JP2844749B2 - Image processing method using texture features - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Summary] Regarding an efficient compression / decompression method of digital image data, an image is divided into texture regions having almost uniform texture features, and only the contour information of the divided texture regions and the texture features are provided. An image input unit that converts an image into a digital image and stores it in an image processing method that divides the image into regions with uniform characteristics and processes the image, with the aim of reducing the amount of data by having Area division that divides the input image into multiple areas by integrating adjacent pixels of the same texture using texture features that represent the density of each pixel in the input image and the density correlation coefficient between each pixel, etc. A texture extracting unit that extracts a texture feature in the divided texture region; a contour extracting unit that extracts a contour of the divided texture region; An image data creation unit is provided that creates image data by associating the data obtained from the contour extraction unit with the data obtained from the texture feature extraction unit.

[Industrial applications]

The present invention relates to an efficient compression / decompression method for digital image data.

In recent years, there has been a demand for a database of images, and accordingly, a technology for efficiently compressing images has been required. As a conventional method of creating a database of images, there has been a method of restoring a high-quality restored image that is not so different from the original image. This is represented by compression and decompression of medical images and the like, and Huffman coding, transform coding, and the like have been proposed.

However, the positional relationship of the original images only needs to be accurate, such as when debugging the movement of the NES character or displaying map information with pictures. There was a demand and demand that it would not be necessary to restore it.

[Conventional technology]

In the conventional image compression method, the main focus has been on restoring the same image as the original image. Hereinafter, transform coding, which is a typical conventional compression technique, will be described.

FIG. 7 shows an outline of the transform coding. In the figure, 71 is an input image. Reference numeral 72 denotes a divided image, which is obtained by dividing the input image 71 into blocks. Reference numeral 73 denotes conversion data, which is data obtained by orthogonally converting the image of each block.

In the configuration shown in the above figure, first, the original image is divided into a plurality of small rectangular areas. This rectangular area is the minimum unit for performing compression. The reason why the image is divided into small regions is to capture a local change in the image that is blurred when taken in a large region. Next, orthogonal transformation is performed on each rectangular area. The orthogonal transform is to perform a discrete cosine transform on the density value of each pixel, and the discrete cosine transform is a function of frequency. The absolute value is called energy.
Since an image has a high correlation between adjacent pixels, energy is concentrated on low-frequency components by orthogonal transformation.

FIG. 8 shows a state in which energy is concentrated on low frequency components by orthogonal transformation. In the figure, the energy is concentrated in the parts shown in dark colors. 8-bit density level of original image
Assuming that there are 256 gradations, the values of each frequency component are compared after performing orthogonal transformation.
It takes a large value such as 250 to 250, and a small value such as -1 to 1 in a portion where energy is not collected.
In order to quantize and represent these values, it is not necessary to prepare the original 8 bits. Bits may be allocated according to the magnitude of the energy, such as 8 bits in a portion where energy is collected and 1 bit in a portion where energy is not collected.

 FIG. 9 shows a diagram of bit allocation.

In the example of FIG. 9, if 8 bits are allocated between -256 and 256, these values can be expressed finely at two-level intervals. On the other hand, if it is in the range of −1 to 1, 1
Allocate bits. In this way, the data amount is reduced by performing quantization according to the allocated number of bits. In the example of FIG. 9, one rectangular area requires only 162 bits.

On the restoration side, the number of quantized values for the allocated number of bits is determined in advance.
For example, when 11 bits are allocated, 11111111 is 256,
It is assumed that 00000000 is -256, 1 when one bit is allocated is 0.7, and 0 is -0.7. The value of the frequency component restored in this manner substantially obtains the original value. Then, if it is inversely transformed, the original image can be restored.

[Problems to be solved by the invention]

In the conventional method, a large amount of data is required because it is necessary to have information at a level close to a pixel unit in order to basically restore the same image as the original image. Further, since the image is divided into rectangular areas regardless of the characteristics of the image, the outlines of different textures in one rectangular area are blurred.

SUMMARY OF THE INVENTION It is an object of the present invention to divide an image into texture regions having substantially uniform texture characteristics, and to reduce the amount of data by having only the outline information of the divided texture regions and the texture characteristics.

At the same time, blurring of the contour can be prevented by providing information on the contour of the texture area. A texture is a repetitive pattern formed by arranging elements according to a certain rule. By analyzing a texture area, a texture characteristic unique to each texture can be extracted. Texture features describe textures,
The attribute of the area is described, and the same texture as the original image can be generated from the texture feature.

[Means for Solving the Problems] FIG. 1 is a diagram (1) for explaining the principle of the present invention.

In the figure, reference numeral 1 denotes an image input unit which digitizes an image and inputs it to a computer. Reference numeral 2 denotes an area dividing unit that divides an input image into the same texture area. Reference numeral 3 denotes a contour extraction unit that extracts a contour of a texture area. Reference numeral 4 denotes a texture feature extraction unit that extracts a texture feature of a texture area. Reference numeral 5 denotes an image data creation unit, and an outline extraction unit 3
Then, compressed image data is created from the texture feature extraction unit 4.

With the above configuration, an image is read by the image input unit 1, the image data read by the region dividing unit 2 is divided into the same texture region, and the contour extracting unit 3 is used for each texture region.
Then, the contour and the texture feature are extracted from the texture feature extracting unit 4 and the image data creating unit 5 creates compressed image data from the contour and the texture feature.

 FIG. 2 is a diagram (2) for explaining the principle of the present invention.

In the figure, reference numeral 6 denotes a contour restoring unit which restores the contour of each texture area according to the chain code. Reference numeral 7 denotes a texture generation unit which generates a texture from the input texture features. Reference numeral 8 denotes an image reconstructing unit that applies a corresponding texture to each area of the area divided image. An image output unit 9 outputs a restored image.

With the above configuration, the contour is restored by the contour restoring unit 6 from the compressed image data, the texture is generated by the texture generating unit, and the above data is input to the image reconstructing unit 8 to restore the image. The restored image is output by the image output unit 9.

[Action]

Using the texture features (density histogram, autocorrelation coefficient, co-occurrence matrix, power spectrum, etc.) of the input digital image data, the image is divided into regions having the same texture feature, and the contour of the texture region and the texture feature are determined. Extract and create compressed image data.

Here, the same texture is not exactly the same image, but is an image that looks different but has the same details. For example, the idea is that the lawn only needs to be reconstructed as a lawn, and it is not necessary for each grass to be the same as the original image. According to this method, the information of each pixel of the image data is not processed, and the information for each area such as lawn is only required.

〔Example〕

 FIG. 3 is a configuration diagram of one embodiment of the present invention.

In the figure, the same components as those in FIG. Reference numeral 21 denotes an image compression unit that compresses an image. An image dividing unit 22 divides the original image into a plurality of small rectangular areas (blocks). Reference numeral 23 denotes a feature extraction unit that extracts a texture feature for each of the above blocks. Reference numeral 24 denotes an area integration unit that integrates blocks of the same texture. Reference numeral 25 denotes a contour tracing unit which codes and compresses the contour of each texture area obtained by the contour dividing unit 2 using a chain code. Reference numeral 26 denotes a code generation unit which generates a chain code from the coordinate values obtained by the contour tracking unit 25. An image database 27 stores compressed image data. An image restoration unit 28 restores a compressed image. Reference numeral 29 denotes a compressed data analysis unit which divides the compressed data into area division information and texture features.

With the above configuration, specific processing steps of each unit will be described below.

First, the image is converted into a digital image by the image input unit 1 and sent to the area dividing unit 2. Here, the specific processing of the area dividing unit 2 will be described below.

First, the image dividing unit 22 divides the original image into a plurality of small rectangular areas (blocks). Next, the feature extracting unit 23 extracts a texture feature for each block. Here, the texture features are the average, variance, and model parameters of the time series model of the density level in the block. The time-series model expresses the texture using the periodicity of the texture, and the density level of a certain pixel is expressed by a linear combination of the value of the previous one-cycle pixel and white noise. The model parameters are the coefficients of this linear combination, which represent the secondary statistics of the texture.

Next, the area integration unit 24 integrates the blocks to form a texture area. Block integration is performed by comparing the texture features of adjacent blocks.
If the difference between all the texture features of the adjacent blocks is equal to or smaller than the threshold value, the two blocks are considered to have the same texture and are set to the same area. If the difference between the texture features is larger than the threshold value, the blocks are not integrated because they are considered to be different textures.

 FIG. 4 shows an example of the area division information.

In the figure, 41 is an input image, and 42 is an image obtained by dividing the area.
43 is the area number, 44 is the value of each pixel in the area division information, 45
Indicates the average density of the texture features.

Next, information of each texture area obtained by the area dividing unit 2 is passed to the contour extracting unit 3 and the texture feature extracting unit 4. The contour extraction unit 3 is composed of a contour tracking unit 25 and a code generation unit 26, and the processing will be described below.

The contour tracking unit 25 checks the coordinate values while tracking the contour of each texture area. At this time, a distinction is made at the same time as to whether the contour is the outer periphery or the inner periphery of the region. Code generator 26
Generates a chain code from the coordinate values obtained by the contour tracking unit 25.

 FIG. 5 shows an explanatory diagram of the chain code.

51 is an 8-way code. 52 is an example of a chain code.

The chain code basically expresses the outline with eight direction codes expressing the coordinates of the viewpoint and the change in direction.

The texture feature extraction unit 4 extracts a texture feature in a texture area. The texture features are the same as those used for region division, and are the average, variance, and model parameters of the time-series model of the density level in the region. The texture in the texture area is described by these texture features.

The area division information including the outline of the area and the texture feature obtained as described above are sent to the image data creation unit 5. The image data creating unit 5 creates compressed data by putting these together.

 FIG. 6 shows an example of the compressed data.

In the figure, reference numeral 61 denotes a header, which is a texture area name and an outer periphery;
Information indicating the inner circumference is stored. Reference numeral 62 denotes a code part, which describes eight directions in units of three bits. 63 is a chain code. Numeral 64 stores information indicating the average and variance of the density level of the texture area. Numeral 65 stores information indicating model parameters. Reference numeral 66 denotes a texture feature, which stores information indicating the texture feature of a certain area.

 Next, the processing of the image restoration unit 28 will be described.

The compressed data called from the image database 27 is sent to the compressed data analysis unit 29. The compressed data analyzer 29
The compressed data is divided into region division information and texture features.
The area division information that is the outline of each area represented by the chain code is sent to the outline restoring unit 6. The contour restoring unit 6 restores the contour of each texture region according to the chain code, and creates a region divided image. On the other hand, the texture feature is sent to the texture generation unit 7.

Here, the processing of the texture generation unit 7 will be described. The texture generator 7 generates a texture from the input texture feature. The texture is generated by applying the Fourier transform of the model parameters as a linear filter to the Fourier transform image of the white noise, and performing the inverse Fourier transform on the result.

The region divided image obtained as described above and the generated texture are sent to the image reconstruction unit 8. The image reconstructing unit 8 applies a corresponding texture inside each region of the region division image. Thus, the image is restored. The restored image is displayed outside by the image output unit 9.

〔The invention's effect〕

As described above, according to the present invention, an image that is not required to be so accurate, such as a rough sketch, is restored from an original image by rough texture features and restored and generated, thereby greatly compressing image data. It is possible to restore an image that retains the "atmosphere" of the original image.

[Brief description of the drawings]

FIG. 1 is a diagram (1) for explaining the principle of the present invention, FIG. 2 is a diagram (2) for explaining the principle of the present invention, FIG. 3 is a block diagram of an embodiment of the present invention, and FIG. Fig. 5 is an example of area division information. Fig. 5 is an explanatory diagram of a chain code. Fig. 6 is an example of compressed data. Fig. 7 is a schematic diagram of transform coding. FIG. 9 is a diagram showing energy concentration by orthogonal transform, and FIG. 9 is a diagram of bit allocation. 1 and 2, 1 is an image input unit, 2 is a region dividing unit, 3 is an outline extracting unit, 4 is a texture feature extracting unit, and 5 is an image data creating unit. , 6 are contour restoring units, 7 is a texture generation unit, 8 is an image reconstruction unit, and 9 is an image output unit.

────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Toshiyuki Goto 1015 Uedanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture Inside Fujitsu Limited (56) References JP-A-1-191273 (JP, A) (58) Fields investigated (Int.Cl. ⁶ , DB name) G06T 9/00-9/40 H04N 1/41-1/419

Claims (2)

(57) [Claims] In an image processing method for processing an image by dividing the image into regions having uniform characteristics, an image input unit (1) for converting the image into a digital image and storing the digital image; A region dividing unit (2) that divides an input image into a plurality of regions by integrating adjacent pixels of the same texture using a texture feature representing a density or a correlation coefficient of the density between pixels; A contour extracting unit (3) for extracting a contour of the texture region, a texture feature extracting unit (4) for extracting a texture feature in the divided texture region, and data obtained from the contour extracting unit (3). An image processing method using a texture feature, characterized by having an image data creation section (5) for creating image data in correspondence with data obtained from the texture feature extraction section (4). 2. A contour restoring unit (6) for restoring a contour of a texture area from the compressed image data created by the image data creating unit (5), and a texture in the texture area from a texture feature of the compressed image data. Texture generator to generate (7)
An image reconstruction unit (8) for applying a corresponding texture among the textures generated by the texture generation unit (7) to the inside of the outline of the texture region obtained by the outline restoration unit (6); An image output unit (9) for outputting an image obtained by the reconstruction unit (8).
The image processing method described in the item.