JPH0269877A - Picture processing device - Google Patents

Abstract

PURPOSE:To enable plural processors to perform picture processing of which they take partial charge without exchanging information between them by determining plural areas by dividing picture data stored in a picture memory so that a coupled component to be an object to be recognized is not divided. CONSTITUTION:A divided area determining means 13 divides the picture data so as not to divide the coupled component to be the object to be recognized, and determines plural areas. Therefore, the coupled component to be the object to be recognized or to be its composing element exists always in a state being included in one area, and it never exists as spreading over plural areas. Accordingly, the whole object to be recognized or its separated composing element can be processed in one local processor 15a to 15n. Thus, the communication of the picture data or feature information among the local processors 15a to 15n can be reduced extremely, and simultaneously, hardware for picture communication among the local processors 15a to 15n can be simplified.

Description

[Detailed Description of the Invention] [Object of the Invention] (Industrial Application Field) The present invention relates to an image processing device that performs processing related to recognition of a recognition target included in image data, and particularly relates to an image processing device that performs processing related to recognition of a recognition target included in image data. The present invention relates to an image processing device that divides each region and processes each region in parallel with a plurality of processors.

(Prior art) In order to recognize and understand the content contained in an image, the image data is first subjected to preprocessing such as noise removal, emphasis, and binarization, and then image features that match the purpose are extracted. After this image processing is performed, it is necessary to use the last extracted image features to analyze and judge the object in the image. Among these processes, in order to perform the image processing part at high speed,
2. Description of the Related Art An image processing apparatus is known that divides image data into a plurality of regions and assigns a plurality of processors to each region to perform parallel processing.

FIG. 8 shows the configuration of such an image processing apparatus.

This processing device includes one host computer 1, a shared image memory 3 connected to the host computer 1 via a common bus 2, and a plurality of local processors 4ar.
4b, 4c,..., 4n. Each local processor 4a-4n is provided with interrupt control means from the common bus 2, a local memory and a communication interface to other processors.

In this image processing device, local processors 4a to 4n
The processing is controlled by the host computer 1. In particular, when performing image processing, the host computer 1 divides the entire image into n regions of the local processors 4a to 4n, and stores the image data of each divided region in the local memory of each local processor 4a to 4n. Responsible for processing. In addition, mouth ~ cal processor 4a ~ 4n
process the image data in their local memory in parallel.

Conventionally, when dividing image data, the entire image is mechanically divided into multiple regions vertically and horizontally, regardless of the content of the image data.
It was arranged to allocate it to each local processor 4a to 4n. This will be explained using FIGS. 9 and 10. That is, in FIG. 9, the entire image 5 is divided into four vertically and horizontally, and in FIG. 10, the entire screen 5 is divided vertically into four, and each region 6a is
~6d has been determined. The target 7b that should be recognized as being divided in this way is the area 6a (FIG. 9) or the area 6.
b (Fig. 10), but object 7a,
7b and 7c are in a state where they each straddle two adjacent areas. Now, host computer 1 is in area 6.
a to processor 4a, area 6b to processor 4b,
The area 6c is connected to the processor 4c, and the area 6d is connected to the processor 4.
d, in order for the processor 4a to extract various characteristics of the object 7a, it must receive data regarding the object 7a that is not included in the area 6a of the processor 4a from the processor 4b in the adjacent area 6b. That is, in the conventional system, there was a problem in that such image data exchange was frequently performed between the local processors 4a to 40, and the overall processing speed was reduced due to communication control.

(Problem to be Solved by the Invention) As described above, in conventional image processing devices in which multiple processors perform image processing in parallel, communication between the processors is frequently performed, so in order to control the communication, the entire processing There was a problem that the speed decreased and the effect of parallel processing was not sufficiently obtained.

The present invention was made in order to solve the above problem, and an object of the present invention is to provide an image processing device capable of high-speed image processing that eliminates communication of information between processors as much as possible and takes full advantage of the advantages of parallel processing. It is said that

[Structure of the Invention] (Means for Solving the Problems) The present invention provides an image memory that stores image data to be processed, and a system in which connected components to be recognized are not separated from the image data stored in the image memory. divided area determining means for determining a plurality of areas by dividing the area as shown in FIG.
It is characterized by comprising a plurality of processors that process image data in each region in parallel.

(Operation) According to the present invention, the divided region determining means determines a plurality of regions by dividing the image data without dividing the connected components to be recognized. It always exists within one area, and never across multiple areas. Therefore, the entire recognition target or its separate components can be processed within one processor. As a result, it becomes possible to significantly reduce communication of image data and feature information between processors. At the same time, the hardware for image communication between processors can also be simplified. Therefore,
The effect of parallel processing can be sufficiently increased.

(Example) Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a diagram showing the configuration of an image processing system according to an embodiment of the present invention. This system includes a host computer 11 that controls the entire system, a divided area determining means 13 connected to the host computer 11 via a control bus 12, a shared image memory 14, and n local processors 15a, 15b. −,15n
and an image bus 16 for high-speed transmission of image data between them.

The divided area determining means 13 divides the image data stored in the shared image memory 14 so that connected components to be recognized are not separated, and divides the image data stored in the shared image memory 14 into each local processor 15a to 15n.
After the area is determined by referring to the shared image memory 14, based on this, the image data stored in the shared image memory 14 is transferred at high speed through the image bus 16 to each local processor. 15.
Store in the local memory of a to 15n. The local processors 15a to 15n process the stored image data in parallel to perform image preprocessing and feature extraction. The host computer 11 controls these processes through the control bus 12. Further, the host computer 11 receives the results of the feature extraction through the control bus 15, and performs analysis and determination processing regarding final image recognition.

FIG. 2 shows an example of the processing procedure of the divided area determining means 13. This processing is a method in which image data is binarized by comparison with a certain threshold, and the binarized image is divided into connected components and assigned to each processor. In this method, the divided region determining means 13 includes a processor for performing image processing, and performs noise removal processing such as filtering (21) and edge enhancement processing (22) on image data.
) Perform region enhancement processing (23). Next, threshold processing is performed to obtain a binary image (24), and labeling processing is performed to obtain information for each connected component (25). Through this labeling process, label numbers to be assigned to each of the processors 15a to 15n are determined (26). In this allocation label determination, if the maximum label number is larger than the number n of processors, several label numbers are allocated to one processor. Then, a circumscribed rectangle containing all the corresponding connected components is determined based on the last determined label number, and this area is designated as the area in charge of each processor (27). Note that the information of each divided area consists of information on two points, the upper left and lower right of the rectangular area, and this information is transmitted to each processor 15a to 15n through the control bus 12.
sent to.

FIG. 3 shows an example in which one side edge data 31 is divided into regions by this division region determining means 13. Note that this example shows a case where the number n of local processors 15a to 15n is four. Area 32 consisting of each circumscribed rectangle
a, 32 b, 32 c.

32d are connected components 33a and 3 to be recognized, respectively.
3b, 33c, and 33d are included without being separated.

Therefore, for example, by allocating the area 31a to the processor 15a, the area 31b to the processor 15b, the area 31c to the processor 15c, and the area 31d to the processor 15d, each connection can be made without exchanging information between the processors 15a to 15d. Features of the components 33a to 33d can be extracted.

FIG. 4 shows another example of area division. Here, an example of region determination in a case where the number of connected components 43 included in the image data 41 after binarization is greater than the number of processors (n=4) is shown. In this example, image data 4 after binarization is
Since the number of connected components 43 included in 1 is 17, these connected components 43 are allocated almost equally to each of the processors 15a to 15d. That is, an area 42a consisting of a circumscribed rectangle that includes all of the first five connected components 43 in ascending order of label number is assigned to the processor 15a, and an area 42b that includes all of the next four connected components 43 is assigned to the processor 15a. Assigned to 15b. Thereafter, the area 42c is allocated to the processor 15c and the area 42d is allocated to the processor 15d in the same manner. By performing such division, each connected component 43 can be divided without exchanging information between the processors 15a to 15d.
The features of can be extracted.

FIG. 5 shows still another example of area division. In this example, the divided region determining means 13 performs threshold processing (24) and then performs thinning processing to convert all the connected components 53 in the image data 51 into line components into regions 52 a, 52 b,
This is an example in which 52 c and 52 d were calculated. In this case as well, the connected components may be allocated to each processor so as to be equally divided into approximately the same number of components, as described in FIG.

Once the regions have been divided in the manner described above, the image data of each region is transmitted from the shared image memory 14 to each of the processors 1.5a to 15n at high speed via the image bus 16.

Each of the processors 15a to 15n performs preprocessing such as noise removal and edge enhancement on each of the transmitted image data, and then performs feature extraction.

Note that, as shown in FIG. 5, the rectangular area handled by one processor may become extremely large. in this case,
This increases the burden on the processor that handles a large area. In order to prevent this, the divided region determining means 13 does not perform preprocessing such as noise removal or edge enhancement in the processors 15a to 15n.
It is also effective to send the binarized or thinned image data from the divided area determining means 13 to each of the processors 15a to 15n via the image bus 16.

FIG. 6 shows divided area determining means 1 according to another embodiment of the present invention.
3 shows the processing procedure. In this method, the model of the object to be recognized is registered in advance in the host computer, and by simply comparing the image data with the model,
The area required for image recognition, particularly target recognition, is extracted from the entire image and assigned to each processor. That is, first, noise removal processing is performed on the image data (61), and then edge enhancement and edge reduction processing is performed (62).
. Next, threshold processing is performed (63) to obtain a binarized image. Line thinning processing is introduced as necessary (64) to obtain line components. Here, matching is performed using a model that has been input into the host computer in advance (65).
The area to be allocated is determined using the result (66).

FIG. 7 shows an example in which regions are determined by the divided region determination procedure shown in FIG. 6. Here, a case is shown in which the model input to the host computer in advance is an ellipse. In this example, the connected component 7 included in the binarized image 71
Among 3.74, the triangular component 74 that exists at the edge of the image
is removed from the recognition target by matching with the model. The 5 pixels near the center of the image are assigned to each processor.
There are only connected components 73.

Note that in each of the above embodiments, a rectangular area circumscribing the connected components is allocated to each processor, but the allocation area may be set to be a little larger with allowances in the vertical and horizontal directions. In this case, even if the process of determining the rectangle is somewhat rough, the recognition target can always be correctly extracted.

[Effects of the Invention] As described above, according to the present invention, connected components to be recognized are not divided by region division. Therefore, it is possible to perform shared image processing without exchanging information between multiple processors, making full use of the advantages of parallel processing, and reducing the amount of hardware required for data communication. There is an effect.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an image processing apparatus according to an embodiment of the present invention, FIG. 2 is a flowchart showing the processing flow of the divided area determining means in the same apparatus, and FIGS. FIG. 6 is a flowchart showing the processing flow of the divided region determining means according to another embodiment of the present invention, and FIG. 7 shows an example of divided regions by the divided region determining means. 8 are block diagrams of a conventional image processing device, and FIGS. 9 and 10 are diagrams each showing an example of region division by the same device. 1.11...Host computer, 2...Common bus, 3, 14...Shared image memory, 4a to 4n. 15a to 15n...Local processor, 12 river control bus, 13...Divided area determining means, 16...
・Image bus. Applicant's Representative Patent Attorney Takehiko Suzue Figure 4 Figure 5 Figure 2 Figure 3 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10

Claims (1)

[Claims] An image memory that stores image data to be processed, and a division that determines a plurality of regions by dividing the image data stored in the image memory so that connected components to be recognized are not separated. An image processing device comprising: region determining means; and a plurality of processors that are respectively assigned to each region determined by the divided region determining means and process image data in each region in parallel.