JPH06347420A - Image area cutting-out device - Google Patents

Abstract

PURPOSE:To eliminate redundant processing, and maintain constant high speed processing regardless of dimensions of significant information by cutting out image information after it is enlarged or thinned out according to a condition of whether or not a rectangular image area containing a significant image enters the prescribed image frame. CONSTITUTION:Digital image information after A/D conversion 1 is sent to a significant information detector 1 and a buffer 3. Significant information is extracted from the image information by the detector 2, and control information is created to read out this significant information from the buffer 3. Through a multiplexer 5, the image information read out from the buffer 3 is inputted to a buffer 7 as it is according to its necessity or after a picture element value is converted by a lookup table 4, and is transmitted to an image processing device after D/A conversion 6 is carried out. That is, the image information corresponding to the image area is sliced out as it is or after it is enlarged or thinned out according to a condition of whether or not a rectangular image area containing a significant image enters the prescribed image frame by considering the prescribed image frame being a unit area for image processing.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention converts image information sequentially input in time series into image information representing a two-dimensional image, and cuts out an image area containing significant information from the two-dimensional image. The present invention relates to an image information cutout device, for example, an image information cutout device suitable for image processing for detecting a steel plate surface flaw or image recognition.

[0002]

2. Description of the Related Art Conventionally, various types of image processing apparatuses have been proposed which can be used for detecting surface defects and surface defects. Of the various image processing devices that have been proposed,
There is a type in which image processing and recognition processing are performed on the entire input image information. Those belonging to this type are hereinafter referred to as "group 1". For example, a plurality of transputers (1CP
U-256 CPU) installed image processing device (Nippon Steel Corporation), image processing device equipped with multiple dedicated processing processors (Fujitsu Ltd.), image processing device equipped with multiple general-purpose processors (Kubota Computer Co., Ltd.) )) Etc.

In the conventional image processing apparatus, the amount of information is reduced by evenly thinning out the entire input image information,
Some have achieved high-speed image processing (Toshiba Corporation). Those belonging to this type are hereinafter referred to as "group 2". Further, among conventional image processing apparatuses, there is an example of an image apparatus having a method of uniformly dividing the entire input image information into a plurality of fixed rectangular portions and processing only a rectangular area including a significant information portion ([US] Isys). Hereinafter, it is referred to as "group 3".

[0004]

As described above, various types of image processing apparatuses have been proposed, but among them, those of the type belonging to group 1 are significant information as described above. Regardless of whether or not the whole input image information is subjected to the image processing and the recognition processing, the processing speed is slow and it cannot be a practical method for the online processing. Alternatively, there is an example in which parallel processing is attempted for the purpose of increasing the processing speed, but the scale of the hardware is large and the cost of the image processing apparatus is high, which is also impractical. Moreover, since the parallelization work of the algorithm for realizing parallel processing is enormous,
Not a practical method.

Further, the type belonging to the group 2 reduces the amount of information by evenly thinning out the input image information as described above, so that when significant information exists in the thinned pixel portion. Therefore, it becomes impossible to extract significant information. In addition, the fact that the amount of information falls evenly to 1 / N (N is an integer representing the degree of thinning) means that the resolution is 1 / N.
It corresponds to the decrease to N. Therefore, even if a sensor having a high resolution is used, there is a problem that only the information thinned out to the processing capacity of the image processing apparatus side to be actually processed can be input. In order to solve this problem, a method is considered in which significant information is always included and information is thinned out, but the amount of information is ultimately reduced and the entire process is performed regardless of the presence or absence of significant information. It has not become an effective solution. There is also a problem in that redundant processing is performed because an insignificant information portion is also processed.

Further, the type belonging to the group 3 selects one rectangular area when the significant information is in one rectangular area, and selects two rectangular areas when it extends over two rectangular areas. It is necessary to further process all the target areas in the case of extending over a plurality of areas. Therefore, there is a problem that the processing speed greatly differs depending on how many fixed rectangular portions the significant information portion corresponds to, and when online processing is performed, the processing cannot be completed.

Further, when an arbitrary rectangular portion is extracted, there is a problem that it is not possible to judge whether or not it is periodic significant information unless position information indicating which portion is extracted is obtained. SUMMARY OF THE INVENTION In view of the above circumstances, an object of the present invention is to provide an image area clipping device capable of omitting redundant processing and maintaining constant high-speed processing regardless of the size of significant information.

[0008]

An image area clipping device of the present invention which achieves the above object is (1) storage means for storing sequentially input image information (2) the image from input image information Significant image information detecting means for cutting out a part of the information as significant image information representing a predetermined significant image (3) Depending on whether or not the rectangular image area including the significant image falls within a predetermined image frame, From the image information stored in the storage means, the image information corresponding to the rectangular image area is cut out as it is or when the rectangular image area falls within the predetermined image frame.
Alternatively, when the rectangular image area extends out of the predetermined image frame, an image area cutting-out means for thinning and cutting out the image area is provided.

Here, the image area cutout means is divided into a plurality of divided areas of the image represented by the image information stored in the storage means in the size of the predetermined image frame. Depending on whether or not one entire significant image is included in the image information, the image information corresponding to one divided area including the significant image is directly or significantly selected from the image information stored in the storage unit. The image information corresponding to a plurality of divided areas including information may be thinned out to the size of the predetermined image frame and cut out, or the image area cutout means may be stored in the storage means. A cutout frame surrounding a significant image is adaptively set in the image represented by the image information, and the image information stored in the storage means is determined according to whether or not the cutout frame falls within the predetermined image frame. From among the above, when the cutout frame falls within the predetermined image frame, the image information corresponding to the image area of the size of the predetermined image frame including the cutout frame as it is or including the cutout frame is included. The image area is enlarged to the image of the size of the predetermined image frame and cut out, or when the cutout frame extends out of the predetermined image frame, the image information corresponding to the image area including the cutout frame May be thinned out to the size of the predetermined image frame and cut out.

In this case, the image area cutout means sets a minimum cutout frame surrounding the significant image, and the image information corresponding to the image area in the minimum cutout frame is
It is preferable to perform an image enlarging process or an image reducing process as necessary so that the image region is filled with the image frame and cut out. Further, in the image information cutout device of the present invention, position information representing the position in the image represented by the image information stored in the storage means of the partial image represented by the image information cut out by the image area cutout means. Alternatively, it is also a preferable aspect to provide an information output unit for obtaining and outputting, together with the position information, magnification information representing a magnification of the partial image corresponding to the image area before clipping corresponding to the partial image. It should be noted that the position information and the magnification information are closely related to each other, and as long as the information represents the position and the magnification as a result, for example, only the position information for indirectly obtaining the magnification from the position information is output. It may be one.

[0011]

The image region clipping device of the present invention considers a predetermined image frame which is a unit region for executing image processing, and determines whether a rectangular image region including a significant image falls within the predetermined image frame. In accordance with the above, the image information corresponding to the rectangular image area is cut out as it is, or enlarged or thinned out. Therefore, when processing the entire image like an image processing apparatus belonging to Group 1, or Redundant processing is prevented as compared with the case where the entire image is thinned out evenly as in the image processing apparatus belonging to group 2.

However, when the significant image is smaller than the predetermined image frame and the information for the predetermined image frame is output without enlarging the significant image, unnecessary portions that deviate from the significant image are also cut out. However, the unnecessary portion can be omitted by separately defining the unnecessary processing information. Further, since the image area cutout device of the present invention thins out and cuts out a significant image when the significant image is large, a plurality of fixed rectangular areas across which the significant images span each fixed rectangular area as in the case of the conventional group 3. Compared to the case of clipping, the shape of the entire significant image and the like can be grasped, and high-speed image processing can always be performed even if the size of the significant image changes. In particular, when it is significant over the entire surface, the recognition rate in the image recognition processing can be greatly improved by reducing the entire image.

Further, in the image area clipping device of the present invention, if the above-mentioned information output means is provided, the position and size of the significant image with respect to the entire image can be grasped, and for example, the presence or absence of repeatability and the size of the significant image. This makes it easy to understand the cause of the occurrence of the significant image. Furthermore, assuming that the recognition process is performed by a neuro, for example, the neuro is not good at recognizing similar shapes. Therefore, it is possible to realize high-accuracy recognition by normalizing them all to the same size by decimation processing etc. can do. Also, for a significant image that is much smaller than a predetermined image frame, instead of cutting it out as it is, the minimum cutting frame that surrounds the significant image is set, and the image inside the cutting frame is set to the maximum size that fits in the predetermined image frame. It becomes possible to normalize by expanding to the value.

[0014]

EXAMPLES Examples of the present invention will be described below. FIG. 1 is a block diagram showing the configuration of an embodiment of an image area cutout device of the present invention. For example, analog image information is obtained by imaging the surface of a steel plate (not shown) conveyed in a predetermined direction in the width direction of the steel plate by a one-dimensional CCD camera (not shown). 1 is input to the A / D converter 1 of the image area clipping device. This A / D converter converts the input image information into a frequency (40 MHz) that is twice the maximum frequency of the CCD camera.
It is quantized at a sampling period corresponding to z) to generate digital image information, and the image information is sent to the significant information detector 2 and the buffer 3.

The significant information detector 2 extracts the significant information from the image information based on the input image information and generates control information for reading the significant information from the buffer 3. Further, in the significant information detector 2, when the significant information is detected, the position indicating the position of the partial image represented by the significant information (this is referred to as “significant image” here) with respect to the entire image is displayed. Since the information and the magnification of the significant image may be changed when the image information is read from the buffer 3 as described later, magnification information representing the magnification is generated and output.

The buffer 3 is provided with a plurality of storage capacities for an image area used for one image processing in an image processing apparatus (not shown), and digital image information is sequentially stored in these buffers 3. . Further, the image information is read from the buffer 3 in accordance with the control information from the significant information detector 2. At the time of reading the image information from the buffer 3, a clock for writing the image information into the buffer 3 is used. Frequency of W_CLK (40 MH
The clock R_CLK having the same frequency as that of z) is used. Since the partially cut-out image information output from the buffer 3 is a burst signal, it is necessary to prepare a buffer 7 to be described later in order to convert it into a continuous signal. For example, 1/5 of the writing frequency to the buffer 7 is required. By reading out at the frequency, conversion into a continuous signal can be realized.

The image information read from the buffer 3 is
If necessary, after the pixel values are converted as they are or by the look-up table 4, one of them is selected by the multiplexer 5 and the buffer 7 is selected.
Is entered once. The image information input to the buffer 7 is read from the buffer 7 with a clock having a frequency of 8 MHz, for example, and is input to the D / A converter 6. In the D / A converter 6, the image information is converted into an analog signal and transmitted to an image processing device (not shown).

FIG. 2 is a block diagram showing a schematic configuration of the significant information detector 2 shown as a block in FIG. The digital time-series image information output from the A / D converter 1 shown in FIG. 1 is input to the shift register 2_1 and is sequentially advanced in synchronization with the clock. The parallel output of the image information input to the shift register 2_1 is input to the comparator 2_2 together with the image information input to the shift register 2_1. Further, ROM2_3 stores reference pattern information for determining whether the image information is significant information, and the reference pattern information is stored in the ROM2_3.
Read out from and input to the comparator 2_2. Comparator 2
In _2, the image information input from the shift register 2_1 and the pattern information read from the ROM and input are compared to determine whether there is significant information, the position of the significant image represented by the significant information with respect to the entire image, and the significant image. Is detected and output. The detected information is used as control information for controlling which partial image of the entire image is read as it is or how much is thinned out when reading the image information from the buffer 3. It is transmitted to an image processing apparatus (not shown) as position information indicating which position of the entire image the read image information corresponds to, and magnification information indicating how much the image information is subtracted from the buffer 3 and read.

FIG. 3 is a schematic diagram showing the image information stored in the buffer 3 shown in FIG. 1 and the significant information detected by the significant information detector 2 as an image. In this figure, the left-right direction corresponds to the width direction of the steel sheet to be inspected, and the vertical direction corresponds to the conveying direction of the steel sheet. Each square 1_1, 2_1, ..., 5_1, 1_2, 2_2, ..., 5 shown in this figure
_2, ..., 1_P, 2_P, ..., 5_P represent each buffer. That is, the steel sheet that is the material to be inspected is divided into five in the width direction, and is divided into P in the longitudinal direction and stored in each buffer. However, in reality, buffer 3
The size of 5 × P must be large enough to store the largest significant image. Considering the difference between the frequency of the write clock W_CLK to the buffer 3 and the frequency of the read clock of the buffer 7, for example, when only one buffer is provided, image information for five squares arranged side by side is input. It is necessary to extract the image information for one square during this period. When two buffers are provided, two rows of five squares arranged horizontally, that is, while the image information for ten squares is input. Image information for two squares may be extracted. Similarly, when P buffers are provided, image information for P squares may be extracted from the image information for 5 × P squares. As the number of buffers increases, the degree of freedom in extracting image information increases, and it is effective when significant information extends to the entire buffer. An appropriate number of buffers is provided in consideration of the properties of the steel sheet as the material to be inspected. To be

Here, by the significant information detector 2 (see FIG. 1), the significant information (significant image) 10_1 as shown in FIG. 3 is obtained.
It is assumed that 10_2 is detected. Significant information 10_1
Does not fit in one image frame (m × n) as it is, so the image information stored in the partial region 11_1, which is similar to the m × n image frame and spans three vertical and horizontal buffers, is ⅓. It is thinned out and read out as an image of size m × n. On the other hand, when the aspect ratio of the partial area spanned by the significant information 10_1 is not similar to the image frame of m × n, the thinning rate may be changed between the vertical and horizontal directions.

Further, at this time, the significant information detector 2 (see FIG. 1) generates and outputs magnification information representing a magnification ⅓. In addition, the partial area 11_ for the entire image
As the position information of No. 1, for example, the partial area 11_1
The coordinates at the upper left of FIG. 3 are output. The magnification information and the position information may be the position information and the magnification information, for example, by outputting the upper left coordinates and the lower right coordinates of the partial area 11_1, and as the information indicating the center coordinates and the magnification of the partial area 11_1. Of course, it goes without saying that various configurations are possible.

Here, the significant information 10_1 is the partial area 11
Nine buffers 1_1, 2_1, 3_ corresponding to _1
1, 1_2, 2_2, 3_2, 1_3, 2_3, 3_3
5 buffers 1_1, 2_
It straddles only 1, 2_2, 2_3, 3_3. Therefore, these five buffers 1_1, 2_1, 2_2,
It is conceivable to read out the image information stored in 2_3 and 3_3 as they are and send them to the image processing device, but in that case, whether the significant information 10_1 represents one continuous significant image or represents a plurality of significant images. Is difficult to perform, and since image processing is performed for each of the five buffer areas, processing time is five times that of image processing for one buffer area. On the other hand, here, since the partial area 11_1 is thinned out and cut out, even if the size of the significant information 10_1 is large, the processing speed is increased, and it is possible to determine that the image is one significant image. . In this case, significant information 10_
Since 1 is large, there is no problem even if the resolution is reduced by thinning. Rather, by thinning out in this manner, the size becomes constant, which also contributes to an improvement in the recognition rate when performing image recognition.

On the other hand, since the significant information 10_2 is in one buffer 4_4, the partial area 10_2 stored in the buffer 4_4 is cut out as it is without thinning out,
And the upper left position coordinate of the partial area 10_2 and the magnification 1 /
Information represented by 1 is output. FIG. 4 is a view similar to FIG. 3, showing another method of cutting out a partial area including a significant image.

First, the cutout frames 12_1 and 12_2 are arranged so as to enclose each of the significant information 10_1 and 10_2 in the smallest size.
Is set. Next, the entire image is scanned with an image frame of size m × n, and it is determined whether or not the cutout frames 12_1 and 12_2 fall within the image frame. In this example, the cutout frame 12
_1 does not fit in the m × n image frame, but the cutout frame 12_2 fits in the m × n image frame.
The image area 13_2 having the same size as the × n image frame is cut out as it is without thinning. Alternatively, the cutout frame 12_2
Is enlarged and output up to the m × n image frame. At that time, the enlargement ratio is also output as magnification information. As a result, the size of the significant information is standardized, which leads to an improvement in the recognition rate as described above.

Next, the entire image is scanned in an image frame having a size of 2m × 2n which is twice the size in the vertical and horizontal directions, and it is determined whether or not the cutout frame 12_2 falls within the image frame having a size of 2m × 2n.
The cut-out frame 12_1 fits in an image frame having a size of 2m × n, and thus the 2m × 2n image area 13_1 is set so as to include the cut-out frame 12_1, and the image area 13_1 is thinned to 1/2. It is cut and cut out. Alternatively, the clipping frame 12_1 may be reduced to the m × n image frame and output. Similarly, 2m x
When there is a clipping frame that does not fit in the 2n image frame, a large image frame similar to the smallest m × n image frame such as a 3m × 3n image frame is set, and the entire image is scanned. It In this way, the image area to be cut out may be adaptively set separately from the frame of each buffer.

[0026]

As described above, according to the image area clipping device of the present invention, a rectangular image area including a significant image is extracted depending on whether or not the rectangular image area is included in the predetermined image frame. Since the image information corresponding to is cut out or cut out by thinning out, redundant processing is prevented, and high-speed image processing can always be performed even if the size of the significant image changes. Further, since the significant image is adaptively enlarged and reduced to have a constant size, when the recognition processing function unit is provided in the subsequent stage, the configuration of the recognition processing function unit is simplified and highly accurate image recognition is performed. Can be realized.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of an image area cutout device of the present invention.

FIG. 2 is a block diagram showing a schematic configuration of a significant information detector shown by a block in FIG.

FIG. 3 is a schematic diagram showing, as an image, image information stored in the buffer shown in FIG. 1 and significant information detected by a significant information detector.

FIG. 4 is a view similar to FIG. 3, showing another method of cutting out a partial area including a significant image.

[Description of Reference Signs] 1 A / D converter 2 Significant information detector 2_1 Shift register 2_2 Comparator 2_3 ROM 3 Buffer 4 Lookup table 5 Multiplexer 6 D / A converter 10_1, 10_2 Significant information 11_1, 11_2 Partial area 12_1 12_2 clipping frame 13-1, 13-2 image area

Claims (5)

[Claims] 1. A storage means for storing sequentially input image information, and a significant image information detecting means for cutting out a part of the image information from the input image information as significant image information representing a predetermined significant image. According to whether or not the rectangular image area including the significant image falls within a predetermined image frame, the image information corresponding to the rectangular image area is selected from the image information stored in the storage unit. When the rectangular image area falls within the predetermined image frame, the image area cutting means is cut out as it is or enlarged, or when the rectangular image area extends out of the predetermined image frame, is thinned out and cut out. An image area clipping device comprising: 2. The image area cutout means is provided in one of divided areas in which the image represented by the image information stored in the storage means is divided into a plurality of grids in the size of the predetermined image frame. Depending on whether or not the entire one significant image is included in the image information stored in the storage unit, the image information corresponding to one of the divided areas including the significant image is directly stored, Alternatively, the image area cutout device according to claim 1, wherein the image information corresponding to the plurality of divided areas including the significant information is thinned out to the size of the predetermined image frame and cut out. 3. The image area cutout means adaptively sets a cutout frame surrounding the significant image in an image represented by image information stored in the storage means, and the cutout frame is the predetermined image frame. The size of the predetermined image frame including the cutout frame when the cutout frame falls within the predetermined image frame from among the image information stored in the storage unit according to whether or not Of the image information corresponding to the image area as it is, or the image area including the cutout frame is enlarged and cut out to an image of the size of the predetermined image frame, or the cutout frame is protruded from the predetermined image frame. The image area cropping device according to claim 1, wherein in this case, the image information corresponding to the image area including the cropping frame is thinned out to the size of the predetermined image frame and cropped. 4. The image area cutout unit sets a minimum cutout frame surrounding the significant image, and sets image information corresponding to an image area in the cutout frame so that the image area fills the image frame. The image area cutting device according to claim 3, wherein the image area cutting processing is performed by performing image enlargement processing or image reduction processing as necessary. 5. Position information indicating the position in the image represented by the image information stored in the storage unit of the partial image represented by the image information clipped by the image area clipping unit, or the position information. Of the partial image,
2. The image area clipping device according to claim 1, further comprising information output means for obtaining and outputting magnification information representing a magnification with respect to an image area before clipping corresponding to the partial image.