number, "000… 01
However, the number of bits 2 required to represent the value is Huffman-encoded as a value representative of this value. Then, only two bits of data "10" are added as additional bits.

If the quantized coefficient is negative, data obtained by subtracting 1 from the additional bit is added. For example, when the quantized coefficient is -2 (decimal number), when expressed in a binary number (two's complement notation), it becomes "111... 110", and the lower two bits become additional bits. “01” obtained by subtracting “1” is added as an additional bit. By doing so, the additional bit starts with 1 when the quantized coefficient is positive, and starts with 0 when the quantized coefficient is negative.

The data thus compressed is decompressed by a process reverse to the processes P1 to P4. That is, Huffman decoding in process P4 ', decoding of DC and AC components in process P3', inverse quantization in process P2 ', and inverse DCT (IDCT) in process P1'.

Next, each pixel data of the image obtained by the inverse DCT is normalized in process P5. As shown in the flowchart of FIG. 4, this normalization is performed first by pixel data G (i, j) (i =
, M, j = 1, 2,..., N, m: the number of pixels in the horizontal direction, n: the number of pixels in the vertical direction) and the maximum value is determined as MAX (step S1). Is calculated as the minimum value and set to MIN (step S
2). The pixel data G (i, j) is corrected by using the following formula based on the maximum value MAX and the minimum value MIN thus obtained.

In this embodiment, since each pixel data of the input and output images is represented by 8 bits, the maximum value max is 255 and the minimum value is
min becomes 0. Accordingly, “max−min = 255”, and the following equation is actually calculated (step S3).

This calculation process is performed for all pixel data G (i, j) for each image. When the calculation for all pixel data is completed (step S4), normalization (process P5) is completed.

By this normalization processing, the change range of the value of each pixel data becomes 0 to 255, and output image data free from the influence of clipping due to overflow or underflow is obtained.

〔The invention's effect〕

According to the present invention, to perform normalization of the pixel data of the output image after the inverse discrete cosine transform at the time of data expansion,
The entire range of the converted image data can be made linear without causing clipping due to an overflow or underflow due to the influence of a quantization error or a calculation error, and a correct value can be output.

[Brief description of the drawings]

FIG. 1 is a diagram showing a processing procedure of an image processing method according to the present invention, FIG. 2 is a diagram showing a quantization matrix of a luminance signal, FIG. 3 is a diagram showing a zigzag scan table, and FIG. And FIG. 5 is a diagram showing a processing procedure of a conventional image processing method.

──────────────────────────────────────────────────続 き Continuation of front page (56) References JP-A-64-60175 (JP, A) JP-A-63-13485 (JP, A) JP-A-1-256817 (JP, A) JP-A 64-64 13817 (JP, A) JP-A-56-169778 (JP, A) (58) Fields investigated (Int. Cl. ⁶ , DB name) H04N 7/24-7/68 H04N 1/41-1/419 H04N 5/50-5/63

Claims (2)

(57) [Claims] An input image is divided into a plurality of blocks each composed of N × N pixels, discrete cosine transform is performed for each block, and N × N transform coefficients obtained by the conversion are represented by N × N transform coefficients.
A data compression step of performing quantization by dividing by each threshold of a quantization matrix composed of N thresholds, and an inverse discrete cosine transform after multiplying the quantized transform coefficient by each threshold of the quantization matrix. In the image processing method including a data decompression process to be performed, maximum pixel data MAX and minimum pixel data MIN are obtained from pixel data of an image obtained by the inverse discrete cosine transform, and a bit representing each pixel data is obtained. For the maximum value max and the minimum value min that can be represented by numbers, the maximum value pixel data MAX and the maximum value max match, and the minimum value pixel data MIN and the minimum value min match. Normalizing each pixel data of an image obtained by performing the inverse discrete cosine transform, and outputting the normalized data as output image data. Image processing method. 2. The normalization is performed on each pixel data G (i, j) of an image obtained by performing the inverse discrete cosine transform. The image processing method according to claim 1, wherein the following arithmetic processing is performed.