JPH06274609A - Compressing and extending method for color image - Google Patents

Abstract

PURPOSE:To improve the compressibility of limited color display by combining a Huffman encoding and predictive encoding for a limited color. CONSTITUTION:The number of limited colors to be calculated is designated (step 1) and the target number of limited colors are extracted by the color distribution of a color image (step 2). The frequency distribution of the number of pixels to be allocated from the color image to the respective limited colors is counted (step 3) and Huffman codes are allocated to the respective limited colors by calculating the probability of generation while using the frequency distribution (step 4). When the limited color of a pixel is same as the limited color of the adjacent pixel, the code of position relation is allocated, in the other case, the code of the limited color is allocated for each pixel, and they are preserved as compressed data (step 6). Finally, the additional information of compressed data is preserved (step 7).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for compressing and expanding a color image into a small amount of data by utilizing limited colors obtained from a digitized color image.

[0002]

Conventionally, many techniques have been proposed for image compression / decompression. Among them, the limited color display method is a commonly used method, in which a representative color is selected by statistical processing of the color distribution of the original image to describe and express the image. For example, when selecting a representative color from 16.7 million colors to 256 colors as a limited color, the number of bits required to represent one point of the image is reduced from 24 bits to 8 bits. The storage capacity required for the storage is reduced to 1/3, resulting in compression of the color image. Several methods have been proposed as statistical processing of the color distribution in the limited color display method, but the Median-Cut algorithm is basically used and often used.

The median cut algorithm is described in the literature ("Color image quantizatio").
n for frame buffer display
y "-Graphics, 16, 3, pp. 297-3
07, 1982), a color cube corresponding to a three-dimensional color space of red R, green G, and blue B is quantized into 5 bits (32) for each of the three color axes, 32 × 32 × 32 = divided into 32768 color cubes,
The frequency (which becomes a three-dimensional histogram) in which each pixel value of the target image enters each solid is created. Here, the color cube is regarded as one box, a one-dimensional histogram is created along the longest direction of the box, and a dividing plane (median) whose frequency is halved on both sides is obtained to obtain two boxes. To divide.

This division process is repeated until the number of boxes reaches a target number, and the colors in each box are averaged to make N representative colors, and the representative color of the color having the smallest distance value is assigned to 32768 colors. . Each pixel is converted to any of 32768 colors, each pixel is represented by the obtained representative color number i, and a color map having a representative color (Ri, Gi, Bi) at the address of the representative color number i is generated. To do. Further, the color map is read out by the number of each pixel, and the image is reproduced by displaying the color map with the representative color.

Various proposals for improving this method have been proposed recently, and as an example, reference can be made to JP-A-2-146685 and a limited color image generating method.

FIG. 4 is a block diagram showing a system configuration to which a conventional limited color image generation method is applied. An image processing device 50 for performing limited color image generation processing, a memory 51 for storing image data and the like, an image data source. 52, a display device 53 for displaying an image and an input device 54. Image data source 5
Reference numeral 2 denotes a video camera, VTR, or other image signal generator, which converts the image signal into digital image data and supplies it to the image processing apparatus 50. Limited color compression is performed in the image processing apparatus 50. However, since there is a problem that a pseudo contour is likely to occur in the part where the color changes gently like human skin according to the limited color display, the area where the pseudo contour is likely to occur is emphasized. This is a modification in which a limited color is specified preferentially and as much as possible is assigned to an area, and a more detailed display is performed.

On the other hand, there is Huffman coding as a conventional character data compression method. Huffman coding for character data means to make a "leaf" corresponding to each character, and the occurrence probability of that character in each "leaf". , A new “node” for the two “leafs” with the smallest occurrence probability.
Make one and connect the node and two "leafs" with "branches",
One of these two "branches" is labeled "0" and the other is labeled "1". Also, the sum of the probabilities of "leafs" is attached to "nodes", and this "node" is considered as a new "leaf". These are repeated until the number of "leaf" becomes one, and the sequence of "1" and "0" obtained by tracing "branch" from "root" to "leaf" corresponding to the character is the code word for the character. Is.

These will be described in detail with reference to codeword generation examples in the conventional Huffman coding shown in FIGS. The examples shown in FIGS. 5 to 10 are A, B, C, D,
The six types of letters E and F are 1/3, 1/6, and
It is a codeword generation example of Huffman coding when the occurrence probabilities of 1/6, 1/6, 1/12, and 1/12 occur. First, "leafs" A to F corresponding to the letters A to F are created, and the occurrence probabilities (1/3 to 1/12 etc.) of each "leaf" are written (Fig. 5). Of these, the two with the smallest occurrence probability are "leafs", which are the letters "E" and "F", and "nodes" G are created from those "leaves" and the label "1" is given to each "branch". "" And "0" are given. The occurrence probability of this "node" G is "leaf"
Is 1/6, which is the sum of the occurrence probabilities of (FIG. 6).

Considering the "node" G as a "leaf", a "node" H is created from the two "leafs" D and G having the smallest occurrence probability of the "leaf" set "A, B, C, D, G". , The labels “1” and “0” are given to each “branch”, and the occurrence probability ⅓ of the “node” H is obtained from the sum of the occurrence probabilities of the “leafs” D and G (FIG. 7). Similarly, the “node” H is considered as a “leaf”, and the set of “leafs” “A,
Two "leafs" with the smallest occurrence probability in "B, C, H"
Make "node" I from B and C, and label "1" on each "branch"
And "0" are given, and the occurrence probability 1/3 of the "node" I is obtained from the sum of the occurrence probabilities of these "leafs" B and C (FIG. 8). This “node” I is considered as a “leaf”, and a set of “leafs” “A, I,
The "node" J is created from the two "leafs" I and H that have the smallest occurrence probability of "H", and the labels "1" and "0" are given to each "branch" to determine the occurrence probability of "leaf" I and H. The probability 2/3 of occurrence of “node” J is calculated from the sum (FIG. 9).

When this "node" J is a "leaf", only two "leafs" remain in the "leaf" set "A, J". "leaf"
Create "node" K from A and J, and label "1" on each "branch"
And "0" are given. This "node" K is a "root" whose occurrence probability is 1 (Fig. 10). As a result, the generated codewords by Huffman coding are "0", "100", "10".
These are 1 ”,“ 110 ”,“ 1110 ”, and“ 1111 ”.

Further, as a method of compressing image data, there is a method called predictive compression which uses the property that there is little difference from adjacent pixels. FIG. 11 is a diagram of a conventional prediction compression example, and there are some prediction methods such as full-value prediction, 1-line prediction, average prediction, and plane prediction for the pixel Xj, i in the previous line and the current line. Those prediction error values are roughly quantized where the value is large and the quantization level is reduced to reduce the code length required to express the error value, resulting in compression. Alternatively, a value that appears many times based on the frequency of the prediction error value is represented by a short code length, and as a result, it is compressed.

[0012]

However, in the conventional image compression method using the limited color display as described above, 24 bits are reduced to 8 bits in spite of little deterioration of the image quality, so the reduction rate ( There is a problem that the compression rate is low.

In view of the above problems, it is an object of the present invention to provide a color image compression / decompression method capable of improving the compression ratio of limited color display by a simple process and reducing the file size required for image storage. I am trying.

[0014]

A color image compression / decompression method of the present invention extracts a target number of limited colors from a color distribution of a color image and determines the number of pixels assigned to each limited color from the color image. Counting the frequency distribution, assigning a Huffman code to each limited color and the positional relationship based on the frequency of the limited color and the positional relationship between the adjacent pixels, and each pixel, if the limited color of the pixel is the same as the limited color of the adjacent pixel. For example, it is characterized in that the Huffman code obtained by encoding the positional relationship is encoded by the Huffman code obtained by encoding the limited color if any of the adjacent pixels is different.

Further, the Huffman code is detected from the compressed data compressed by the above-mentioned compression method, and in the case of the Huffman code in which the adjacency relationship is encoded, the corresponding original adjacent position data is obtained, and in other cases, the original Huffman code is obtained. It is characterized in that the compressed data is expanded by converting and decoding it into limited color data and outputting R, G, and B values representing the color of each pixel.

[0016]

According to the above configuration, when each pixel is encoded,
When the limited color of the pixel is the same as the limited color of the adjacent pixel, it is encoded by the Huffman code that encodes the positional relationship, and when it is different from any of the adjacent pixels, the Huffman code that encodes the limited color by frequency is used. Encode. When decompressing or playing back an image, it is converted / decoded by the procedure reverse to the encoding,
Since the R, G, and B values are output, the compression rate can be improved by simple processing.

[0017]

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of a system to which the color image compression / decompression method of the present invention is applied. FIG. 2 is a diagram showing a procedure for creating a color image compression / decompression method according to the present invention.

The CPU 10 for performing the compression code processing of the image system shown in FIG. 1 has a memory 20 for storing various data including image data, an input device 30 for inputting commands and text data, and an image. A display device 40 is coupled to display data including the data on the display.
The CPU 10 reads a color image to be processed from the memory 20, creates a compression code, converts the read color image into a code, and returns the code to the memory 20.

When a color image is displayed on the display device 40, the display command and the image name of the display target are input from the input device 30, and the CPU 10 reads the image stored in the memory 20 in the form of a code. The color image is expanded by the procedure opposite to the compression, and the expanded color image is transferred to the display device 40 and displayed.

Next, the processing operation will be described.

There is a bias in the frequency of use of the limited colors used to represent the original color image, and by using this bias in the frequency of use, the higher the frequency of use, the shorter the code becomes by Huffman coding. If it is realized, the average bit length expressing the limited color can be shortened, and as a result, the compression rate can be increased. Furthermore, since each pixel of an image is often the same as an adjacent pixel, the compression rate can be further increased by combining the concept of predictive coding, which is the same as which pixel is coded.

The procedure for creating a compressed code will be described below with reference to FIG. First, the number of limited colors to be obtained is designated (step 1). The specified number of limited colors are extracted from the color image to be processed by the limited color extraction method described in the related art (step 2). Subsequently, a limited color corresponding to the color value of each pixel is obtained, and a frequency distribution having a frequency of the number of pixels included in each limited color is created (step 3). The occurrence probability of each limited color is obtained from the frequency distribution, and the Huffman code is assigned to each limited color by the Huffman coding method described in the prior art (step 4). Step 4 is repeated for all pixels of the color image to be processed (step 5).

Next, as shown in the coding example in the predictive compression of FIG. 3, in the case of the coding of the positional relationship between adjacent pixels, if the leading bit is "0", it means that the code is of the positional relationship. In this case, the second bit encodes whether the pixel is the same as the pixel in the previous line or the same as the pixel immediately before. In FIG. 3, for the pixel Xj, i, the pixel on the previous line is Xj-i,
i, the immediately preceding pixel is Xj, i−1. If the first bit is "1", the subsequent bit string is the code of the limited color, so that the code of the positional relationship and the code of the limited color can be treated together.

According to the above encoding process, the positional relationship code and the limited color Huffman code are assigned to each pixel and stored as compressed data in the memory 20 (step 6).
Then, the image size, the code of the positional relationship, the Huffman code of each limited color, and the color values of red R, green G, and blue B are stored as additional information of the compressed data (step 7).

The above is the color image compression process. In the case of the decompression process, the CPU 10 reads the compressed data stored in the memory 20, detects the Huffman code from the compressed data, and the detected Huffman code is the first bit. "0"
In the case of the code having the positional relationship of, the corresponding position data is converted / decoded. In the case of the limited color Huffman code with the first bit being “1”, it is decoded into the corresponding limited color data, and the R, G, B values representing the color of each pixel are output. -It is to be regenerated.

Up to now, the limited color has been obtained by dividing the color cubes of red R, green G, and blue B, but the compression effect can be obtained even if a uniform color space matching the human color sense is used as the color cube. It goes without saying that it will be done.

As described above, in this embodiment, the average bit length for expressing the limited color is shortened by the Huffman coding utilizing the bias in the frequency of use of the limited color, and the position of the limited color by the code and the predictive coding. Since the bit strings of the related codes are distinguished and handled together, the synergistic effect of compression is enhanced as compared with the conventional example.

[0028]

As described above, according to the present invention,
The effect of achieving a higher compression rate than the limited color display, because it is compressed by a combination of the encoding for obtaining the Huffman code based on the occurrence probability of the limited color obtained from the image to be processed and the encoding for predictive compression. There is.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an image system to which a color image compression / decompression method of the present invention is applied.

FIG. 2 is a diagram showing a creation procedure of a color image compression / decompression method of the present invention.

FIG. 3 is a diagram showing a coding example of predictive compression according to the present invention.

FIG. 4 is a system configuration diagram to which a conventional limited color image generation method is applied.

FIG. 5 is an explanatory diagram of “leaf” in codeword generation in conventional Huffman coding.

FIG. 6 is an explanatory diagram of a “node” G created from the “leaf” shown in FIG. 5;

FIG. 7 is an explanatory diagram of “node” H created from “leaf” shown in FIG. 6;

8 is an explanatory diagram of a "node" I created from the "leaf" shown in FIG.

9 is an explanatory diagram of a "node" J created from the "leaf" shown in FIG.

10 is an explanatory diagram of a "node" K created from the "leaf" shown in FIG.

FIG. 11 is a diagram showing a conventional prediction compression example.

[Explanation of symbols]

 10 CPU 20 Memory 30 Input Device 40 Display Device

Claims (2)

[Claims] 1. A desired number of limited colors are extracted from a color distribution of a color image, and a frequency distribution of the number of pixels assigned to each limited color is counted from the color image, and the frequency of the limited colors and the positions of adjacent pixels are counted. A Huffman code is assigned to each limited color and a positional relationship from the relationship, and each pixel is a Huffman code that encodes the positional relationship if the limited color of the pixel is the same as the limited color of an adjacent pixel. A method for compressing a color image, characterized in that, if different, the limited color is encoded by a Huffman code. 2. A Huffman code is detected from the compressed data compressed by the color image compression method according to claim 1, and in the case of a Huffman code encoding an adjacency relationship, the corresponding original adjacent position data Otherwise, the compressed image data is expanded by converting and decoding the original limited color data and outputting the R, G, and B values representing the color of each pixel.