JPS6391784A - Image identifying system - Google Patents

Abstract

PURPOSE:To contrive to remarkably simplify and accelerate a circuit, by setting the coefficient of a filter at + or -2+ or -<n> (n=0, or positive integer), or 0, except the center of a coefficient matrix. CONSTITUTION:The coefficient other than the center picture element of the filter is set at + or -2+ or -<n> (n=0, or positive integer), or 0, and multiplication by a multiplier is applied only on the center picture element, and an attenuation quantity is shifted by weighing. In other words, a filter arithmetic circuit 2 is provided with the multiplier for the coefficient of the center of the filter which decides an enhanced frequency band area basically, and a data shift circuit 21 is provided for each of the coefficient in the neighborhood which occupies a large part of the coefficients to be calculated, and the outputs of the multiplier 22 and the data shift circuit 21 are added at an adder 23. In this way, it is possible to perform the large parts of the arithmetic calculation of a product in a filter processing by a simple shift processing, and to simplify the circuit, and to perform an operation at high speed.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an image identification method used in an image processing device or the like.

[Conventional technology]

In an image processing device or the like, in order to perform appropriate processing depending on the type of image, it is necessary to identify images such as character line images, 1M 4-point images, photographic images, and the like.

As a method for image identification, Japanese Patent Application Laid-Open No. 58-20
There are many methods such as those described in Japanese Patent Laid-Open No. 5376 and Japanese Patent Laid-Open No. 56-132061, and one of them is an image identification method using a filter. This method uses the fact that text line images, halftone dot images, and photographic images have different frequency characteristics for identification.

Figures 11 to 13 are photographic images and halftone dot images, respectively.

This is the frequency characteristic of a character line image. As can be seen from this figure, photographic images and halftone dot images have a high proportion of frequency components near DC, and halftone images are characterized by the presence of halftone frequency components in addition to this. On the other hand, character line images are characterized by continuously containing frequency components up to high frequency regions.

Here, in order to identify a character line image, for example, a mid-range frequency component that is not included in other images may be extracted using, for example, a digital filter.

Therefore, as a digital filter, a filter having a coefficient matrix of MXN size is used to emphasize the mid-range frequency components included only in the character line image, and to
In particular, the frequency characteristic may be such that the DC component and the halftone frequency component are attenuated. At this time, the pixel data at the center of the image and the pixel data around it, that is, the grayscale data of the pixel, are multiplied by each coefficient of the M x N filter to obtain the product,
Filter processing is performed by finding the sum of these products.

[Problem that the invention seeks to solve]

The digital filter has a circuit configuration consisting of a product-sum calculation circuit, and after performing a product calculation for each coefficient of the filter, a calculation is performed to obtain the sum of each product.

Therefore, as the filter size increases in terms of −i, the number of elements in the filter increases and the circuit scale increases. Therefore, when the filter size is large, there is a problem in the feasibility of creating a digital filter.

Furthermore, as an operation, addition can be processed relatively quickly, whereas multiplication requires processing time and requires a large number of operations, which makes the processing time even longer and has the problem of not being able to be processed at high speed. Ta.

Regarding the problem of processing time, the use of a ROM table enables high-speed processing, but there is a problem in that the circuit configuration becomes even larger.

The present invention is intended to eliminate and improve the drawbacks of the prior art described above, and aims to simplify and speed up the filter circuit configuration for identifying types of images.

[Means and effects for solving the problem] In order to achieve the above object, the image identification method of the present invention includes multiplication in which pixel data of an image is multiplied by each coefficient of a filter having a coefficient matrix of a predetermined size. By performing calculations,
An image identification method that performs filter calculation processing that selectively attenuates frequency components of the image, and identifies the type of the image according to the value of the signal after the cough filter calculation processing,
The coefficients of the filter are set to ±21 (n is 0 or a positive integer) or O except for the center of the coefficient matrix.

In the present invention, the coefficients other than the center of the coefficient matrix are set to ±2±1. Since the multiplication of ±2±η can be realized only by a data shift operation, it is possible to extremely easily perform a product operation on surrounding pixels that account for most of the pixels to be filtered. Therefore, significant simplification and speed-up of the circuit are realized.

〔Example〕

DETAILED DESCRIPTION OF THE INVENTION Hereinafter, features of the present invention will be specifically described based on examples with reference to the drawings.

FIG. 2 is a schematic diagram showing an image identification device to which the image identification method according to the present invention is applied.

In the figure, 1 indicates a two-dimensional memory into which image data So is written. As shown in FIG. 1, the two-dimensional memory 1 is composed of a plurality of line memories 11 and a memory shift circuit 12 provided corresponding to the vertical size of the filter. Each line memory 11 stores grayscale data of the pixel of interest and its surrounding pixels, and is transferred by the memory shift circuit 12 to the subsequent filter calculation circuit 2.

The filter calculation γ circuit 2 performs arithmetic processing of multiplying the grayscale data of the pixel of interest and its surrounding pixels by a predetermined coefficient based on the data in the two-dimensional memory 1, and calculating the sum of the data. This calculation performs a so-called filter calculation that selectively attenuates image frequency components.

As a method for performing the above filter calculation, for example, there is a method using a digital filter.

Digital filters can realize various frequency characteristics such as low-pass filters and band-pass filters by changing the filter size and coefficients. Utilizing this feature, among the frequency components, the DC neighborhood component and the halftone frequency component are selectively attenuated.

In this example, in order to simplify the circuit scale and achieve high-speed processing, the coefficients of the filters other than the center are set to ±2''''.
(n is O or a positive integer) or 0. The reason for selecting the filter coefficients in this manner will be described in detail later.

The output of the filter calculation circuit 2 is supplied to a comparison and determination circuit 3, and based on the results processed by the filter calculation circuit 2, character line images, photographic images, and halftone dot images are determined. That is, as described above, since the frequency characteristics of the character line image 2w 4-point image and the photographic image are different, the type of image can be identified by comparing the output of the filter calculation circuit 2 with a predetermined level. Note that as the predetermined level, a fixed darkness value or a darkness value in which a pixel of interest is weighted a certain amount can be used.

The reason why the filter coefficients are set to ±2±1 will be explained with reference to FIG.

For example, consider multiplying the 8-bit grayscale data shown in FIG. 3 [al] by four. For digital data, a 4x operation can be achieved by shifting the data 2 bits to the MSB side (right side in the figure) as shown in Figure 3 (bl).Similarly, 8x is 3 bits, 16x is 4 bits, and 16x is 4 bits.
obtained by bit shifting (Fig. 3 (C1
, (d)). In this way, the operation of multiplying data by ±2' (n is an integer) can be obtained by simply shifting the data. Therefore, since multiplication is realized only by shift operations, the circuit configuration is simplified. In addition, since only shift operations are required, processing can be executed at high speed. Furthermore, multiplication by ±1/2' can also be realized by setting n to a negative integer.

Specifically, this can be achieved by shifting the data to the LSB side (left side in the figure). Note that n is 0.
If so, it will be multiplied by 1, and it can be output as is without shifting.

This embodiment utilizes this fact to change the coefficients of the filter for pixels other than the center pixel to ±2" (n is O or a positive integer) or O.
Then, only the central pixel is multiplied by a multiplier, and the attenuation amount is shifted by weighting.

That is, as shown in the filter arithmetic circuit 2 in FIG. A data shift circuit 21 is provided for each coefficient, and an adder 23 adds the outputs of the multiplier 22 and each data shift circuit 21.

With such a filter configuration, a pure product operation is only required once for the central coefficient, and the rest are only shift operations and additions. Therefore, most of the product calculations in filter processing can be accomplished by simple shift processing, which simplifies the circuit and speeds up the operation.

An example of filter coefficients satisfying such conditions is shown in FIG. Moreover, the frequency characteristics at this time are shown in FIG.

In the frequency characteristics shown in Fig. 5, if the areas where the gain is low, that is, the areas where the attenuation is large, are matched to the DC neighborhood band and the halftone frequency band, the output will be large only when a character line image is input. Therefore, character line images can be distinguished from photographic images and halftone dot images.

Here, a design method when using the filter configuration of this embodiment will be explained with reference to FIGS. 6 to 10. Note that in this embodiment, it is the emphasized frequency band that is a problem as a filter characteristic, and the gain is not so important.

In this embodiment, the frequency band to be emphasized is basically determined by the value of the center coefficient of the filter. As shown in Fig. 6, when the frequency bands obtained for the desired emphasis frequency bands fa and fb are fo and f+, by setting the center coefficient of the filter to an appropriate value, , desired characteristics as shown in FIG. 7 can be obtained. However, as shown in FIG. 8, the obtained frequency band fo
, L are significantly shifted from the emphasis frequency bands fa and fb, it is not possible to obtain the desired characteristics simply by changing the coefficients at the center of the filter. In such a case, by changing the position or value of the coefficients other than the center of the filter, or by changing the size of the filter, the frequency bands f, , f, as shown in FIG. 9 can be changed.
can change its position.

When changing the value of the coefficient, the value of the coefficient should be ±2±T.
' (n is O or a positive integer) or 0, but there is no restriction on the gain, so there is a large degree of freedom. Furthermore, by setting the center coefficient to an appropriate value, the position of the frequency bands f, , f can be brought closer to the emphasis frequency bands fa, fb, as shown in FIG.

By such a method, filter characteristics that emphasize only a desired band can be obtained.

In the embodiments of the present invention, a two-dimensional filter has been described, but the invention is not limited to this.
A similar configuration is also possible for a one-dimensional filter.

Furthermore, the circuit configuration can be further simplified by utilizing the symmetry of the filter. For example, if the coefficient of the filter is 0, the calculation speed can be increased by removing it from the calculation target from the beginning and not performing the calculation. In addition, if there are identical coefficients in the filter, by first adding the pixel data of the same coefficients and then performing multiplication, that is, shift operation, the number of shift operations is reduced, further increasing speed and simplifying the circuit configuration. can be converted into

Therefore, it is also possible to realize an ultra-high-speed LSI for digital filters, which was previously difficult to realize due to the complexity of the circuit configuration and the limitations of operating speed.

In addition, in the above-mentioned embodiment, only the identification of character line images was explained, but the invention is not limited to this.
Halftone images and photographic images can also be identified by using filters with other frequency characteristics. For example, if a filter that emphasizes the halftone dot frequency is used, a halftone dot image including the halftone dot frequency becomes an image in which the halftone dot pitch is emphasized. On the other hand, photographic images are less affected by filters because they do not have halftone frequency components. Therefore, halftone images can be identified by making a correlation with the output image.

〔Effect of the invention〕

As described above, in the present invention, in the filter calculation for identifying the type of image, the coefficients of the filter are set to ±2 η (n is O or a corrected integer) or 0 except for the central coefficient. This makes it possible to perform multiplication on peripheral pixels that account for most of the pixels to be filtered using a simple shift operation. Therefore, the number of multipliers, which conventionally required a complicated configuration and long processing time, can be significantly reduced. This simplifies the circuit configuration, reduces the time and number of multiplications, and enables high-speed processing.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a schematic block diagram for explaining the present invention, FIG. 3 is an explanatory diagram of multiplication using shift, and FIG. 4 is a block diagram for explaining the present invention. A diagram showing an example of the filter coefficients used, FIG. 5 is a graph showing an example of the frequency characteristics of a filter having the filter coefficients of FIG. 4, and FIGS. 6 to 10 are filters for explaining the filter design method. Graphs showing examples of characteristics, FIGS. 11, 12, and 13 are graphs showing examples of frequency characteristics of photographic images, halftone dot images, and character line images, respectively. Patent applicant Fuji Xerox Co., Ltd. Agent
Pediatric benefit (2 others) Figure 6 Figure 7 Figure 8 519 Figure 10 Figure 11 Figure 13 Frequency Figure 12 Frequency

Claims (1)

[Claims] 1. Performing a filter operation process that selectively attenuates the frequency components of the image by performing an operation including multiplication of multiplying the pixel data of the image by each coefficient of a filter having a coefficient matrix of a predetermined size; In an image identification method that identifies the type of the image according to the value of the signal after arithmetic processing, the coefficients of the filter are set to ±2^±^n (n is 0 or a positive integer) except for the center of the coefficient matrix, or An image identification method characterized by setting the value to 0.